Data types used for network planning 177

Figure 4.5 – Current usage of data types across case study group129

There is no common data type that is used across all airlines. Flight sched-
ules, operating cost, and traffic rights are the most widely used data types
(7 out of 9 airlines). Rather innovative data sources, such as clickstream
data from booking websites and real-time data feeds from global distribu-
tion systems, are the least used data types, corresponding to the initial
hypothesis that NP departments rely mostly on traditional sources for NP.
Data on meteorological events is only used by two NP departments, since

129
Source: Own illustration.
178 Status quo of strategic network planning in airlines

in most airlines, the NP process is concluded before reliable meteorologi-

cal forecasts are available. Every airline seems to have its own traditional
data pool, which is not standardized across the sample.

4.2.3 Reliability and usefulness of data types

In the second part of the structured survey, participants were asked to rate
the reliability and usefulness of the data types for their specific NP process.
While reliability was only ranked by the airlines that are currently using the
data type or have used it in the past five years, the usefulness was evalu-
ated by all participants. The participants could rate the data types on a four-
point scale130, from very useful/reliable to not useful/reliable. Figure 4.6
displays the average survey results.

The average reliability score across all 22 data types is 3.23, with most
evaluations within the 75% quantile. This observation indicates that airline
network planners are generally satisfied with the reliability level of the cur-
rently used data sources. The most reliable data type is flight event data
from actually flown flights with an average score of 4.0 that is also the max-
imum score. Three data types – flight schedules, airport characteristics,
and aircraft specifications – are also perceived as highly reliable with an
average score of 3.7. Clickstream data from booking websites is perceived
the least reliable with a score of 2.4, followed by ready-made market fore-
casts (2.6) and political event forecasting (2.7).

Average usefulness across all data types was evaluated at 3.30, which is
slightly higher than the average reliability. This indicates that the currently
used data sources provide satisfactory information. Operating cost are
deemed most useful with an average score of 3.9, followed by real-time
booking curves (3.8), airport slot allocation (3.7), and flight schedules (3.7).

130
Compare section 2.3.3: scale following the very positive – slightly positive – slightly nega-
tive – very negative logic, as suggested by Berg (2004).
Data types used for network planning 179

Meteorological event forecasting is perceived as the least useful with an

average score of 2.4, followed by political event forecasting (2.8).

Figure 4.6 – Perceived usefulness and reliability of the 22 currently used data types131

The difference in scores for usefulness and reliability allow for a clustering
of data types. When plotted on a matrix graph (Figure 4.7) with reliability
along the X-axis and usefulness along the Y-axis, four generic clusters
could be defined. If a data type is both useful and reliable (value over 3.0),

131
Source: Own illustration.
180 Status quo of strategic network planning in airlines

it should be maintained for fully meeting its purpose. These are the data
essentials for airline NP, comprising exactly half of all data types132.

Figure 4.7 – Usefulness and reliability analysis of the currently used data types133

On the contrary, if a data type is neither useful nor reliable, it does not add
any value to the NP process. It can even destroy value through bad deci-
sions. Two data types fall in this cluster: political event forecasting and me-
teorological event forecasting. If airlines are not able to improve either the
reliability or usefulness of these data sources, they should eliminate such
hands-off data.

132
Demographic data, real-time booking curves, real-time booking data from global distribu-
tion systems (GDS), oil price forecasts, MIDT data, fare data, airport traffic statistics, airport
slot allocation, real-time slot availability, traffic rights, including overflight rights, aircraft spec-
ifications.
133
Source: Own illustration.
IT tools used for network planning 181

Data types that are reliable but not useful can be retained if they do not
incur additional cost; these are nice-to-have data. Operating cost and air-
port characteristics are the only ones in this category. Operating cost infor-
mation is usually available from internal sources, and airport characteris-
tics are procured for flight operations in any case.

The last cluster are data types that are very useful but not “yet” reliable.
These “unfulfilled data needs” include seven data types, namely ready-
made market forecasts, economic forecasts, clickstream data from book-
ing websites, airport and traffic fees, flight schedules, flight event data on
actually flown flights, and competitor-related event forecasting.

This clustering is the first step towards identifying BDOs for airline NP. In
principle, big data can generate value in three different ways. First, it can
improve the reliability of currently used data types by shedding light on
“blind spots” or improving the validity of a data type. Second, it can improve
the usefulness of a current data type by allowing for additional analyses
that extract the right information for airline network planners. Finally, big
data can create entirely new data types that complement the currently used
set of data types. Clustering provides a starting point for the first two value
generation mechanisms, as it highlights the improvement potential of the
currently used data types. Sub-chapter 5.2 discusses the findings of this
section to develop potential BDOs.

4.3 IT tools used for network planning

The data used for NP has to be processed and analyzed to generate useful
information for NP staff. This sub-chapter analyzes the current market of
IT tools for airline NP (section 4.3.1) and presents the practical use of these
systems by the airlines of the case study group (section 4.3.2). The last
section 4.3.3 draws on additional interviews with an IT system provider as
well as on publicly available information in order to map out the data capa-
bilities of existing NP IT solutions.
182 Status quo of strategic network planning in airlines

4.3.1 Market of IT tools for airline network planning

The easiest differentiation method from an outside-in perspective is the

scope analysis. Taking the NP process developed in section 3.5.7 as the
baseline, the coverage of available systems can be mapped.

Figure 4.8 – Market overview over commercially available IT systems for airline network
planning134

134
List derived from Airline Software (2017); specific information derived from solution web-
sites: AIMS (2017); APM Technologies (2017); AVES – Flight Operations Management Sys-
tems (2017); Aviation Research Technologies (2017); Aviolinx (2017); HITIT (2017); ICF
IT tools used for network planning 183

The Digital Solution Directory from Airline Software (2017) contains 66 IT

solutions in the NP and scheduling field. Only 14 of these solutions cover
more than two planning steps, all of which are included in Figure 4.8.

Three different types of IT solutions can be discerned. Integrated network

planning solutions cover the entire NP process and may include various
individual solutions bundled as a single suite. Currently, there are three all-
in-one solutions provided by Sabre (AirVision Suite), Lufthansa Systems
(NetLine Suite), and Seabury-Accenture (APG Suite). The Network Plan-
ner developed by Aviation Research Technologies encompasses almost
the entire NP process and lacks only crew planning and flight operations
functionality. Conversely, the RAIDO solution from Aviolinx omits network
structure and demand estimates, otherwise covering the entire process
from route selection to the flight operations phase.

Strategic network planning solutions focus on the strategic NP steps up to

scheduling and fleet assignment. Typical examples for this are the solu-
tions from L.E.K. Consulting and ICF, partly also the NP suite offered by
Lift, a relatively new company in the market.

Tactical network planning solutions focus on technical schedule optimiza-

tion and usually start with the flight scheduling process step. The IT tools
from AVES, APM, and AIMS cover all process steps up to the flight oper-
ations phase, whereas the Crane Suite from HITIT has no flight operations
functionality. The solutions developed by Optym and PDC only cover the
core process steps of scheduling, fleet assignment, and aircraft routing.

System developer firms can also be distinguished in three groups. The first
one includes large aviation system developers, covering a wide range of
business processes. Sabre, Lufthansa Systems, and APM Technologies
belong to this group. Boutique system developers that focus only on NP
and flight operations are the second group, which includes Optym, Lift,

(2017); L.E.K. Consulting (2017); Lift – airline planning solutions (2017); Lufthansa Systems
(2017); Optym (2017); PDC Aviation (2017); Sabre (2017); Seabury Consulting (2017)
184 Status quo of strategic network planning in airlines

PDC Aviation, Aviolinx, AVES, and AIMS. The third group are consultan-
cies that have started to market their proprietary decision-support tools.
L.E.K, Seabury (now part of Accenture), ICF, and Aviation Research Tech-
nologies are part of this last group.

It is remarkable that all the available solutions are provided by system de-
veloping companies with a specific focus on aviation. Large commercial
software development companies such as Infosys, SAP, Oracle, or IBM do
not offer any NP solutions for airlines. This further supports the hypothesis
that airline NP is highly specialized and interlinked with other airline busi-
ness processes.

4.3.2 Current use of IT tools in the case study airlines

Despite the diverse offer of IT solutions for NP, the participating case study
airlines utilize solutions from only three providers. The NetLine suite devel-
oped by Lufthansa Systems is the most widely used NP system in the case
study group. The NetLine/Sched application that covers the tactical plan-
ning process is used by five airlines, the strategic NetLine/Plan by one air-
line. AIMS and Sabre solutions are each used by two airlines. LCC 1 does
not use any airline NP solution, relying solely on Microsoft Excel. Eight out
of the nine case study airlines use Microsoft Excel as an important tool,
making it the most important unspecific IT tool.

Furthermore, two airlines built their own market forecasting models, as they
deemed existing ready-made market forecasts to be unreliable (compare
previous section). One cargo airline employs a cargo-specific revenue
management system (CHAMP) to inform NP decisions. Figure 4.9 sum-
marizes the currently used IT systems.
IT tools used for network planning 185

Figure 4.9 – IT systems currently used by the case study participants for network plan-
ning135

The reliance on Microsoft Excel and IT solutions for tactical NP has also
been reflected in the current degree of process automation. While demand
forecasting is partly automated in the NP process, strategic NP mostly re-
mains a manual process with little IT support. Standardized business case
calculations are136 the only noteworthy automation, even in large airlines,
before the flight scheduling process step. While FSC 1 and FSC 2 have
started to automate the first steps beyond business plan creation, all other
airlines rely on a fully manual strategic NP process.

With the exception of LCC 1 and LCC 2, the tactical NP process is more
automated than the strategic planning process. In most airlines, the stand-
ard AIMS and NetLine functionality is used. FSC 1 and CAR 1 are currently
working towards a complete automation of the tactical planning process.
However, the extreme complexity of airline NP requires non-exact optimi-

135
Source: Own illustration.
136
With a fleet size of more than 100 aircraft.
186 Status quo of strategic network planning in airlines

zation heuristics that can cause significant re-planning in case of a param-

eter change (Kölker & Lütjens, 2015, p. 901). Figure 4.10 displays the cur-
rent state of automation in airline NP graphically.

Figure 4.10 – Degree of automation in current NP process137

4.3.3 Data capability of current IT tools

Information on the data processing capability of current systems is difficult

to obtain, since publicly available information exists for only 7 of the 14 IT
solutions presented in section 4.3.1. When analyzing the publicly available

137
Source: Own illustration.
IT tools used for network planning 187

information (right column of Figure 4.8), the commonalities among the dif-
ferent IT solutions are evident at the onset. Most IT systems rely on existing
schedules, including revenue and QSI information and an O&D level fore-
cast of expected demand. Airline internal data, such as operating cost and
past booking data, as well as strategic assumptions on market and demand
development can be handled by most systems. Not a single IT solution for
NP claims to support real-time data, such as live bookings or search-en-
gine data.

One system provider is included in the additional source tri-angulation and

it provided information on the data capabilities of the NP solution. System
developers distinguished primary from secondary data integration. The pri-
mary data integration occurs when databases are processed and analyzed
directly by the system. In secondary data integration, assumptions based
on primary data are entered without replicating the primary data structure.
Only two data types from the analysis in section 4.2.2 are integrated by
default, namely airport characteristics and publicly available statistical data
on actually flown flights. Seven additional data types are mandatory to run
the NP application. These need to be provided by the airlines and include
flight schedules, MIDT data, fare data, operating cost, slots and aircraft
characteristics. Furthermore, there are mandatory secondary data that
need to be translated as assumptions. Market estimates, economic fore-
casts, and oil price forecasts are key assumptions to be entered for this.

Besides the mandatory data types, there are additional optional data types
that can be processed by the NP IT system. Real-time internal booking
data and external GDS data can be incorporated in addition to airport traffic
statistics and traffic rights. Data from booking platforms (such as Skyscan-
ner) is currently in a pilot testing phase.

Five data types cannot be integrated in the NP solution. Neither demo-

graphic data nor real-time slots and any data on external shocks are en-
visaged. Interestingly, these data types are widely used among the case
188 Status quo of strategic network planning in airlines

study airlines138. Figure 4.11 summarizes the data integration in an exem-

plary NP IT solution.

Figure 4.11 – Exemplary integration of the currently used data sources in airline net-
work planning IT solution139

4.4 Performance measurement of airline network planning

The performance measurement in airline NP is closely connected to the

role and responsibilities of the NP department in the organization (Goede-
king, 2010, p. 99). Revenue, cost, or profit responsibility should define the
primary KPIs in airline NP; however, secondary KPIs too can complement

138
Among these, 66% use demographic data, 56% use real-time slot information, and 78%
use at least one external shock data type.
139
Source: Own illustration.
Performance measurement of airline network planning 189

the performance measurement. Section 4.4.1 introduces the different NP

responsibilities across the case study group and links it to the primary NP
targets. Section 4.4.2 then presents the currently used primary and sec-
ondary KPIs across the case study group.

4.4.1 Planning and optimization goals for NP departments

Airline NP departments can be charged with revenue, cost, and profit re-
sponsibilities (see section 3.2.4). In our case study group, five airlines as-
signed profit responsibility to their NP department, while three airlines
opted for cost responsibility and one chose revenue responsibility for this
(see Figure 4.12).

One of the SCAs follows a shared-revenue responsibility model, in which

revenue management and NP are jointly accountable for revenue genera-
tion. This makes the NP department a de-facto revenue center and re-
quires their close collaboration with P&RM. Both the LCCs and CAR 3
charge their NP departments with a cost responsibility, which necessitates
close engagement with flight operations and other operative departments.
All other airlines place NP as a mediator between revenue centers – usu-
ally marketing and P&RM – and cost centers. Thus, NP holds a global profit
responsibility that underlines its strategic importance for airlines’ financial
success.

The targets of an organizational unit should align with its role and respon-
sibilities. If a NP department is considered a profit center, the primary target
should ideally be profit maximization. Similarly, a cost center should strive
to minimize cost and a revenue center should maximize revenue. Across
our case study group, only seven airlines have aligned the stated primary
target of NP with the assigned role for the department.

All profit centers are measured against aggregate route profitability (which
excludes network effects) or network profitability. Here, the targets are
aligned with NP departments’ role as profit centers. In CAR 3, the sole
190 Status quo of strategic network planning in airlines

purpose of the NP department is to minimize operating cost to an accepta-

ble service level. Similarly, LCC 2’s NP department is tasked with maxim-
izing their asset utilization, hence minimizing unit cost. In all of these air-
lines, the targets are clearly aligned with the role of the NP departments.

Figure 4.12 – Performance management in NP departments140

LCC 1 assigns a cost center role to its NP department but ranks aggre-
gated route profitability as most important target. Although secondary NP
targets, such as aircraft utilization and operating cost minimization, are
aligned with the departments’ cost center role, the mismatch of role and
primary target may result in difficulties in prioritizing big data investments
for NP. SCA 2 exhibits an even greater mismatch in this regard. Despite
assigning a joint revenue responsibility to its NP department, none of the

140
Source: Own illustration.
Performance measurement of airline network planning 191

primary or secondary targets focus on revenue generation. The primary

target “asset utilization” is a cost target, and all secondary targets are cost
or profit targets as well.

In summary, the targets in airlines, barring a few, are well aligned with its
role and responsibilities. However, SCA 2 and LCC 1 present misalign-
ments that can potentially complicate the benefit analysis pertaining to big
data investments.

4.4.2 KPIs currently used for airline network planning

The primary cost, revenue, or profit objectives can be measured through a

variety of direct and indirect KPIs. While direct KPIs express the immediate
change in profit, revenues, and cost, indirect KPIs contribute towards the
desirable improvements in direct KPIs. Since no pre-defined KPI frame-
work for airline NP exists in the literature, a new framework has been de-
signed on the basis of general airline KPIs summarized by Sterzenbach et
al. (2013, pp. 389–416). This framework incorporates the KPIs used in the
NP departments of the case study group (see Figure 4.13).

Figure 4.13 – KPI framework for airline network planning141

Direct KPIs squarely contribute to the profit and loss statement of an airline
and are thus global performance measures within these organization. Air-
lines with a significant share of connecting traffic usually rely on network

141
Source: Own illustration following Sterzenbach et al. (2013).
192 Status quo of strategic network planning in airlines

profit calculations, whereas other airlines use aggregate route profit fig-
ures142. Revenues depend on the available seat kilometers, which is con-
sidered the unit measure of the total offered capacity controlled for route
length143, the average load factor, and the yield per seat kilometer. In sim-
ple words, revenues can be raised if more capacity is offered, a larger
share of capacity is sold, and the charged ticket price per kilometer in-
creases. Airline cost are typically segmented as direct and indirect operat-
ing cost and non-operating overhead cost (see section 3.4.4).

Direct KPIs are influenced by many different organizational units of an air-

line. However, indirect KPIs can typically be associated with one specific
department within the airline. NP departments across the case study group
feature four major indirect KPIs, which are mostly under the complete con-
trol of the respective NP department. If demand forecasting is in the remit
of NP, forecast accuracy is a crucial measure to assess planning quality,
which enables network optimization, capacity planning, and pricing.
Hence, it has an indirect effect on both profit and revenues. NP can also
indirectly affect cost KPIs in three ways. First, a schedule’s robustness re-
duces irregular operation events, thereby incurring high direct operating
costs. Significant delays or flight cancellations result in costly repayments,
re-routings, and accommodation cost for stranded passengers. Second,
high asset utilization reduces the capital cost per ASK, since the fixed cap-
ital cost can be distributed across more flown hours and thus more ASK.
Third, the NP department itself can become more efficient and reduce
overhead cost. A typical efficiency measure is the number of routes

142
See section 3.4.4 for a detailed discussion on the differences between network profit and
route profit.
143
Take a flight with an Airbus A320 from Berlin to London as example: 180 seats are flown
950 kilometers, which results in 180 * 950 = 171,000 ASK for this specific flight.
Performance measurement of airline network planning 193

planned/managed per full-time equivalent (FTE) 144 , which can be in-

creased by using more useful information, IT system support, and automa-
tion.

The most frequently used NP KPI among the case study group is net-
work/route profitability. It serves as the primary KPI for six NP departments
and as secondary KPI for one other (see Figure 4.14).

Figure 4.14 – Primary and secondary NP performance targets in case study airlines145

Asset utilization ranks second, being the primary target for two NP depart-
ments and secondary target in four airlines. The minimization of direct and
indirect operating cost is the primary KPI for CAR 3 and secondary target
for two more airlines. Forecast accuracy, FTE efficiency, schedule robust-
ness, and load factor are other subsidiary KPIs for NP in the airline case
study sample. Three potential KPIs are not employed at all, two of those

144
An FTE of 1.0 is equivalent to one full-time employee (40h per week).
145
Source: Own illustration.
194 Status quo of strategic network planning in airlines

being the revenue KPIs “ASK” and “yield” and another being the direct
overhead cost KPI. The fact that only one airline chooses a direct revenue
KPI for its NP department also validates the finding that NP departments
are considered profit or cost centers rather than revenue centers.
5 Big data opportunities for airline network planning

This chapter builds on existing research and practice to develop tangible

BDOs for airline NP. First, the information needs of airline network planners
are derived leveraging the currently used data sources and interview re-
sults from the case study group (sub-chapter 5). The BDOs that address
the identified information requirements are described in detail in sub-chap-
ter 5.2. In the last sub-chapter 5.3, the potential of BDOs to improve airline
NP is qualitatively evaluated by network planners, and the feasibility of im-
plementation is assessed. This chapter concludes with a classification of
BDOs based on their potential and feasibility.

5.1 Information needs for airline network planning

Data needs to be translated into useful information to generate value for

decision making (Ackoff, 1989, p. 4). In section 4.2.2, four major infor-
mation objects, which airline network planners require information on, were
identified. These can further be specified in information needs, which de-
scribe a cluster of logically similar information on a specific information ob-
ject. Section 5.1.1 defines the relevant information needs for this study. In
order to prepare the discussion on potential BDOs in sub-chapter 5.2, sec-
tion 5.1.2 associates currently used data sources to specific information
needs and indicates the gaps to be addressed with big data opportunities.

5.1.1 Definition of information needs for airline NP

Air transport demand is a broad concept involving all relevant actors in the
market. Hence, airline network planners need information on various
groups of actors as well as on the entire market system. Especially early
in the NP process, long-term demand forecasts depend on a holistic per-
spective of the entire transportation market, including the air transport mar-
ket as a specific sub-market. Furthermore, relevant trends in travel and
transportation must be identified and evaluated. This results in two con-
crete information needs, the holistic transportation demand forecast and a

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2020

M. Schosser, Big Data to Improve Strategic Network Planning in Airlines,
Schriftenreihe der HHL Leipzig Graduate School of Management,
https://doi.org/10.1007/978-3-658-27582-2_5
196 Big data opportunities for airline network planning

more specific trend identification. Since both information needs focus on

the entire market, the differences must be pointed out.
Transportation demand forecasts ultimately result in a concrete estimation
of the people willing to travel over a specific time window between a certain
O&D pair. The granularity of these estimates varies as do the forecast ac-
curacy. High-level transportation demand forecast estimates the number
of travelers across all transportation modes between large geographical
entities, such as countries or continents on a yearly base. Whereas, gran-
ular transportation demand forecasts distinguish demand according to
transportation mode and specify geographical areas down to individual
catchment areas and hourly time windows. The value for airline network
planners obviously increases with a higher degree of granularity and accu-
racy. Transportation demand forecasts tend to become more precise with
a decreasing time horizon. Airline network planners use this information in
all steps of strategic NP and even for tactical fleet assignment decisions
(see Figure 5.1).
Strategic network planning Tactical network planning

Demand Fleet Network Route Flight Flight Fleet Aircraft Crew Disruption
Information forecast planning selection frequency scheduling assignment routing assignment management
structure
objects

Transportation demand forecast

Trend identification

Demand Competitor monitoring

Real movements of passengers and goods

Internal booking situation

Cost Cost forecast

Transparency on planning constraints

Planning
constraints
Real-time monitoring of constraints

External
Early warnings for relevant incidents
shocks

Figure 5.1 – Information needs throughout the airline NP process146

146
Source: Own illustration.
Information needs for airline network planning 197

Trend identification focuses on systematic changes in the transportation

demand. On a higher level, this includes the identification of growth mar-
kets, mega-trends, and general travel preferences of customers. On a
more granular level, trend identification aims to discover the recurring mi-
cro-seasonality for specific destinations (e.g., due to fairs or holidays) or
the mismatches between accommodation and transportation supply for a
specific region. While the end result of transportation demand forecasts is
an estimated absolute demand figure, trend identification indicates the ex-
pected relative changes, thereby complementing the transportation de-
mand forecast. Thus, airline network planners mostly rely on trend identifi-
cation during the strategic NP phase up to the final scheduling.

Besides the information on the entire market, airline network planners re-
quire information on its main actors, mainly competitors, customers, and
the airline itself. Information on competitors can be pooled as competitor
monitoring. Competitor monitoring focuses either on regular or irregular
operations. Information on regular competitor operations includes the
backward-looking analysis of actually flown flights, delays, traffic volume,
and reported yield. Forward-looking information consists mainly of pub-
lished schedules, real-time fare data, and news analysis of planned routes
or fleet decisions. The monitoring of regular operations is most relevant for
strategic NP, as it informs an airline’s route and schedule decisions. In
contrast, irregular operations monitoring aims to exploit the short-term
changes of a competitor’s operation. This includes strikes, flight cancella-
tions, and other unforeseen events that may justify the offering of additional
flights or changes in fleet assignment. Consequently, information on irreg-
ular competitor operations are most relevant for tactical NP.

Information on airline customers has become possibly the most important

need of P&RM departments (Chen et al., 2017, p. 25). While the focus of
P&RM is mainly on understanding the customers’ individual preferences,
airline network planners try to understand the granular movement patterns
of passengers and cargo. These so-called real movements of passengers
198 Big data opportunities for airline network planning

and goods provide complete information on the real origin and the real
destination of a passenger or a cargo item. Ideally, airline network planners
should wish to know the exact address of both origin and destination for
every passenger and every cargo item. Furthermore, even the travel pat-
terns of non-airline customers are highly relevant for the identification of
potential routes that can compete with railway and individual traffic. The
real movements of passengers and goods are most relevant to the strate-
gic planning steps, from route selection to flight scheduling. Cargo airlines
especially benefit from the real-time location information of cargo items
when a delayed delivery requires short-term re-planning.

The internal demand perspective of airlines is mostly shaped by the inter-

nal booking situation. Based on their experience and the current forecast,
airlines define expected booking levels for specific flights during a defined
period before departure. If the bookings are significantly above or below
their expected level, schedule adjustments become necessary. The poten-
tial actions to adjust the schedule range from a permanent change in flight
frequency to the addition or cancellation of individual flights up to a change
in fleet assignment to reflect the specific internal demand. Accordingly, in-
formation on the internal booking situation is of utmost relevance in the
mid- to short-term airline NP.

The information object “cost” is typically associated with the need for only
one type of information: the cost forecast. Cost forecasts are typically struc-
tured in line with the airline cost classification scheme (Niehaus et al.,
2009, p. 177). Because overhead costs are not in the remit of NP, the need
for information is focused on direct and indirect operating cost. Most air-
lines utilize internal cost calculations to satisfy this need for information.
Cost forecasts are relevant in all the strategic and tactical NP process
steps, as they are required for any profitability analysis.

Information on planning constraints can be distinguished as two specific

information needs. Transparency on planning constraints is concerned
with the collection and organization of all relevant information for the NP
Information needs for airline network planning 199

process. This includes airport and aircraft characteristics, an updated list

of available slots, and existing traffic right agreements. Airline network
planners apply this information during the strategic NP phase to avoid tech-
nical planning mistakes147. The real-time monitoring of planning constraints
checks for both the short-term changes in existing constraints (e.g., tem-
porary runway closures) and the availability of slots and traffic rights for all
tactical NP decisions. This need for information is relevant when airline
network planners adjust the planned schedule by changing aircraft types
or re-scheduling flights.

Finally, external shocks create the need for early warnings for relevant in-
cidents. Airline network planners are primarily concerned with the political
and meteorological events that may have a permanent impact on the travel
demand or passengers’ capability to fly to a specific destination. Political
events include long strikes, travel restrictions, or conflicts that affect the
stability of a destination. Meteorological events are usually less relevant
for NP, since they usually cause only short-time interruptions. However,
major catastrophes such as floods, tsunamis, or tropical storms may influ-
ence travel demand in the long term. Thus, early warnings for relevant in-
cidents are relevant to both strategic and tactical NP, depending on the
impact and severity of the shock.

5.1.2 Satisfaction of information needs with current data use

In this section, the currently used data sources are assigned to the infor-
mation needs derived in the previous section. Figure 5.2 visualizes the al-
location of currently used data sources to information needs. The useful-
ness and reliability classifications are included as the shaded areas of the
data sources.

147
Exemplary technical planning mistakes: aircraft too heavy for airport, night curfew not re-
spected, aircraft range not sufficient for stage length, slot or traffic rights unavailable.
200 Big data opportunities for airline network planning

External
Information Demand Cost Planning constraints shocks
object
Information Transpor- Trend Competitor Real move- Internal Cost Transpa- Real-time Early
need tation identifica- monitoring ments of booking forecasts rency on monitoring warnings
demand tion passengers situation planning of for
forecast and goods constraints planning relevant
constraint incidents
s
Currently MIDT data Cost data Political
used data
Information on unconstrained demand Airport data events
types
Market estimates Service Oil price Aircraft Aircraft
bench- forecasts data data
Economic data
marking
Demographic data data Traffic Real-time Meteoro-
rights slot logical
Real time GDS data
availability events
Published flight schedules Airline slot
portfolio
Real traffic data Competito
r-related
Fare data Internal booking data
events

Data essentials Data dreams Nice-to-have data Hands-off data

Figure 5.2 - Allocation of currently used data sources to airline NP information needs148

Transportation demand forecast

The current transportation demand forecasts are mostly developed around

ready-made market forecasts and MIDT data, which are enriched by as-
sumptions on economic development and demographic factors. One of the
airlines uses clickstream data from booking websites to model their uncon-
strained demand, which is already considered a big data type. The useful-
ness of all five currently used data sources is rated as high; however, the
reliability varies widely (see Figure 4.6, p. 180). Demographic data, MIDT
data, and real-time GDS data are considered fairly reliable, whereas ready-
made market forecasts, economic forecasts, and clickstream data from
booking websites are considered rather unreliable. Furthermore, MIDT and
GDS data cover only part of the market, namely the tickets sold via Global
Distribution Systems. All of the data sources, excepting unconstrained de-
mand, deliver data on actual bookings but not on the booking intentions.

148
Source: Own illustration.
Information needs for airline network planning 201

Thus, there are multiple big data opportunities to improve the need for in-
formation on transportation demand forecast. First, BDOs may improve the
reliability of economic forecasts, general market estimates, and uncon-
strained demand. Second, the data pool can be extended using data on
booking intentions to predict actual bookings better. Third, a suitable proxy
for real-time data on cargo demand is urgently required, as MIDT and GDS
data cover passenger bookings only.

Trend identification

Trend identification relies on the same data sources as transportation de-

mand forecast. In addition, airlines reference published flight schedules,
real traffic data, and fare data to perform ex-post analyses and create time
series in order to discover any relevant change. All currently used data
sources exclusively contain airline-specific data. However, travel trends
are often influenced by factors independent from the airlines, such as in-
creasing tourism interest in a destination or specific events, such as fairs
or concerts. Big data can add an outside-in perspective to complement the
currently used data sources. In addition, big data needs to improve the
reliability of the five data sources in the “data-dream” cluster (see Figure
5.2).

Competitor monitoring

Present day competitor monitoring occurs on multiple dimensions. Past

traffic and fare data are evaluated through ex-post competition analyses.
The current offering can be compared based on published schedules and
service-benchmarking data. Bookings of competitors can be analyzed ex-
post employing MIDT data or real-time GDS data. Flight schedules and
traffic data appear to be the weak links, presenting great improvement po-
tential in data reliability. Short-term schedule changes and aircraft type
changes are usually not recorded in published flight schedules nor are
charter flights. Big data can address these reliability issues by providing a
complete picture on performed flights, complementing the currently used
data sources.
202 Big data opportunities for airline network planning

Real movement of passengers and goods

Most data sources provide only the air travel–specific origin and destina-
tion, which includes the first and the last airport on the itinerary. MIDT data
also contains the booking agent, but neither any personal address nor the
internet protocol (IP) address used while making the booking. If a booking
is made through the airline’s website, most airline can locate the IP ad-
dress, which indicated either the real origin or the real destination. How-
ever, no current data type can provide comprehensive real O&D infor-
mation. Cargo airlines usually have the real shipping O&D for individual
shipments, since it is a part of the air waybill. However, many freight for-
warders consolidate individual shipments and issue new air waybills for
such shipments. The shipping addresses are then the forwarder’s airport
facilities, which do not provide any further information on the real O&D of
the consolidated cargo. Both passenger and cargo airlines can use big
data opportunities to fill this gap and produce reliable information on the
real O&D of their customers.

Internal booking situation

The internal booking situation is analyzed according to the airline’s own

booking data. In many airlines, P&RM has already established sophisti-
cated analytics and forecasting tools, often linking their booking data to a
360-degree view of the customer. Furthermore, the airlines in the case
study group consider real-time booking curves as both highly reliable and
useful. There exist big data applications for hyper-granular customer pre-
diction. However, improving the analysis of the internal booking situation
does not appear to be a priority of airline NP departments.

Cost forecasts

Two of the three main data sources are highly reliable but of limited utility.
Both operating cost and airport fees are considered must-have factors by
airline network planners, since these need to be included in the route and
network profitability calculations. However, they are not in the remit of the
Information needs for airline network planning 203

NP departments. Flight operations departments across airlines are cur-

rently investing heavily in big data applications in order to reduce operating
cost; however, little is invested to improve cost forecasts. From the airline
NP perspective, there seems to be little need to invest in BDOs to improve
cost forecasts.

Transparency on planning constraints

All of the currently used data sources are highly reliable for airline NP.
However, many interview participants confessed that the data sources are
not connected, and the actual useful information has to be compiled man-
ually. Big data not only consists of new data sources but also a combination
of existing data sources. Therefore, there is great big data harmonization
and automation potential in planning constraint data. The harmonization
aspect is considered part of the real-time planning constraint solutions,
which are introduced subsequently.

Real-time monitoring of planning constraints

At present, the real-time monitoring of planning constraints is limited to the

use of slot information systems, which mostly cover airports in free market
economies149. These platforms are meant for information purposes only; a
trading or swapping function does not exist. Furthermore, there is no
means to track the use of traffic rights, e.g., for overflight contingents.
These shortcomings of the current data sources are concrete opportunities
to establish big data solutions and add a new real-time dimension in plan-
ning constraint monitoring. At the same time, a real-time platform should
link all the available existing data sources to guarantee transparency in the
status quo of planning constraints.

149
ACL International covers UK, Canada, Ireland, Luxemburg, New Zealand, Poland, and the
United Arab Emirates. e-Airportslots covers Australia, France, Belgium, Germany, Nether-
lands, Philippines, Spain, Tunisia, and Mexico.
204 Big data opportunities for airline network planning

Early warning systems for relevant incidents

The current early warning systems on meteorological and political events

have been proved to be both unreliable and useless. While meteorological
warning systems may benefit from a larger time horizon150, political warn-
ing systems must deliver more airline-specific results. New big data tech-
nologies, such as natural language processing, may be able to increase
relevance significantly. The same applies for competitor information, which
is currently compiled via daily news briefings.

5.2 Development and evaluation of BDOs for airline NP

A two-step idea generation process identified potential BDOs for airline

NP. First, the big data types in the literature were reviewed and contrasted
to news articles and reports on the big data types used in other airline
functions. In the second step, concrete ideas are developed to address
specific information needs in preparation for the second interview round
with airline network planners. Section 5.2.1 demonstrates the development
of 23 specific BDOs. These are subsequently detailed in sections 5.2.2–
5.2.7, which are structured as per the primarily addressed information
needs.

5.2.1 Specific BDOs for airline NP information needs

Yaqoob et al. (2016, p. 1234) elaborated on social media, the Internet of

Things (IoT), mobile device data, and other web-based content as primary
big data types. These data types have emerged over the past 15 years.
They produce large volumes of both structured and unstructured data,
which can often be transmitted in real time. Traditional “small” data types

150
The current systems warn for specific meteorological events, such as a thunderstorm or a
hurricane. However, a generalized analysis, such as one on the occurrence of fog at a specific
airport, can indicate specific times of day during certain months with a high fog risk, thus
informing scheduling.
Development and evaluation of BDOs for airline NP 205

complement the big data types. The airline-specific big data types identi-
fied in section 3.6.5 can be classified under these categories (see Table
5.1).
Table 5.1 – Classification of airline big data types151

Airline big data Social Internet Mobile Other web- Traditional

type media of Things device based con- small data
data data tent
Flight tracking x
data
Airport data x
Meteorological x
data
Airline info data x x x
Market data x x x
Passenger data x x x x
Aircraft data x x
Air Safety Re- x
ports
Social media x
data
Mobile location x
data
Search engine x
data

Social media comprise services such as social networks, (micro-)blogs,

and review platforms, which present valuable information on passengers,
competitors, and the general market development. Social networks contain
an abundance of demographic data, which can be accessed to improve
granularity and reliability of demographic customer profiling beyond aggre-
gated statistical bureau data. Social media data can also be used to im-
prove economic forecasting. Consumer confidence can be measured in
real time based on unstructured social network and social blogging data

151
Source: Own illustration.
206 Big data opportunities for airline network planning

(Tuckett et al., 2014). Social media data can then be used in combination
with search engine data to determine the specific or aggregated consumer
price elasticity (Ghose, Ipeirotis, & Li, 2018). Travel-specific social media,
such as travel rating platforms or travel blogs, present valuable data on
transportation streams. Reviews and visitor volumes can indicate shifts in
travel trends (Bowen, 2015). Clickstream data152 from travel rating plat-
forms can also reveal travel patterns of consumer groups. The analysis of
specific events in social networks may yield a better prediction of one-time
travel flows, e.g., for large concerts or industry fairs.
The IoT provides multiple BDOs, especially for air cargo network planners.
The location transmitting sensors attached to individual shipments can re-
veal the real O&D and further indicate short-term delivery delays. Alterna-
tively, a cheaper RFID-chip can be attached to cargo carrying an improved
electronic air waybill. Flight tracking data can complement competitor mon-
itoring and fill gaps in MIDT data and published flight schedules. The in-
creasing number of weather-tracking devices can improve the usefulness
of long-term meteorological forecasts. Similarly, machine learning-based
image recognition tools can use geo-spatial images to predict the impact
of extreme meteorological events and even improve economic forecasting.
Aggregated and anonymized mobile location data can draw a seamless
picture of all travel movements, independent from the chosen transporta-
tion mode. Airline network planners can discern a real-time view on travel
behavior, market shares, and real O&Ds.
Furthermore, there are multiple BDOs using other web-based content. The
most prominent opportunity is conducting an analysis of search engine
data to predict future bookings. This can either be based on general search
engine data (e.g., Google, Baidu) or on specific clickstream data from
meta-search websites for flights (e.g., Kayak, Skyscanner). Another im-
portant data source is e-commerce platforms (e.g., Amazon, Alibaba).

152
A clickstream is the sequence of hyperlinks followed by a unique website visitor. The click-
stream is presented chronologically and allows retracing the search pattern of customers.
Development and evaluation of BDOs for airline NP 207

Data on cross-border orders can reveal trends in air cargo demand from
e-commerce; this is predicted to become a major air cargo growth driver in
the coming years (IATA, 2017b). Moreover, e-commerce pricing data can
be used to track inflation in real-time in order to better inform economic
forecasts (Cavallo & Rigobon, 2016).
Improved scraping153 technology enables the collection of information from
different web-based services. Airlines can analyze daily hotel prices to
identify micro-seasonality in specific destinations or identify a supply mis-
match between transportation and accommodation. Similarly, cargo air-
lines can collect data on industry fairs and related exhibitors in order to
predict cargo flows. Online forwarders quote public rates on the major ship-
ping lines, which can indicate the expected booking situations in a manner
similar to passenger airline fare intelligence. The scraping of unstructured
data in combination with natural language processing (NLP) analytics can
vastly improve news monitoring and political risk assessments.
Two of the BDOs identified do not fit the classification scheme presented
in Table 5.1. First, an integrated real-time slot trading platform can facilitate
short-term schedule adjustments and decrease the need to freeze the
schedule before the IATA slot conferences. A real-time traffic right infor-
mation platform can bundle all the relevant information on global traffic
rights and indicate potential violations, e.g., overusing an overfly contin-
gent. In total, 23 BDOs have been identified. Figure 5.3 schematizes the
BDOs by their primarily addressed information need154.

153
Scraping describes a technique in which a computer program extracts data from human-
readable output, coming from another program or website.
154
Please refer to Appendix 17 for a different view structured by the NP process steps.
208 Big data opportunities for airline network planning

Information needs
A B C D E F G H I

Real-time monitoring of
Transportation demand

passengers and goods

Competitor monitoring

planning constraints

planning constraints
Real movements of
Trend identification

Early w arnings for

relevant incidents
Transparency on
Internal booking

Cost forecasts
situation
forecast
Big data
opportunities

A1 Geo-spatial image analysis     

A2 Real-time inflation forecasts   

A3 Consumer confidence from social media  

A4 Detailed demographic profiling with social networks 

A5 General search engine data    

A6 Click-stream data of meta-search websites     

A7 Social media mining to determine price elasticity  

A8 Analysis of rates from online air cargo portals    

A9 Analysis of exhibitor data of industry fairs   

A10 Shipping data from e-commerce platforms     

B1 Analysis of reviews in travel rating communities 

B2 Click-stream data from travel platforms  

B3 Analysis of local hotel prices   

B4 Analysis of events in social networks 

C1 Real-time flight tracking with transponder data    

C2 Network-relevant competitor news monitoring  

D1 Mobile location data from smartphones    

D2 Extended electronic AWB with shipper address  

D3 Location-transmitting sensors for cargo shipments   

H1 Real-time slot trading platform  

H2 Real-time traffic right information platform   

I1 Global meteorological warning platform   

I2 Global political risk assessments  

 Primary information need addressed  Secondary information need addressed

Figure 5.3 – Addressed information needs of BDOs155

Since many of the BDOs address more than one specific need for infor-
mation, the secondarily addressed information needs are also displayed in

155
Source: Own illustration.
Development and evaluation of BDOs for airline NP 209

light grey. In line with the argumentation presented in the previous sub-
chapter, no BDO is included for internal booking situation and cost fore-
casts. The transparency in planning constraints is addressed as a second-
ary need for information in real-time platforms.
The following sections, which are organized according to the addressed
information needs, describe each BDO in detail.Appendix 18 summarizes
the big data properties of each opportunity and indicates the potential data
providers and applicable BDA methods to generate useful information from
the data.

5.2.2 Transportation demand forecast

A1 – Geo-spatial image analysis

The current economic forecasts mostly consist of econometric extrapola-

tion models, using indicators at the country or regional level. The most
widely used economic growth indicator – GDP – is calculated on the basis
of a variety of secondary indicators, such as industrial production and un-
employment rate (Angelini et al., 2011, C30). This bears three shortcom-
ings for its use in airline NP. First, ensuring a high data quality of the input
indicators requires a thorough statistical recording of economic activity.
Developing countries often do not have acceptable data quality standards
in place to support meaningful economic growth forecasts (Klein, 2001,
pp. 65–69). Second, current growth forecasts are rarely available on the
local level. Especially in the route selection process, accurate local eco-
nomic activity forecasts can inform the choice between several airport can-
didates within a shared catchment area. Third, global growth forecasts are
210 Big data opportunities for airline network planning

usually updated only on a quarterly base156. Thus, these changes in eco-

nomic activity are only incorporated after a significant time lag.

The analysis of geo-spatial images, which usually are captured using sat-
ellites but more recently also with drones, can help overcome all three
shortcomings of traditional growth forecasting. In the scientific community,
Henderson, Storeygard, and Weil (2012) pioneered the approach and
showed that the nighttime light intensity analysis is a better predictor for
economic growth in developing countries than traditional forecasts. Simi-
larly, Jean et al. (2016) have predicted poverty levels with this approach.
Keola, Andersson, and Hall (2015) have contributed land cover data as an
additional image-based indicator for economic forecasts. Gennaioli et al.
(2014) proved that geo-spatial image analysis generates robust economic
forecasts on sub-regional level.

Several start-ups are now commercializing geo-spatial imaging data. The

Japanese company Nowcast uses nighttime data to estimate sub-regional
economic growth (Allan & Ito, 2016). The American start-ups Orbital Insight
and SpaceKnow have developed industry-specific image analysis applica-
tions, including hyper-granular economic forecasts (Komissarov, 2017). All
of these companies use supervised machine learning to inform the algo-
rithms to recognize specific patterns within satellite or aerial images. The
list of recognizable items is becoming longer every day. Orbital Insight an-
alyzes mass retail revenue by counting cars in supermarket parking lots or
construction activity by the deployment of construction cranes (Orbital In-
sight, 2018). SpaceKnow predicts trade flows by counting the shipping
containers in port storages and truck movements on highways (Space-
Know, 2018).

156
The International Monetary Fund updated the World Economic Outlook quarterly, Word
Bank, and OECD every six months.
Development and evaluation of BDOs for airline NP 211

The primary use case for airline NP is the precise analysis of local eco-
nomic activity, which can improve both the reliability and usefulness of eco-
nomic forecasts. Furthermore, a geo-spatial image analysis can reveal the
travel and trade trends by counting vehicles in tourist destinations or ships
and trucks on trade lanes. If the frequency of imaging is further in-
creased157, image analysis can be used to monitor competitors by reveal-
ing the position of spare aircraft, maintenance activities, or cargo shipment
routing. Similarly, real O&D monitoring can become feasible, perhaps more
so for cargo than for passengers. Outsized cargo can be tracked from the
real origin to destination and reveal all the involved transportation provid-
ers. Several companies already use this technology for their early warning
systems for political risks and natural disasters, which can also become
relevant for airline NP (Komissarov, 2017).

Geo-spatial imaging data is of utmost relevance for the NP process steps

of demand forecast and route selection. Demand forecast can be based
on much more granular economic forecasting, allowing for more accurate
O&D-specific forecasts. Route selection can benefit as granular growth
data informs idea generation (e.g., what is the next best Chinese airport to
serve) and airport decisions in large agglomerations with multiple airports.

A2 – Real-time inflation forecasts

High inflation rates directly influence the air transportation demand and in-
directly endanger revenues if tickets are sold in local currencies (Wadud,
2014). Inflation measurement and forecasts are often unreliable, especially
in countries with high inflation rates. In a sample of Latin American coun-
tries, Jalles (2017) found that inflation forecasts are highly volatile and not
dependable. Extreme cases, such as that of Venezuela, show that pro-
longed high levels of inflation significantly affect the demand for air travel,
as it lowers the purchasing power of the population (Echenique, 2018).

157
The current maximum data frequency is daily (Orbital Insight, 2018).
212 Big data opportunities for airline network planning

Furthermore, the official inflation statistics are not reliable in some coun-
tries.

Driven by the manipulated official inflation statistics in Argentina, research-

ers at the Massachusetts Institute of Technology (MIT) began recording
online retail prices to monitor inflation (Cavallo & Rigobon, 2016). The so-
called “Billion Prices Project” has been expanded subsequently, and it is
now providing daily inflation figures for 22 countries based entirely on
online retail data (The Billion Prices Project, 2018). The data is commer-
cialized by the spin-off PriceStats (PriceStats, 2018).

Real-time inflation measuring and forecasting can improve both the use-
fulness and reliability of economic forecasts. The data can also reveal
short-term trends and thus serve as an early warning system for political
risks, e.g., for protests due to high inflation or poor economic development.
Airline network planners can anticipate routes with high demand risk better
due to inflation and thus adjust flight frequencies or the route mix. It impacts
the three NP process steps of demand forecasting, route selection, and
flight frequency determination.

A3 – Consumer confidence measurement with social media

Consumer confidence indices are the central input factors for GDP growth
forecasts (Angelini et al., 2011, C30). Currently, consumer confidence is
measured through representative polls conducted on a monthly basis by
statistical bureaus and research facilities (Dominitz & Manski, 2004,
pp. 52–53). Since regular large-scale polls are expensive and fail to meas-
ure real-time results, scholars have proposed to employ the sentiment
analysis of social media postings to measure consumer confidence in real
time. Daas and Puts (2014) have shown that the sentiment analysis of
Dutch Twitter and Facebook posts resembles the traditional consumer
confidence index while delivering faster results. Similarly, Tuckett et al.
(2014) utilize the sentiment analysis of broker and analyst reports to meas-
ure business confidence.
Development and evaluation of BDOs for airline NP 213

Despite scientific pioneering, there are no commercial providers of a real-

time consumer confidence index. Nevertheless, airlines can use the pub-
lished methodology to build their own consumer confidence index for coun-
tries, but that does not provide sufficient reliable data. This data can com-
plement the assumptions regarding consumer confidence in economic
forecasts on the country level to improve their reliability. Furthermore,
trends in changing consumer confidence can be detected easily to adjust
economic forecasts instantaneously. This would directly impact the de-
mand forecasting process step, although there may be little impact on other
steps.

A4 – Detailed demographic profiling with social networks

Airlines currently use a variety of demographic indicators to estimate travel

demand from the O&D level (Itani, O‫׳‬Connell, & Mason, 2014, p. 129). One
SCA of the case study group utilizes the ancestry data provided by Amer-
ican airports to estimate source demand to Europe, since a large propor-
tion of Americans travels to their ancestor’s home country. One LCC uses
official statistical data on migration background in order to identify routes
for ethnic travel. Other airlines use trade data to estimate the extent of
business connections between two different regions. However, the current
sources of demographic data are fragmented and of varying quality. Data
provided by airports may differ by methodology and definitions, which
makes it difficult to compare across route candidates. Data from official
statistical bureaus is usually reliable but restricted to larger geographic en-
tities, and they sometimes miss important variables. In Germany, for ex-
ample, a breakdown by country of origin is only available for foreigners
living in Germany but not for naturalized Germans (Statistisches Bun-
desamt, 2017). A German of Kurdish origin, for example, may still travel
frequently to Turkey or Iraq to visit his extended family. For business travel,
trade volumes are only a rough proxy for the actual business travel.

Scientific research on innovative demographic data has concentrated

mostly on extraction methodologies from social networks (Filippova, 2012)
214 Big data opportunities for airline network planning

and micro-blogging services such as Twitter (Sloan et al., 2015). Despite

the increasing importance of ethnic travel in the aviation industry (Itani,
2017), there exists no scientific study on the use of social media to improve
the current data sources of demographic data.

Three categories of demographic data are most relevant for airline NP,
namely ethnic background, business contacts, and economic status. Large
social networks, such as Facebook, contain vast information on ethnicity
and economic status. Professional social networks, such as LinkedIn, con-
tain data on direct global business contacts. Facebook offers a free ana-
lytics tool called “Audience Insights” that allows the segmentation of the
targeted audience through both demographic and socio-economic factors
(Facebook, 2018a). The user can either segment a specific audience (e.g.,
the followers of his/her page) or the entire public audience. It offers seg-
mentation criteria, such as language, gender, and educational background.
Until August 2018, the data could be extracted through an open application
programming interface (API)158, but Facebook discontinued this API after
a massive data scandal (Johnson, 2017). Now, companies have to use the
integrated “Audience Insights” tool to process raw data.

LinkedIn currently does not commercialize the network connection raw

data but only pre-filtered audience data for marketing and recruiting pur-
poses (Cohen, 2016). However, LinkedIn provides researchers with net-
work raw data if they join the LinkedIn research platform (Agarwal, 2017).
Since data on global business contact is valuable to many business mod-
els, LinkedIn may start commercializing this data as well. Besides social
networks, there is a group of aggregator companies that facilitate data
management and analytics across platforms159. Most companies started

158
An open API, also known as a public API, is an application programming interface that
allows the owner of a network-accessible service to give universal access to other develop-
ers.
159
The main competitors include SproutSocial, Spredfast, agora Pulse, Salesforce Social
Studio, and Socialbakers.
Development and evaluation of BDOs for airline NP 215

as cross-platform marketing tools, which then developed analytical func-

tionalities. However, almost all companies only offer analytics for a com-
pany’s own social media audience and not on the entire user population.
SproutSocial is currently developing a cross-platform analytics solution,
but the launch date is yet to be announced (York, 2018).

Airline network planners can use demographic data from social networks
in various ways to improve the usefulness and reliability of current demo-
graphic data. Data from global platforms and LinkedIn allow for seamless
analyses across locations, thereby overcoming the issue of varying data
quality from different data sources. Moreover, demographic data can be
narrowed down to the city level (Facebook, 2018a), allowing for a greater
data granularity. Filtering by language or interests may yield a much more
accurate picture of people with a migration background than official statis-
tics. If professional networks, such as LinkedIn, begin marketing direct
business contact data, the demand for business travel may be forecasted
more accurately. All of this affects the need for information pertaining to
transportation demand forecast. Demographic data is mostly utilized in the
strategic NP process steps from demand forecast up to route selection.

A5 – General search engine data for detailed O&D demand

All the booking data that airlines currently use for demand forecasting is
based on actual bookings, even if it is updated in real time. Thus, it reflects
booking decisions that have already been made. Many potential passen-
gers search for flights in search engines before they actually process the
booking. Analyzing search engine data could thus indicate future booking
intentions, which makes it a truly future-oriented data source. In addition,
search engine data can indicate real O&D demand, since people are likely
to search for their real destination. IP analytics allow for the detection of
the geographic origin of the search, so that airlines have indications for
both the real origin and the real destination.

Search engine data has attracted significant attention in the scientific re-
search community. Disease prediction was the first proven application
216 Big data opportunities for airline network planning

(Lazer et al., 2014), although the range of opportunities is infinite (Choi &
Varian, 2012). Carrière-Swallow and Labbé (2013) employed Google
Trends data to measure economic indicators in emerging markets in real
time. In tourism research, Google Trends has been widely used to predict
tourist flows (Önder & Gunter, 2016; Park, Lee, & Song, 2016; Yang et al.,
2015) and hotel room demand (Pan, Chenguang Wu, & Song, 2012; Ri-
vera, 2016). Indeed, there is no research on predicting air travel demand
with search engine data, but a similar methodology as for hotel demand
forecasts may be applicable.

All major research is conducted using Google data, which can be explained
by its high market share of about 90% in the global search engine market
(StatCounter, 2018). Google offers its proprietary “Google Trends” webtool
to analyze search volumes. It allows the comparison of different search
terms, which are displayed in a normalized scale. One can obtain absolute
search volumes using the Google Adwords plug-in “Keywords.” The search
statistics can be filtered geographically down to the city level in North
America or to the regional level in other continents (Google, 2018).
Searches can then be displayed as a time series with daily search data for
up to ten years. The data is also available through a Google Trends API
(npm Inc., 2018).

Airlines can combine the search term “flight” with a list of potential destina-
tions and filter these results by the intended origin geographies. The results
should indicate changing preferences and seasonal patterns. The data
needs to be controlled for different booking lead times and conversion rates
for different O&Ds. On some routes, people may search and book several
months ahead, whereas on other routes, the lead time is much shorter. In
any case, search engine data indicates booking intentions and may thus
be a highly valuable complement for actual external and internal booking
data. The broadness of information comes at the cost of specificity and
reliability. Google Trends data have to be adjusted to booking lead times,
Development and evaluation of BDOs for airline NP 217

conversion rates, and search terms. This requires significant data pro-
cessing before it becomes reliable.

Network planners can use general search engine data to fine-tune demand
forecasts and identify unserved markets. Furthermore, this data can reveal
trends in booking and travel behavior. Due to its general character, this
data contributes to various information needs. Primarily, it can improve
transportation demand forecasts and also inform trend identification and
the internal booking situation. In combination with competitor search terms
(e.g., Aeroflot + Moscow), it can enrich competitor monitoring and serve as
an early warning indicator. The broad range of applications makes the data
valuable for almost all NP process steps. Long-term demand forecasts can
benefit from 10-year trend patterns, while mid- and short-term forecasts
can be directly informed by concrete booking intentions. This not only af-
fects the strategic process steps like route selection and flight frequency
determination but also the tactical scheduling and fleet assignment deci-
sions.

A6 – Clickstream data of meta-search websites for flights

Meta-search websites for flights can provide much more focused data than
general search engines. However, the market is fragmented and features
multiple players, such as Skyscanner, Kayak, and Expedia (Reals & Had-
wick, 2015). Since every player can commercialize only their own click-
stream data, these datasets have limited market coverage. In turn, the data
contains complete search information, including set filters, clickstreams,
and IP addresses (Skyscanner, 2018). The IP address can indicate the
real origin of the passenger.

There is no recent literature on the use of clickstream data from meta-

search engines, which may be due to the recency of data availability.
Around mid-2018, only Skyscanner had commercialized its clickstream
data to airlines. Skyscanner established its “Travel Insights” platform in
2016 and entered into a partnership with the air route portal Anna.aero to
promote the “unserved route of the week”, a weekly feature identifying the
218 Big data opportunities for airline network planning

next-best direct flights according to the Skyscanner clickstream data

(Anna.aero, 2017). Their direct competitor Kayak, a brand of Booking Hold-
ings, launched a “Travel trends” feature in 2018. This service does not pro-
vide actual clickstream data but only processed trending searches and
destinations (KAYAK, 2018). Expedia offers “creative partnerships” for
customized data solutions (Expedia Group, 2018) but no standard click-
stream product that is comparable to the Skyscanner solution.

The clickstream dataset allows airline network planners to identify O&Ds

by search volumes, thereby revealing unserved potential non-stop routes.
In addition, airline network planners can retrace how many passengers
have booked with competitors or have not booked at all160. All data is trans-
mitted daily, so that booking intentions can be tracked in nearly real time.
If a sample can be generalized based on the assumed market coverage
and possible biases, such as a skew towards leisure travelers, airlines can
model O&D demand much more granularly than with the existing market
forecast models.

Clickstream data from meta search websites can improve the usefulness
and reliability of transport demand forecasts by increasing the granularity
and specificity of O&D demand estimates. If the data is compiled to a time
series, airline network planners can also identify demand trends. The click-
stream data supports competitor monitoring and market share modelling,
since competitor bookings are visible in this data. At the same time, internal
bookings can be traced. With the IP address included, the real origin of
passengers may be approximated. This information primarily serves the
demand forecast as well as route selection and scheduling. Scheduling,
especially, can be improved through this, since clickstream can contain
preferred departure and arrival time windows. Fleet assignment can also

160
All the results can also be refined by the filters that users have set, e.g., to identify price-
sensitive or time-sensitive customers.
Development and evaluation of BDOs for airline NP 219

be informed by improving the forecast accuracy of the internal booking sit-

uation.

A7 – Social media mining to determine price elasticity

Traditionally, price elasticity in airlines was calculated based on published

ticket data (e.g., the US Department of Transportation DB1B database) or
MIDT-data (Brons et al., 2002). Price elasticity was determined on a route
level and based on the total fare, omitting ancillary revenues. With the in-
creased unbundling of air fares, researchers strived to calculate the price
elasticities for individual service items (Granados et al., 2012). The availa-
bility of booking data from online booking portals and airlines’ internal book-
ing systems allow for elasticity calculation at the flight level (Mumbower,
Garrow, & Higgins, 2014). The next granularity step is calculating the price
elasticity on an individual customer level, which was pioneered by Ghose
et al. (2018) on a Google Product Search dataset. They enriched a struc-
tural econometric model with unstructured social media data in order to
understand individual customer preference factors. These preference fac-
tors were then mapped to a set of hotel rooms reserved by the same cus-
tomer group to determine the price elasticity. Airlines can use a similar ap-
proach to estimate the individual price elasticity of their own customer
base.

In general, price elasticity in airlines is crucial for P&RM departments. How-

ever, the price elasticity informs revenue estimates for route forecasts and
is thus also vital for airline NP. Network planners can prioritize destinations
with relatively low price sensitivity and adjust flight frequencies and fleet
assignment accordingly. Besides adding an additional layer of granularity,
price elasticity at the customer level improves the interpretation of the in-
ternal booking situation, as abnormal booking levels may be correlated
with changing price elasticities.
220 Big data opportunities for airline network planning

A8 – Analysis of air cargo rates from online freight portals

While passenger airlines can rely on the real-time price intelligence of es-
tablished providers, such as Infare or DataCrops, cargo airlines only have
access to their internal accounting data on freight rates and to the market
rate levels published one month after being cleared by the IATA or
WorldACD. Rates between forwarders and cargo airlines have long been
negotiated bilaterally, so there was no data source to access real-time
prices. With the digitization of the logistic transportation chains, cargo rate
intelligence providers have entered the market in recent years and are try-
ing to bring transparency similar to those of passenger airlines.

Digitization is changing the air cargo industry at lightning speed. Online

freight forwarders, such as Freighthub or Instafreight, are challenging the
established freight forwarding companies (Vaske, 2017). They offer not
only tailored services but also price transparency to the shipper. Moreover,
they offset volume disadvantages with slim service charges due to a lean
digital business model. Many incumbents, such as DHL, have reacted and
now offer online quotes as well; cargo airlines, such as Lufthansa Cargo
(Lufthansa Cargo, 2018a) and Emirates Sky Cargo (Emirates Sky Cargo,
2018), have also started to circumvent the issues with traditional forwar-
ders and enable direct quotes and bookings. The availability of online
quotes is fueling the business model of freight rate comparison websites,
such as Freightos (Carey, 2017) and Xeneta (Air Cargo News, 2017). With
an increasing number of cargo rates being available, Cargoguide (2018)
has developed a freight rate intelligence software, which is comparable to
the Infare solution for passenger fares.

Real-time price intelligence is even more important to air cargo NP than to

passenger airlines, since their planning cycles are much shorter. Cargo
airlines can be required to start new routes with only a few weeks prepa-
ration; thus, sensing the rate development of potential cargo routes can be
a significant leap forward. Network planners for cargo airlines can use real-
time price information to improve the reliability of demand forecasts and
identify trends rapidly. Furthermore, cargo rate intelligence can improve
Development and evaluation of BDOs for airline NP 221

competitor monitoring and explain the internal booking situation. This has
an impact on almost all NP process steps from demand forecasting until
the final aircraft routing, which can be changed if short-term demand in-
creases in the spot market are detected.

A9 – Analysis of exhibitor data of industry fairs

Fairs for industrial goods are the main driver of the micro-seasonality in
cargo airlines. Especially those fairs that involve high value goods such as
cars or specialized machinery, usually rely on air cargo services. Identify-
ing cargo flows associated with industry fairs requires two components: an
exhaustive list of industry fair events must be complemented with the anal-
ysis of the manufacturing facilities of the exhibitors to predict the micro-
seasonality at the O&D-level. While the cargo airlines in our case study
group maintain an exhaustive list of the most important industry fairs, they
all lack transparency in terms of the entire industry fair landscape. In addi-
tion, trade flows are only recorded by experience, not predictively. Cargo
airlines have currently no transparency in manufacturing sites from exhib-
itors that can help them predict the exact demand origins.

A machine learning-supported data mining tool can increase transparency

in industry-related trade flows. Data on industry fairs, exhibitors, and their
manufacturing sites, as well as the products per manufacturing site can be
derived from publicly available web-sources, using professional scraping
services, such as Data Crops (DataCrops, 2018b). Such data can be linked
to historical internal booking information to identify industry fair-related
shipments. Furthermore, the data can be linked to published cargo vol-
umes and rates from the WorldACD or IATA Cargo Accounts Settlement
System (CASS) databases. With sufficient data points, industry fair-spe-
cific trade lanes can be identified and incorporated in the NP process.

Air cargo network planners can use such data to plan for industry fair-spe-
cific extra stops that can be actively marketed to potential shippers. By
offering a more direct flight connection, the trucking cost can be reduced
and a value proposition for exhibitors can be created. Let us consider the
222 Big data opportunities for airline network planning

Tecnoshow Comigo as an example. It is the largest agricultural technology

show in Brazil, featuring more than 500 exhibitors and over 100,000 annual
visitors (Tecnoshow Comigo, 2018). It occurs in Rio Verde, a medium-
sized town, southwest of Brasilia. As of 2018, only domestic cargo airlines
serve the closest airports of Brasilia and Goiânia, since most international
cargo arrives in the airports of the Sao Paulo agglomeration. Offering an
industry fair-specific stopover in either Brasilia or Goiânia can attract cargo,
which would otherwise have been flown to Sao Paulo and then be trucked
to Rio Verde for one day. Big data analytics on the exhibitors and their
manufacturing facilities can indicate the most likely origins of the cargo.
Network planners can leverage this information to create a better product
for potential shippers and incorporate the micro-seasonality with one or two
additional stops on regular cargo flights to Brazil.

Predicting industry fair-related cargo streams contributes towards both

transportation demand forecast and the identification of trends, including
micro-seasonality. In addition, it may reveal the real movement of goods
by detecting the true origin and destination. This information can be used
both for long- and short-term demand forecasts. Network planners may
choose to add a new route temporarily or add a frequency on an existing
route to serve the demand. They can also schedule additional stops on an
existing route and adjust the aircraft type, based on the expected cargo
volume and characteristics.

A10 – Shipping data from e-commerce platforms

Cross-border e-commerce is one of the major growth factors of air cargo

demand (IATA, 2017b). Large e-commerce platforms, such as Amazon,
Alibaba, eBay, and JD.com, facilitate most e-commerce transactions
(Minges et al., 2016). They own the transaction data, including the real
O&D of the shipment, size of the goods, and chosen delivery mode. While
Amazon is increasingly integrating logistic services, including air cargo in
its own product portfolio (Chamlou, 2018), Chinese platforms Alibaba and
JD.com have not yet entered such logistics fulfillment yet. Furthermore, the
Development and evaluation of BDOs for airline NP 223

growing cross-border e-commerce market opens new route opportunities

for air cargo network planners. Belgian ASL Airlines, for example, started
a direct service from the West-Chinese city of Urumqi to Europe, suppos-
edly driven by such e-commerce demand (Kauffman, 2018).

E-commerce platforms without integrated logistics service have a genuine

interest in offering their customers the optimal shipping services. They may
be willing to share their data with carriers for this purpose. If airlines can
anticipate e-commerce cargo flows, they may be able to react quickly and
adjust their own network dynamically to market needs. If Alibaba, JD.com,
and similar e-commerce platforms share shipping data on a daily base,
airline network planners can refine their demand forecasts and inform new
route decisions, frequency changes, schedule adjustments, or changes in
aircraft types. Thus, cross-border e-commerce shipping data influences
the entire NP process up to actual aircraft routing.

Besides improving the demand forecasts, e-commerce shipping data can

be used to identify trends and extract real movements of e-commerce ship-
ments. Detailed shipping data can help classify airline internal bookings
and monitor competitor actions pertaining to e-commerce demand.

5.2.3 Trend identification

B1 – Analysis of reviews in the travel rating communities

The popularity of travel destinations can change over time, as illustrated

by Cuba. Many Cuban hotels and tourist facilities increased their prices,
anticipating the relaxation of visa regulations in the United States, which
would surge tourist numbers. However, they failed to invest in better ser-
vices, so the perceived value for money in their tourism industry dropped
sharply (Whitefield, 2017). European tourists opted for other destinations,
and the overall air travel demand to Cuba dropped sharply in 2016. Euro-
pean charter airlines did not anticipate this trend and were faced with mas-
sive overcapacity during winter, which had to be adjusted on a short notice.
224 Big data opportunities for airline network planning

Data from large travel communities, such as TripAdvisor or Foursquare,

can indicate such travel trends. Monitoring the average reviews for a des-
tination and its tourist facilities (hotels, restaurants, etc.) can indicate
whether the destination’s image is improving or deteriorating (Kladou &
Mavragani, 2015). Similarly, the number of reviews over a specific period
reflects whether the popularity of the destination changes. In theory, this
data can be obtained from a range of travel communities, but data com-
mercialization has started only recently. TripAdvisor publishes curated
travel trend analyses, without disclosing raw data (Tripadvisor, 2018b).
However, TripAdvisor promotes its “Business Advantage” product as cus-
tomizable analytics solution (Tripadvisor, 2018a). The rating communities
Foursquare (2018) and Yelp (2018) already have both commercialized re-
view data. Since the data is publicly available, it can also be retrieved using
a data scraping service. In this case, the data processing and cleaning
must be performed by the airline.

Airline network planners can use trends in travel reviews to inform the mar-
ket growth expectations of their demand forecasts. Destinations with an
increasingly positive review trajectory can be identified as potential new
routes. Similarly, destinations with decreasing volume and quality of re-
views may be candidates for frequent reviews. Thus, travel review data is
most important for the strategic NP steps, as it indicates long-term trends
rather than the transient short-term demand.

B2 – Clickstream data from travel platforms

Clickstream data from travel platforms, such as TripAdvisor, comprises in-

formation regarding travelers’ choice preferences and individuals’ travel
patterns. While reviews reflect the image of a location, clickstream data
displays its popularity. It includes absolute page view numbers und unique
page view numbers (Tripadvisor, 2018a), which proxy the general interest
in a destination. While reviews contain historical information161, page views

161
The reviewer has already been at the reviewed point of interest.
Development and evaluation of BDOs for airline NP 225

may indicate travel intention. It thus has a more forward-looking character

and may be useful in predicting the next season’s travel flows.

NP managers can use clickstream data for three purposes. First, they can
forecast future travel flows as per travel intentions. Second, they can rank
potential destinations by their popularity to identify route candidates. Third,
they can monitor the trends in popularity using time-series clickstream
data. While the first use case supports a mid-term demand forecast, the
rest aim to primarily identify long-term trends. Therefore, demand forecast-
ing is the most impacted NP process step. Network planners can further
use clickstream data for route selection. Mid-term travel flow forecasts may
enable schedule adjustments and fleet assignment decisions.

B3 – Analysis of local hotel prices

Transportation represents only one side of the coin in a travel system. The
other is accommodation, which largely includes hotels, where leisure and
business travelers are lodged (Bieger & Wittmer, 2006, p. 40). Hotels are
bound to their location and usually know the local demand specificities
much better than any transportation carrier. They may reflect events in their
pricing, of which airline P&RM departments are completely unaware
(Herrmann & Herrmann, 2014). It may be a good strategy for airline net-
work planners and P&RM departments to leverage the local pricing
knowledge for their own decision making.

Hotel prices present two important pieces of information for network plan-
ners. They indicate micro-seasons driven by specific events, which are ex-
pressed through temporary price spikes. It is likely that the demand for
transportation also increases during these price spikes, which may justify
an additional frequency or the deployment of a larger aircraft. If hotel prices
are structurally lower than comparable hotels in other destinations, airlines
should check their own booking ramp up. Low hotel prices may be a symp-
tom of a transportation supply shortage, leading to a concrete business
opportunity – adding transportation capacity to a destination. If a route is
226 Big data opportunities for airline network planning

extraordinarily well booked while hotels remain cheap, network planners

may add flights to benefit from the misalignments.

Many rate management software providers offer a hotel rate intelligence

functionality. The most advanced rate intelligence product in the market is
RateGain, which features a proprietary service quality index to compare
hotels across destinations according to service standard (RateGain, 2018).
Amadeus recently acquired the market leader, TravelClick, for more than
Euro 1 billion (O'Neill, 2018). Octorate (2018) features a similar functional-
ity and includes Airbnb and vacation rental apartments. All these hotel rate
intelligence solutions are highly powerful analytics tools, with much more
functionality than pure rate comparison. It may be sufficient for an airline
to scrape hotel rates from one online travel agency, such as Booking.com,
to understand local micro-seasonality. Specialized data scaping services,
such as DataCrops (2018a), can provide a daily data batch with the desired
prices.

Airline network planners can use hotel prices to fine-tune their demand
forecasts at the O&D level and make the schedule more dynamic for local
micro-seasons. This implies short-term route addition, flight frequency ad-
justment, and fleet assignment changes.

B4 – Analysis of events in social networks

A different approach to detect micro-seasonality is the analysis of events

in social networks. It starts with the detection of events accommodating a
critical audience size among the thousands of public events in social net-
works, such Facebook or MeetUp. These events then need to be assigned
to (potential) destination airport catchment areas. Ideally, the event data
includes demographic audience information, e.g., the stated location of
participants. This would enable network planners to anticipate the actual
travel demand and allow for the filtering of events with a large number of
non-local attendants.
Development and evaluation of BDOs for airline NP 227

Scientific research has mostly focused on event analysis based on un-

structured data, such as pictures and videos (Liu, Wang, & Huet, 2016). In
this approach, a pre-existing event list is used to tag social media content,
based on image recognition or a simple metadata analysis (Stieglitz et al.,
2018). However, analytics related to events without a pre-defined event list
have not widely been researched so far (Liu et al., 2016, p. 445). Until mid-
2018, Facebook allowed access to public event data via the Export API
(Facebook, 2018b). However, the data scandal of 2017 drove the depreci-
ation of the API (Burgess, 2018). The only option remaining is to scrape
event data randomly via the “Discover Events” function, which allows for
filtering by time, location, and category. MeetUp provides free access to
group metadata via API (MeetUp, 2018).

Airline network planners can use social network event data for similar pur-
poses as hotel price patterns. The major applications entail the fine-tuning
of demand forecasts and the dynamization of the schedule, which affects
NP process steps from frequency selection to fleet assignment.

5.2.4 Competitor monitoring

C1 – Real-time flight tracking with transponder data

Airlines experience multiple blind spots in competitor monitoring. Tradi-

tional MIDT data do not convey information on direct airline website book-
ings, which account for the increasing amount of total bookings
(Sheivachman, 2017). Published flight schedules do not include charter
flights, e.g., for sport events or cruises. If airlines lose tenders for charter
flights, they often do not know which airline gained the opportunity. Finally,
airlines have limited visibility in terms of the operational issues of compet-
itors, such as grounded aircrafts or heavy delays.

The emergence of the Automatic Dependent Surveillance Broadcast

(ADS-B) technology enabled the real-time tracking of ADS-B-equipped air-
craft. Most commercial aircrafts now have an ADS-B transponder installed,
which permanently transmits data on position (longitude and latitude),
228 Big data opportunities for airline network planning

speed, altitude, hex-code for aircraft identification, and the aircraft registra-
tion or flight number (Flightradar24, 2018). Companies such as
Flightradar24 and FlightAware maintain their own receiver network to cap-
ture the ADS-B data in relevant regions. The data is already commercial-
ized and can be purchased through a real-time data feed subscription or
as batch data.

Research on ADS-B data has focused almost exclusively on air traffic con-
trol applications (Strohmeier et al., 2014) and air traffic flow research
(Schäfer et al., 2014). Older research has focused on the security aspect
of the transmission protocol and general technology adoption issues
(McCallie, Butts, & Mills, 2011).

Batch data can be used to fill the gaps of MIDT and flight scheduling data.
Since almost all the commercial airplanes are equipped with an ADS-B
transponder, the dataset includes a seamless recording of flight move-
ments in the covered geographies. By being aware of the outline data of
charter tenders, airlines can retrace the winning competitors. Real-time
data also enables the detection of the operational issues of competitors,
which can be used to selectively increase frequencies or change aircraft
types to the airline’s benefit. The recorded flight paths can support cost
calculation for new routes, since the likelihood of specific air routings can
be estimated. In summary, transponder data primarily serves competitor
monitoring and has secondary applications in cost forecasting and as an
early warning system.

While airline network planners conduct cost forecasting and competitor

monitoring throughout the entire NP process, the real-time monitoring com-
ponent is most vital for short-term tactical NP. Airlines may re-fleet or re-
schedule if a competitor has severe operational issues, but this is not rel-
evant to the strategic planning and scheduling process steps.
Development and evaluation of BDOs for airline NP 229

C2 – Network-relevant competitor media monitoring

The bankruptcy of Air Berlin in 2017 represented the impact of competitor

events on airline NP. Competitors took over or opened more than 100 new
routes from Berlin and Düsseldorf in the summer of 2018, which posed a
huge NP challenge for all airlines (Recklies, 2018). As a consequence, on-
time performance suffered dramatically, as airlines operated with non-ro-
bust schedules to secure the slots that they inherited from Air Berlin
(Reichert, 2018). Early warnings concerning such significant competitor in-
cidents, such as bankruptcies, strikes, or forced groundings thus become
increasingly important. Airlines in the case study group mostly rely on tra-
ditional news briefings and scenario planning to monitor competitor events.

Research on new media monitoring technologies has mostly focused on

social media (Batrinca & Treleaven, 2015). However, the advent of ma-
chine learning-based NLP technologies has also transformed established
media monitoring companies. Machine-learning algorithms allow for a
more holistic news coverage, including foreign language media and ad-
vanced sentiment analysis to predict the likelihood of certain events (Ger-
suni, 2016). The American media analytics company Meltwater (2018) of-
fers an “Executive Analytics” product that supports strategic decision mak-
ing, based on media monitoring that is aided by artificial intelligence. The
large incumbent Universal Information (2018) offers cross-platform analyt-
ics that is capable of processing all types of unstructured data, including
videos and podcasts.

Airline network planners can use these specialized media monitoring tech-
niques to filter those competitor events that may have an impact on NP.
Larger events such as bankruptcies mostly have a lasting impact and in-
fluence strategic NP. Temporary events such as strikes or groundings may
call for tactical actions, such as the scheduling of additional flights or a
change in fleet assignment. In any case, advanced media monitoring ben-
efits the airlines in terms of competitor monitoring and serves as an early
warning system for critical events.
230 Big data opportunities for airline network planning

5.2.5 Real movements of passengers and goods

D1 – Mobile location data from smartphones

Airline network planners have only extremely limited data on the real origin
and the real destination of their passengers. Recorded IP numbers of direct
bookings are often misleading, and location tracking with an airline’s mo-
bile app covers only a small portion of the passengers. The widespread
prevalence of mobile phones has created a completely new opportunity for
location tracking. Every time a cell phone receives a call or message or
data package, mobile network operators can link the phone to the trans-
mitter catchment area. In the early years of mobile phones, operators could
locate users only when they placed or received a call or text message.
However, the increasing popularity of data flat rates have caused constant
flows of data package exchanges and allows mobile operators to track
smartphones seamlessly in their networks (Bonnel et al., 2015, pp. 381–
385).

Researchers started using mobile location data for urban travel flow anal-
ysis around ten years ago (Calabrese et al., 2011). Subsequently, more
studies on urban travel flows based on mobile location data have been
conducted (Çolak et al., 2015; Iqbal et al., 2014). Tolouei, Psarras, and
Prince (2017) demonstrated that properly processed mobile location data
is as reliable as traditional travel surveys, but it has the advantage of a
much larger sample size. Mobile network operators cover a significant part
of the population, resulting in highly representative samples.

Telecommunication providers are increasingly commercializing their mo-

bile location data. In Europe, Deutsche Telekom founded the subsidiary
Motionlogic only to process and market location data (Motionlogic, 2018).
Similarly, Telefónica created Telefónica NEXT, which cooperates with the
Swiss location analytics company Teralytics (Telefónica Next, 2018). All
three companies agreed to share certain information for this study and
were included in the data provider group. Current data analytics can record
movements down to the radio cell level (in Germany) and identify the mode
Development and evaluation of BDOs for airline NP 231

of transportation. Data protection laws require companies to anonymize

users and display only aggregate movement flows, including at least five
users. The current datasets are limited to national mobile networks. If a
user crosses a country’s border and connects to a new network, he is as-
signed a completely new anonymized random code, making cross-border
tracking difficult. However, all providers are currently working to integrate
their European mobile networks into one coverage zone.

The American start-up Streetlight Data, which also joined the data provider
group, follows a different approach. As against using mobile phone signals,
Streetlight Data purchases the GPS-tracked location data from over 400
application providers and now covers over 30% of the mobile phone users
in the US (StreetLight Data, 2018). GPS-based location data has the ad-
vantage of being recorded independent from the mobile network and thus
can be tracked across country borders.

Mobile location data provide a broad application range for airline network
planners. They draw an accurate picture of the relevant travel streams
across all transportation modes, which can be filtered by origin, location,
time of travel, and mode of travel. No other data type currently provides a
comparable information depth across transportation modes. Airline net-
work planners can employ mobile location data to precisely define the
catchment areas of airports. Moreover, in case of low flight frequency, they
can track the real origins and destinations of their passengers162. If GPS
data is used, passengers can be matched to flights, even at busier airports.
The data allows also for competition monitoring, since location data is not
limited to their own passengers. The time series of mobile location data
reveal changing travel patterns, including the used transportation mix. The
data can be combined with an event list (see B4 – Analysis of events on
social networks) to identify the origin of event visitors. Micro-seasonality

162
Data is currently available as 15-minute time stamps; if a limited number of flights arrive
or depart during a 15 minutes time window, passengers may be allocated to the individual
flights.
232 Big data opportunities for airline network planning

can also be identified by analyzing the absolute numbers of travelers per

day.

This information can be used across the strategic NP process. Demand

forecasts can be enriched with real O&D granularity, and the alternative
transportation modes can be included. Transparency pertaining to the
catchment areas of airports can influence hub decisions and route selec-
tion. Information regarding time-of-day travel patterns can inform schedul-
ing and flight frequency decisions. The application for tactical planning is
rather limited, since the data reflects actual travel and not intended travel.

D2 – Extended electronic air waybill with shipper address

Air cargo network planners lack transparency in terms of the real O&D of
shipments when the shipments are consolidated by a freight forwarder.
The current paper-based air waybills (AWBs) require the logistics facilities
only before and after the air transport, so there is little information on the
real origin and destination addresses. The digitization of air waybills is now
a major initiative of the IATA, who defined the report and information stand-
ards for electronic AWBs (IATA, 2018a). However, the e-AWB resembles
the data entry fields of the paper-based AWB, thus not solving the infor-
mation transparency issue for air cargo network planners.

The digitization of logistical supply chains was mentioned by researchers

as early as 2001, when Forster and Regan found that the e-AWB would
aid significant operational performance improvement. However, the role of
freight forwarders as intermediaries impeded their support in electronic
data exchange with other supply chain members for another decade. Only
in recent years has the electronic waybill (which also covers terrestrial and
sea cargo) attracted the attention of scholars. Bakhtyar, Holmgren, and
Persson (2013) have found positive synergies between electronic waybills
and intelligent transportation systems. Bakhtyar and Henesey (2014) em-
ployed standard e-waybill data fields to train a machine learning algorithm
Development and evaluation of BDOs for airline NP 233

to predict cargo volumes and products for land transport. However, there
exists no air cargo-specific scientific literature on electronic waybills. The
IATA started the e-AWB initiative in 2010, aiming to digitize all AWBs by
2025 (IATA, 2018a).

In order to improve the usefulness of e-AWB for air cargo NP, the standard
template defined in the IATA Resolution 672 needs to be changed. A de-
tailed composition list of the consolidated cargo must be added as a man-
datory data field. This change has to be backed by an air cargo industry
initiative, since all cargo airlines would benefit from the increased real O&D
transparency. However, it is unlikely that freight forwarders will agree to
this change, since they fear that direct sales from cargo airlines will endan-
ger their business model.

If such changes are applied and all e-AWBs contain real O&Ds, the need
for information related to real O&Ds would be fully satisfied. Real O&D data
for cargo shipments is absolutely crucial for strategic NP decisions. It would
enable a substantially more granular demand forecast based on the real
O&D areas, thereby allowing the optimization of the route portfolio and
flight frequencies. Since a large portion of air cargo is trucked over large
distances, a reconfiguration of the network structure may be useful to cre-
ate shared value for airlines and shippers.

D3 – Location-transmitting sensors for cargo shipments

Another solution for the lack of transparency of real O&Ds for cargo ship-
ments is the attachment of location tracking sensors to cargo shipments.
An increasing number of shippers request real-time tracking of location and
environmental parameters (e.g., temperature, air pressure, gravity peaks),
especially for expensive and highly sensible cargo (Boylan, 2016). Cargo
airlines, such as Cathay Pacific and Lufthansa Cargo, cooperate with sen-
sor producers to offer real-time cargo tracking as a service to their custom-
ers (Air Cargo News, 2018; Lufthansa Cargo, 2018b). However, the cur-
rently available tracking solutions are costly, since the devices need to be
234 Big data opportunities for airline network planning

leased for about USD 1.50 per day (Lufthansa Cargo, 2018b). Multiple pro-
viders of cargo tracking devices offer sensors directly to shippers, without
leasing them (Cargo Signal, 2018; Sendum, 2018; Sensitech, 2018).

The increasing importance of Internet of Things (IoT) solutions has

boosted the demand for a low bandwidth communication network. The ex-
isting mobile networks provide increasing transmission bandwidth but are
restricted in the number of devices that can be connected at the same time.
Many IoT sensors do not require a high bandwidth to upload their data, but
the sheer number of sensors requires networks with a much larger device
capacity than current mobile networks. Thus, industry associations and
mobile network operators have founded the LoRa Alliance (2018) to pro-
mote a new Low Range Wide Area Network (LoRaWAN) standard that can
complement the planned 5G networks with a low-bandwidth alternative for
IoT sensors. Poenicke et al. (2018) have already described the applicability
of the LoRaWAN technology for air cargo tracking sensors, which are ex-
pected to be significantly cheaper than the current solutions.

The increasing popularity and decreasing cost of real-time tracking sen-

sors for shipments present a great opportunity for air cargo network plan-
ners. Similar to mobile location data for passengers, they can get a com-
plete picture of the real O&Ds of cargo shipments. However, data owner-
ship remains a question. Since shippers would probably not share location
data automatically, airlines can only access data sent to LoRaWAN receiv-
ers in their airplanes. Ideally, the mobile network operators should anony-
mize and aggregate location data as with mobile location data, which can
then be commercialized for cargo airlines.

Cargo location data can have a substantial impact on the entire NP pro-
cess. Air cargo network planners can employ this data to conduct hyper-
granular O&D analyses to retrace the end-to-end shipment streams. This
data can be used to optimize network structure, route portfolio, and flight
frequencies. If the data is available in real-time, prediction models for short-
Development and evaluation of BDOs for airline NP 235

term spot bookings can inform scheduling and fleet assignment as a reac-
tion to foreseeable demand shifts. Real-time location data can also inform
aircraft routing and short-term dispatch, since it increases visibility in terms
of whether booked shipments will arrive on time or not.

5.2.6 Real-time planning constraint monitoring

H1 – Real-time slot trading platform

Airports with high traffic volumes assign landing time windows (i.e., slots)
to airlines to avoid congestion in the available infrastructure. Airport slots
are allocated using the grandfathering principle, implying that the slot
owner keeps the slot if he complies with a specified use quota (e.g., 80%
of days used). If an airline fails to comply with the use quota or stops a
service, the slot ownership is retracted by the airport or the regulator (de-
pending on the country). The slot are then re-allocated based on the pub-
lished allocation criteria (Madas & Zografos, 2006, p. 54).

Scholars have long argued that the grandfathering concept is economically

inefficient and proposed alternative allocation strategies. Brueckner (2009)
argued that congestion can be regulated most efficiently either through
congestion-dependent landing fees or by auctioning slots. Some scholars
go even further and demand a secondary market for slots, which would
enable the transfer of the ownership right from airports to airlines once the
slot has been allocated or auctioned (Madas & Zografos, 2006; Verhoef,
2010). While monetary slot trading is not permitted in most regulatory re-
gimes (except for the US and the United Kingdom), airlines can swap slots
through bilateral agreements (Madas & Zografos, 2006, p. 53).

Airline network planners are faced with two transparency issues with re-
gard to airport slots. First, many countries do not support transparency in
the slot register, i.e., the database containing all the allocated and available
slots. Second, slot changes require in almost all countries (except for the
236 Big data opportunities for airline network planning

US and the United Kingdom163) regulatory approval and thus significant

time investment. Therefore, slot swaps are often agreed during the semi-
annual IATA slot conferences, which drastically reduces scheduling flexi-
bility.

Currently, only two companies have transparency in the slot registers of

mostly free-market economies. The UK-based Airport Cooperation Limited
(ACL) covers slot-restricted airports in eight countries, including UK, Can-
ada, and New Zealand (Airport Coordination Limited, 2018b). The German
company e-Airportslots serves nine markets, including Germany, Spain,
France, Mexico, and Australia (e-Airportslots.aero, 2018). ACL provides a
slot-trading platform for the UK, which has facilitated 55 slot swaps or
trades in 2017 (Airport Coordination Limited, 2018a).

A more flexible trading or exchange mechanism would reduce the required

lead time for the scheduling process dramatically, since the required slots
can be acquired simply. The slot acquisition cost must be integrated in the
business plan for the new route or frequency. Any trading scheme comes
with complete transparency on the slot database, thus solving both trans-
parency issues. The European Commission has proposed the liberaliza-
tion of secondary slot trading in the entire EU (Buyck, 2018), and the US
Federal Aviation Authority proposed a similar deregulation three years ago
(Federal Aviation Authority, 2015). To realize the full potential in this area,
more countries need to liberalize slot allocation mechanisms, including
growing markets such as China and India.

Real-time slot trading platforms can potentially be integrated in standard

scheduling solutions to automatically indicate when a desired slot is avail-
able. Airline network planners will be able to use the information from the
route selection process to test the financial and operational feasibility more

163
UK enjoys a special exception from the current EU regulations; slot trading in the US is
restricted to a limited number of congested airports (Fukui, 2010).
Development and evaluation of BDOs for airline NP 237

rapidly than today. Flight frequency changes and (re-)scheduling will also
become more dynamic.

H2 – Real-time traffic right information platform

Traffic rights, including overflight rights and transport freedom rights164, are
usually negotiated between two countries and codified in a bilateral agree-
ment. The exceptions are multilateral agreements, such as the European
Single Sky Agreement, where one agreement includes multiple countries.
Those agreements are legal documents and usually comprise a public and
a classified part.

Both the International Civil Aviation Organization (ICAO, 2018a) and the
World Trade Organization (WTO, 2013) maintain global traffic right data-
bases, which suffer from similar shortcomings. The databases contain only
the public information regarding the agreements, which sometimes lacks
penalty and contingent information. Furthermore, the databases are up-
dated infrequently (the last WTO update was in 2013), so they may not
reflect the latest legal situation. In addition, the coverage of agreements
between countries with poor reporting standards is low. Consequently,
most airlines from the case study group maintain their own databases on
traffic rights, which are evaluated as reliable and useful in section 4.2.3.
Airlines usually contact the respective government authority to gather in-
formation on relevant existing traffic rights in case they plan to serve a new
route. Moreover, the airlines sometimes experience denied overflights due
to a non-public overflight contingent, which is not recorded in official data-
bases and is not tracked real time.

This process is both time and labor intensive and can potentially be simpli-
fied by a global traffic right database, which is linked to a real-time tracking

164
Transport freedom rights regulate how airlines can transport passengers and goods from
and to a specific country. The ICAO (2018b) defines nine freedom rights, from simple over-
flight rights (First Freedom) to the right for a foreign airlines to serve domestic routes within
the home country (Ninth Freedom).
238 Big data opportunities for airline network planning

tool, such as flight transponder data. Modern text mining algorithms can
simplify the labor requirement to scan and analyze bilateral agreements
and deliver instead just the relevant information. Such information, e.g.,
number of permitted overflights per day, can then be cross-checked
against the published schedules and real-time flight tracking data.

Real-time transparency in traffic rights can save airline network planners

their manpower in the strategic planning steps, as they can easily extract
information. It would also flag risks, such as late flights, which may lead to
denied overflight permits or additional fees. The risks of paying additional
fees and the cost of diversions can then be considered in the cost forecast
of a route. Tactical NP can benefit from real-time information on the use of
traffic rights, e.g., if the overflight quota has already been fulfilled. In this
case, the aircraft routing can be adjusted to avoid additional cost. All of
these are of utmost relevance for charter and cargo airlines that frequently
change routings and schedule additional flights at short notice.

5.2.7 Incident monitoring

I1 – Global meteorological warning platform

Two types of meteorological events may influence airline NP. Recurring

meteorological phenomena, such as the occurrence of a morning fog in a
specific location, can influence scheduling decisions. The forecasting and
impact assessments of meteorological disasters, such as floods, hurri-
canes, and droughts, should indicate the shifts in short-to mid-term de-
mand, which may cause re-routings or temporary route suspension.

Scholars and practitioners are applying big data analytics to both meteor-
ological forecasts and disaster impact assessments. Larsen (2013) pro-
posed to use historical meteorological data to inform flight planning sys-
tems, which is a central IT tool of the flight operations process. Artificial
intelligence is being used to forecast turbulence development (Williams,
2014) and provide decision-making tools in case of significant weather
phenomena (Choi et al., 2016; McGovern et al., 2017). A research project
Development and evaluation of BDOs for airline NP 239

conducted by the University of Michigan has developed an algorithm to

detect weather patterns, which can inform airline scheduling (Moore,
2014). In practice, The Weather Company (2018), a subsidiary of IBM, is
already commercializing turbulence forecasts for aviation.

Disaster forecasting and assessment is a predestined application for BDA.

High volume real-time data from a variety of sources must be examined to
obtain useful information. Akter and Wamba (2017) found that data mining
techniques using social media and geo-spatial information are most com-
monly used for disaster management applications. IBM is researching ar-
tificial intelligence forecasting algorithms with data from Australia and New
Zealand (Mareels, 2018). However, there is no commercialized big data
disaster forecasting solution yet.

A global meteorological warning platform addresses three types of infor-

mation needs. It is primarily an early warning platform to mitigate the im-
pact of meteorological events. It further informs short- and mid-term de-
mand forecasts in case of natural disasters and cost forecasts by calculat-
ing the expected cost due to irregular weather situations. Disaster impact
assessment is most useful for route selection and flight frequency determi-
nation to react to demand changes. Finally, airline network planners can
use weather pattern forecasting for schedule optimization.

I2 – Global political risk assessments

Like natural disasters, unexpected political events, such as strikes, military

coups, and civil wars can significantly influence travel demand. Airline net-
work planners rely on risk reports evaluating risks based mostly on human
judgement and media analysis. These are provided by professional risk
consultancies. Airline network planners of the case study group rated these
risk reports low both in usefulness and reliability.

Big data analytics may provide a means to improve in both dimensions.

The American start-up Geoquant (2018) uses 250 risk indicators to calcu-
240 Big data opportunities for airline network planning

late a real-time political risk index (Shieber, 2017). Other companies de-
velop media monitoring algorithms that scan relevant sources every 30
seconds and flag potentially critical events (Muñoz, 2016). The Chinese
company Grisk combines NLP and machine learning to produce sectorial
risk indices, considering the different vulnerability levels of specific indus-
tries towards political events (Grisk, 2018).

Airline network planners can use political risk assessments to adjust their
demand forecasts and subsequently route portfolio and flight frequencies.
The applicability in tactical NP is rather limited. However, unexpected po-
litical events may be relevant to flight operations if they require short-term
re-scheduling or re-routing of flights.

5.3 Evaluation of BDO potential

This sub-chapter presents the qualitative potential assessment of the 23

BDOs identified. All the BDOs have been introduced to the airline network
planners in the case study group, including data properties, NP applica-
tions, and potential data providers. The airline network planners then
scored the potential of the BDOs for their specific airline NP process on a
scale of 0 to 5. A score of 0 means that the respective BDO has no potential
use in the airline’s NP process, and it will not be considered even if the
data is available for free. Conversely, a score of 5 indicates that the airline
would invest significantly in data acquisition and data analytics to leverage
the specific BDO.

5.3.1 Qualitative assessment of BDO potential

All nine case study airlines participated in the qualitative potential assess-
ment. The evaluation scale ranged from 0 for no potential to 5 for very high
potential. The arithmetic mean across all the BDOs is 2.2, which is slightly
less than the median of 2.5. This indicates that the overall potential of the
BDOs is considered only low to medium. Only four BDOs were scored with
an overall potential of more than 3 (A5, A6, H1 and H2). A9 was assessed
Evaluation of BDO potential 241

by the airline network planners with the lowest potential across all BDOs,
with a score of 0.5. Figure 5.4 visualizes the overall potential scores (black
line) and the score range across all individual assessments (grey line).

No Very high
potential potential
Answer
0 1 2 3 4 5 variance
A1 Geo-spatial image analysis 0.79
A2 Real-time inflation forecasts 1.98
A3 Consumer confidence measurement with social media 1.27
A4 Detailed demographic profiling with social networks 2.84
Transportation
A5 General search engine data for detailed O&D demand 0.55
demand
A6 Click-stream data of meta-search websites for flights 2.84
forecast
A7 Social media mining to determine price elasticity 1.64
A8 Analysis of air cargo rates from online freight portals 3.43
A9 Analysis of exhibitor data of industry fairs 0.86
A10 Shipping data from e-commerce platforms 4.41
B1 Analysis of reviews in travel rating communities 2.41
Trend B2 Click-stream data from travel platforms 2.00
identification B3 Analysis of local hotel prices 2.57
B4 Analysis of events in social networks 2.55
Competitor C1 Real-time flight tracking with transponder data 0.98
monitoring C2 Network-relevant competitor news monitoring 1.13
D1 Mobile location data from smartphones 4.50
Real move-
D2 Extended electronic AWB with shipper address 4.57
ments
D3 Location-transmitting sensors for cargo shipments 4.41
Planning H1 Real-time slot trading platform 1.27
constraints H2 Real-time traffic right information platform 1.07
Incident I1 Global meteorological warning platform 1.13
warnings I2 Global political risk assessments 2.13

Average Answer space

Figure 5.4 – Overall qualitative potential assessment of BDOs165

The variance in the assigned scores for potential across the case study
group is 2.27 on an average, indicating significant differences in BDO po-
tential for airlines. Location transmitting cargo sensors (D3) exhibited the
largest differences. Two CARs estimated the potential to be very high (5.0),
while five out of the six passenger airlines see absolutely no potential in
this BDO. The differences seem to be business model specific and are
examined further in the section 5.3.2.

165
Source: Own illustration.
242 Big data opportunities for airline network planning

Furthermore, there are also BDOs with comparably uniform potential esti-
mates. Some BDOs for economic forecasts (A1, A3) presented similar po-
tential for all business models. Moreover, the potential for a real-time traffic
right information platform and the incident warnings did not differ dramati-
cally between airlines.

Looking at the need for information clusters, BDOs addressing planning

constraints presented the highest overall potential score of 3.6. Trend iden-
tification, competitor monitoring, and incident warnings reached an aver-
age potential score of 2.3, followed by real movements that was scored
2.0. Transportation demand forecast, which is the most prominently ad-
dressed need for information, with 10 primary BDOs, presented the lowest
overall potential score of 1.9.

The variance within the need for information clusters seems to be inversely
related with the average potential score. The highest rated cluster “plan-
ning constraints” had a comparably low answer variance of only 1.17,
which is much lower than the average answer variance of 2.27. On the
contrary, real movements had an extraordinarily high variance of 4.41, and
it was rated below average with a potential score of 2.0. However, a simple
correlation analysis performed on BDO level does not support this assump-
tion. The correlation166 between the variance and average score is -0.23,
which is a negligible value, at most indicating a mild association between
variance and potential score.

5.3.2 Differences in BDO potential by business model

Due to the specificities of airline business models, airlines may evaluate

the potential of BDOs differently according to their information needs. Fig-
ure 5.5 visualizes the assessment results by business model.

166
The Pearson correlation coefficient has been calculated using the score variance as the
independent variable and the average score as the dependent variable.
Evaluation of BDO potential 243

No Very high
potential potential
0 1 2 3 4 5

A1 Geo-spatial image analysis

A2 Real-time inflation forecasts
A3 Consumer confidence measurement with social media
A4 Detailed demographic profiling with social networks
Transportation
A5 General search engine data for detailed O&D demand
demand
A6 Click-stream data of meta-search websites for flights
forecast
A7 Social media mining to determine price elasticity
A8 Analysis of air cargo rates from online freight portals
A9 Analysis of exhibitor data of industry fairs
A10 Shipping data from e-commerce platforms
B1 Analysis of reviews in travel rating communities
Trend B2 Click-stream data from travel platforms
identification B3 Analysis of local hotel prices
B4 Analysis of events in social networks
Competitor C1 Real-time flight tracking with transponder data
monitoring C2 Network-relevant competitor news monitoring
D1 Mobile location data from smartphones
Real move-
D2 Extended electronic AWB with shipper address
ments
D3 Location-transmitting sensors for cargo shipments
Planning H1 Real-time slot trading platform
constraints H2 Real-time traffic right information platform
Incident I1 Global meteorological warning platform
warnings I2 Global political risk assessments

FSC SCA
LCC CAR

Figure 5.5 – BDO qualitative potential evaluation patterns by business model167

Full-service carriers find the highest potential in data from meta-search

websites (5.0) and from general search engines (4.0). Mobile location data
(4.5) was also highly ranked, whereas cargo-specific data sources A8-A10
were mostly not deemed to have any potential in FSC NP. The average
potential assigned by the FSC network planners was 2.3 and thus slightly
higher than the global average of 2.2. FSCs see more potential in the travel
demand forecasting and trend identification BDOs than other airlines (see
Appendix 19). This can be explained by the longer planning horizon of
FSCs, which emphasizes on the need for accurate long- and mid-term de-
mand forecasts and the early identification of trends. In contrast, competi-
tor monitoring and incident warnings were rated with less potential than the
global average. Due to their inflexible schedules, FSC may benefit less

167
Source: Own illustration.
244 Big data opportunities for airline network planning

from real-time information, which would allow them to react to competitor

problems or other events at a short notice.

In most cases, low-cost carriers evaluate the potential of BDOs lower than
FSCs, which is confirmed by the below-average potential score of 2.0
across BDOs. Both planning constraint BDOs bear high potential for LCCs,
since they would benefit from real-time data due to their short planning
cycles. The identification of micro-seasonality with local hotel prices and
events is also considered to have high potential by LCC network planners.
Demographic profiling can further benefit the identification of underserved
ethnic markets, which is a considerable traffic driver for LCCs. Looking at
need for information clusters, LCCs rated BDOs above average in the plan-
ning constraint cluster and in the trend identification cluster. Whereas, the
potential of BDOs for transportation demand forecast and real movements
was scored lower than average. This aligns with the findings on the LCC
NP process, which is highly dynamic due to short lead times. LCCs want
to identify unserved revenue pockets, which is reflected in the serving of
ethnic routes and schedules reflecting micro-seasonality.

The ranking profile of scheduled charter airlines combines aspects from

the FSCs and LCCs. Search engine and mobile location data were evalu-
ated with high potential in addition to demographic profiling and mobile lo-
cation data. SCAs have longer planning cycles than LCCs, which in-
creases the need for trend identification and transportation demand fore-
casting information. Moreover, SCAs also target ethnic travel and try to
identify the underserved point-to-point connections. This makes demo-
graphic data and mobile location data valuable. SCAs can also leverage
the current hotel prices to identify supply shortages in tourist destinations
and benefit from adding capacity. Overall, SCAs evaluate BDOs similar to
how LCCs do so, with an average score of 2.0, which is lower than the
global mean. Trend identification is the only need for information cluster
rated higher by SCAs than by the global average of all airlines.
Evaluation of BDO potential 245

Cargo airlines show a completely different evaluation pattern than the other
business models. They see high potential in a real-time slot trading plat-
form, which is in line with LCCs and caused by the short planning cycles
of the cargo NP process. However, besides the real-time slot trading plat-
form, cargo airlines highly rated those BDOs that have an explicit role in
cargo demand forecasting and tracking. Overall, CARs rated the 23 BDOs
with an average potential of 2.3, which is similar to the FSCs and higher
than the LCCs and SCAs. They rated all the need for information clusters
except for trend identification higher than the average of the case study
group. Table 5.2 shows the BDOs ranked by business model.
Table 5.2 – BDOs ranked by business model168

Code Big data opportunity Overall rank Rank by business model

FSC CAR LCC SCA

H1 Real-time slot trading platform 1 4 1 1 9

Clickstream data of meta-search 2 1 12 11 1
A6
websites for flights
General search engine data for 3 3 9 6 4
A5
detailed O&D demand
Real-time traffic right information 4 7 10 2 6
H2
platform
Mobile location data from 5 2 20 7 5
D1
smartphones
Detailed demographic profiling 6 5 21 5 3
A4
with social networks
B3 Analysis of local hotel prices 7 10 19 4 2
Network-relevant competitor 8 16 7 12 11
C2
news monitoring
Analysis of events in social net- 9 6 17 3 13
B4
works
Global meteorological warning 10 13 6 10 14
I1
platform
Analysis of reviews in travel rat- 11 11 22 8 7
B1
ing communities

168
Source: Own illustration.
246 Big data opportunities for airline network planning

Code Big data opportunity Overall rank Rank by business model

FSC CAR LCC SCA

I2 Global political risk assessments 12 18 13 14 8

Real-time flight tracking with 13 17 8 13 15
C1
transponder data
Click-stream data from travel 14 12 23 9 10
B2
platforms
Consumer confidence measure- 15 14 16 15 12
A3
ment with social media
A1 Geo-spatial image analysis 16 15 11 16 16

A2 Real-time inflation forecasts 17 8 15 17 17

Location-transmitting sensors for 18 19 2 19 19
D3
cargo shipments
Extended electronic AWB with 19 20 3 20 20
D2
shipper address
Social media mining to deter- 20 9 18 18 18
A7
mine price elasticity
Shipping data from e-commerce 21 21 4 21 21
A10
platforms
Analysis of air cargo rates from 22 22 5 22 22
A8
online freight portals
Analysis of exhibitor data of in- 23 23 14 23 23
A9
dustry fairs

Real-time slot trading platforms were the highest ranked BDO for both
LCCs and CARs, and the fourth highest for FSCs. It was also the only
shared top 5 BDO of CARs with any other business model. Within the pas-
senger airlines, demographic profiling appears in the top 5 of all three busi-
ness models. Four BDOs appeared twice in the business model-specific
top 5, namely clickstream data of meta-search websites, mobile location
data from smartphones, general search engine data (all FSCs and SCAs),
and analysis of local hotel prices (LCCs and SCAs). In total, 12 different
BDOs appeared at least once in the top 5 of one business model, which
are highlighted in grey in Table 5.2.
Evaluation of BDO potential 247

Comparing the business model-specific evaluation patterns, significant

contrasts are detected. The sharpest contrast is between cargo airlines
and passenger airlines. Cargo airlines assessed the potential of seven
BDOs (A4, A7, B1, B2, B3, B4, and D1) with a score below 1, thereby
denying any potential of these BDOs in their airline NP. In six out of seven
cases (except for A7), there was not a single passenger airline that rated
the potential of the BDO less than 2. Thus, these BDOs seem to create
value for passenger airlines but not for cargo airlines. Conversely, five
BDOs (A8, A9, A10, D2, and D3) seem to create significant potential for
cargo airlines but little to no potential for passenger airlines. This indicates
that there are three application classes of BDOs.

Data opportunities can be applicable to either cargo airlines or passenger

airlines, or they can generate potential for all airline types. Figure 5.6 illus-
trates the allocation of BDOs. Only transportation demand forecast and
real location clusters include both passenger and cargo-specific BDOs.
The trend identification cluster is only relevant to passenger airlines, while
competitor monitoring, planning constraints, and incident warnings exclu-
sively make for general BDOs.

If the low assessments of potential for cargo and passenger-specific BDOs

are excluded from the analysis169, the global arithmetic mean increases
from 2.2 to 3.0. From the 23 BDOs, 12 reached a recalculated mean of 3.0
or higher, indicating medium to high potential for the respective business
models. These 12 BDOs coincide fully with the 12 BDOs identified in the
analysis by business model earlier. Appendix 20 presents the BDOs sorted
according to the recalculated mean, and it indicates the individual business
model-specific evaluation scores.

169
The assessments of passenger airlines for cargo-specific BDOs and the assessments of
cargo airlines for passenger-specific BDOs have been excluded.
248 Big data opportunities for airline network planning

Cargo airline specific Passenger airline specific

General BDOs
BDOs BDOs

A8 Analysis of air cargo A4 Detailed demographic A1 Geo-spatial image analysis

rates from online freight profiling with social
portals media mining A2 Real-time inflation forecast
A9 Analysis of exhibitor data
Transpor- of industry fairs A3 Consumer confidence
tation measurement with social media
A10 Shipping data from e-
demand commerce platforms A5 General search engine data for
forecast detailed O&D demand
A6 Click-stream data of meta-search
websites for flights
A7 Social media mining to determine
price elasticity
B1 Analysis of reviews in
travel rating communities
B2 Click-stream data from
Trend travel platforms
identification B3 Analysis of local hotel
prices
B4 Analysis of events in
social networks

C1 Real-time flight tracking with

Competitor transponder data
monitoring C2 Network-relevant competitor news
monitoring
D2 Extended electronic AWB D1 Mobile location data from
with shipper address smartphones
Real move-
ments D3 Location-transmitting
sensors for cargo
shipments
H1 Real-time slot trading platform
Planning
constraints H2 Real-time traffic right information
platform

Incident I1 Global meteo warning platform

warnings I2 Global political risk assessments

Figure 5.6 – Cargo vs. passenger-specific BDOs170

5.3.3 Implementation of BDOs in current IT systems for NP

This list of 23 BDOs was presented to a major system developer for IT NP

solutions in order to evaluate the ease of implementation. The evaluation
was not based on the feasibility in gathering and analyzing the data from

170
Source: Own illustration.
Evaluation of BDO potential 249

the BDO (which is discussed in the next section) but the possibility of the
current NP software solution to integrate the analytics result. The scores
were given on a 6-point scale from 0 (not possible to implement) to 5 (very
easy to implement). Since the current system is focused on passenger air-
line NP, all cargo-specific BDOs were evaluated with the lowest score of
0.

No potential / Very high potential /

Not possible to implement Very easy to implement
0 1 2 3 4 5
A1 Geo-spatial image analysis
A2 Real-time inflation forecasts
A3 Consumer confidence measurement with social media
A4 Detailed demographic profiling with social networks
Transportation
A5 General search engine data for detailed O&D demand
demand
A6 Click-stream data of meta-search websites for flights
forecast
A7 Social media mining to determine price elasticity
A8 Analysis of air cargo rates from online freight portals
A9 Analysis of exhibitor data of industry fairs
A10 Shipping data from e-commerce platforms
B1 Analysis of reviews in travel rating communities
Trend B2 Click-stream data from travel platforms
identification B3 Analysis of local hotel prices
B4 Analysis of events in social networks
Competitor C1 Real-time flight tracking with transponder data
monitoring C2 Network-relevant competitor news monitoring
D1 Mobile location data from smartphones
Real move-
D2 Extended electronic AWB with shipper address
ments
D3 Location-transmitting sensors for cargo shipments
Planning H1 Real-time slot trading platform
constraints H2 Real-time traffic right information platform
Incident I1 Global meteorological warning platform
warnings I2 Global political risk assessments

Airlines potential Ease of implementation

assessment assessment by system
developer

Figure 5.7 – Ease of implementation assessment for current NP IT system171

The answers displayed in Figure 5.7 show that the analytical results from
BDOs are in most cases either very easy or very hard to implement, with
a few exceptions that show a medium degree of ease of implementation.
All BDOs that influence assumption values in the NP software solutions

171
Source: Own illustration.
250 Big data opportunities for airline network planning

are easy to integrate. This group includes economic growth and inflation
forecasts as well as assumptions concerning O&D demand and price elas-
ticity. A second group of BDOs that are comparably simple to integrate are
those providing structured demand data, which can be calibrated to inform
O&D growth forecasts. This includes search data (both Google and flight
websites), clickstream data from travel platforms, and event data from so-
cial media.

The third group of easy-to-implement BDOs includes slots and traffic right
information. The current IT systems already feature a slot and traffic right
database, currently updated manually by airline network planners. The
data pull can be automated if the data is available in a suitable format, i.e.,
a table of current slots and codified traffic rights.

BDOs that provide data on information categories which are not included
in the current IT systems are difficult to implement, because the entire sys-
tem would need to be recalibrated for the new data. Mobile location data,
for example, includes information on non-airline movements, which would
add a completely new dimension to the current IT systems. Similarly, real-
time flight tracking and micro-seasonality forecasting are not included in
present IT systems.

Existing NP software are not able to process unstructured data. Thus,

BDOs that are likely to produce unstructured output data, such as all inci-
dent warning platforms, cannot be integrated into the current IT systems.

5.3.4 Feasibility assessment of BDOs

This section evaluates the implementation feasibility of the 23 BDOs to

complement the potential assessment of the last sections. In the first step,
a feasibility evaluation framework must be constructed to meaningfully as-
sess the BDOs. In the literature, there is no data-specific feasibility assess-
ment framework available. Most existing frameworks have focused on IT
project feasibility evaluations, resulting in a large body of literature being
published in the early 2000s. However, the focus of most research lies on
Evaluation of BDO potential 251

the financial feasibility evaluation (Milis & Mercken, 2004) instead of tech-
nological and legal feasibility aspects. Other literature follows a strategic
perspective and classifies the feasibility assessment as an important pro-
cess step in IT project planning (Willcocks, 2009, pp. 239–259). However,
it does so without further specifying the feasibility criteria apart from finan-
cial feasibility. Similarly, Ehie and Madsen (2005) found that feasibility is
the third most important step in an implementation process for enterprise
resource planning through an empirical analysis.

Although there is no data-specific assessment framework, scholars have

proposed various frameworks to assess the readiness of big data analyt-
ics. Chen et al. (2014, pp. 175–176) listed the challenges for big data im-
plementation, including data availability, data representation, legal re-
strictions, and the availability of analytical methods. Comuzzi and Patel
(2016) developed a maturity model for BDA, constituting a specific data
dimension focusing on data acquisition, management, and analytics.
Based on the scarce literature, a five-dimensional model is developed to
cover data availability, legal restrictions, and analytics. The evaluation
model comprises four feasibility levels, ranging from 2 to minus indefinite.
A score of 2 always indicates the highest feasibility of a dimension,
whereas a minus indefinite specifies that either data collection, processing,
or analytics are impossible. If a BDO receives a score of minus indefinite
on any dimension, it is excluded from further analysis immediately. The
ease-of-implementation assessment provided by one system provider in
the previous section was not considered, because it represents the iso-
lated view on one NP system. Figure 5.8 summarizes the assessment
model and presents the five dimensions and feasibility scores.
252 Big data opportunities for airline network planning

Scoring

2 1 0 -∞
A How technologically feasible is the data Feasibility proven Feasible, but not Potentially Not feasible
collection? yet proven feasible

B What are legal or political restrictions to the No restrictions Minor restrictions Severe Use of data not
use of data? restrictions permitted

C How is data ownership organized? Internal data Data Data potentially Data not available
commercially available
available

D How aggregated (holistic) is the available Single provider of Various providers Various providers Data not available
data? aggregated data of partly of un-aggregated
exists aggregated data data

E How accessible are analytics tools to Analytic tools Analytic tools Analytic tools Analysis not
process the data? integrated with commercially need to be possible
data available developed

Figure 5.8 – Feasibility assessment model for big data acquisition and analytics172

The data collection and acquisition related to a BDO needs to be techno-

logically feasible. Ideally, the specific dataset should already have been
recorded, which proves the technological feasibility (score 2). If data col-
lection is technologically not feasible, the BDO needs to be dropped. This
would hypothetically apply for real-time satellite images, since that would
require an unrealistically high number of geostationary satellites. Further-
more, there can be legal or political restrictions on data collection. Data
privacy laws may restrict the use of raw data or make it completely illegal.
Individual, non-anonymized mobile location tracking is an example of le-
gally restricted data collection. Even if there are no technological and legal
barriers to data collection, the data may simply not be available because
data ownership lies with a company that does not want to share it. Data on
the real O&Ds of cargo shipments exists already, but freight forwarders are
data owners and have no interest in commercializing the data for cargo
airlines. The fourth feasibility criterion is the variety of data sources. Ideally,

172
Source: Own illustration.
Evaluation of BDO potential 253

all data should be available at the same source, which is the case for tran-
sponder data or Google Trends. BDOs with multiple data sources, such as
mobile location data or clickstream data from meta-search engines, require
data aggregation, thus reducing the feasibility of the BDO. Finally, there
should be analytic tools to process the data. In an ideal case, the analytic
tool is integrated in the data set, which is true for the hotel rate intelligence
software. In the worst-case scenario, analytic tools cannot handle the data
meaningfully, e.g., when data quality is poor.

Table 5.3 summarizes the individual feasibility assessments for all 23

BDOs. Examining the individual dimensions, technological and legal feasi-
bility are the least restrictive criteria, whereas data ownership and acces-
sibility of analytic tools hinder the implementation feasibility most.

For 16 out of the 23 BDOs, it could be proven that data collection is tech-
nologically feasible. This proof stems from either a commercialized use or
scientific research using the data source. Three BDOs gave no proof of
technological feasibility. Extraction of exhibitor manufacturing sites (A9)
may be feasible, but there is no real-life example for such a scraped data
set. Similarly, there is no example for a real-time information platform for
traffic rights that connects transponder data with text mining from bilateral
agreements. The extended e-AWB suffers resistance from forwarders who
have blocked a technological proof of concept. However, all three BDOs
appear potentially feasible and are rated with a score of 0.
254 Big data opportunities for airline network planning

Table 5.3 – Feasibility assessment of BDOs173

How accessible are analytic tools to

How technologically feasible is the

How is data ownership organized?

How aggregated (holistic) is the

What are legal or political re-
strictions to the use of data?

process the data?

data collection?

available data?
Total
Code Big data opportunity score
A1 Geo-spatial image analysis 2 2 1 1 1 7
A2 Real-time inflation forecasts 2 2 0 2 1 7
A3 Consumer confidence measure- 2 2 0 0 0 4
ment with social media
A4 Detailed demographic profiling 2 1 1 0 0 4
with social media mining
A5 General search engine data for 2 2 1 2 0 7
detailed O&D demand
A6 Clickstream data of meta-search 2 2 1 0 0 5
websites for flights
A7 Social media mining to determine 1 1 0 0 0 2
price elasticity
A8 Analysis of air cargo rates from 2 1 1 2 2 8
online freight portals
A9 Analysis of exhibitor data of in- 0 2 0 0 0 2
dustry fairs
A10 Shipping data from e-commerce 2 2 0 1 0 5
platforms
B1 Analysis of reviews in travel rating 2 2 0 1 0 5
communities
B2 Clickstream data from travel plat- 2 2 0 2 0 6
forms
B3 Analysis of local hotel prices 2 2 1 2 2 9

173
Source: Own illustration.
Evaluation of BDO potential 255

How accessible are analytic tools to

How technologically feasible is the

How is data ownership organized?

How aggregated (holistic) is the

What are legal or political re-
strictions to the use of data?

process the data?

data collection?

available data?
Total
Code Big data opportunity score
B4 Analysis of events in social net- 1 1 0 1 0 3
works
C1 Real-time flight tracking with tran- 2 2 1 2 0 7
sponder data
C2 Network-relevant competitor news 2 2 1 0 1 6
monitoring
D1 Mobile location data from 2 0 1 1 1 5
smartphones
D2 Extended electronic AWB with 0 1 2 0 0 3
shipper address
D3 Location-transmitting sensors for 2 1 0 0 0 3
cargo shipments
H1 Real-time slot trading platform 2 0 0 0 0 2
H2 Real-time traffic right information 0 1 0 0 0 1
platform
I1 Global meteorological warning 1 2 0 0 0 3
platform
I2 Global political risk assessments 1 2 1 0 1 5
Average score 1.57 1.52 0.52 0.74 0.39 4.74

Two BDOs suffer severe legal restrictions, and seven BDOs have minor
legal or political restrictions. A real-time slot trading platform relies on the
cooperation of national regulatory authorities, and so far, only free-market
economies have provided the necessary access to slot data. Transparency
in slots may not be desirable for countries that wish to protect their national
aviation industries or use slots as a means of political pressure. Mobile
location data suffers from severe legal restrictions, since all the data needs
256 Big data opportunities for airline network planning

to be anonymized and aggregated, making it impossible to track individual

customers. BDOs with minor legal restrictions are mostly constraint from
data privacy legislation.

Only the extended e-AWB can be counted as internal data, since it needs
to be transmitted for every shipment. If the real O&D were to be added as
a mandatory field, the BDA can use the internal AWB database as a pri-
mary data source. Further, 12 BDOs are not yet commercialized, which
requires bilateral negotiations with the current data owners to access these
data. Six BDOs have a single source of data, which reduces the effort of
data matching and processing. Another 12 BDOs work with dispersed and
unstructured data, which, in turn, makes complicated data matching and
processing even more complicated. Intelligence software for cargo rates
and hotel prices include analytic tools to draw useful information from given
data. Five other BDOs offer analytics as an add-on service that needs to
be procured. Integrated platform analytic tools, such as Google Trends or
Facebook Audience Insight, were not rated in this category, since they are
not customizable. The output data from Google Trends or Audience Insight
must be processed and analyzed subsequently for the specific BDO pur-
pose of airline NP.

If we contrast the potential and feasibility of all 23 BDOs, four clusters can
be derived (see Figure 5.9). Top priorities are the BDOs with high potential
(3.0 or above) and high feasibility (score of 5 or above). Six BDOs are
located in this cluster, including the two most feasible BDOs. Solutions for
cargo rate and hotel price analytics are readily available and need to be
only subscribed. Search engine and clickstream data is also already avail-
able, although not all meta-search websites have commercialized their
data yet. Mobile location data is being commercialized, and first pilot appli-
cations for public transportation do exist. Finally, shipping data from e-com-
merce platforms exist, but airlines need to negotiate with platform opera-
tors, such as Alibaba or JD.com, to access the data.
Evaluation of BDO potential 257

Data add-ons Long-term value creators

A1 Geo-spatial image analysis
Value traps Top priorities
Potential A2 Real-time inflation forecasts

5.0 A3 Consumer confidence measurement with social media

D3 A4 Detailed demographic profiling with social networks
A5 General search engine data for detailed O&D demand
4.5 D2 A10
A6 Click-stream data of meta-search websites for flights
A7 Social media mining to determine price elasticity
4.0 A8 A8 Analysis of air cargo rates from online freight portals
H1
A9 Analysis of exhibitor data of industry fairs
D1
A10 Shipping data from e-commerce platforms
3.5 A4
A6 A5 B1 Analysis of reviews in travel rating communities
H2
B3 B2 Click-stream data from travel platforms
3.0 B4 B3 Analysis of local hotel prices
B1
B4 Analysis of events in social networks
B2
2.5 C1 Real-time flight tracking with transponder data
I1 C2 C2 Network-relevant competitor news monitoring
C1
I2 D1 Mobile location data from smartphones
2.0 A9 A1
A3 D2 Extended electronic AWB with shipper address
D3 Location-transmitting sensors for cargo shipments
A2
1.5 H1 Real-time slot trading platform

A7 H2 Real-time traffic right information platform

I1 Global meteorological warning platform
1.0
0 1 2 3 4 5 6 7 8 9 10 I2 Global political risk assessments
Feasibility

Figure 5.9 – Potential and feasibility evaluation matrix for BDOs174

Long-term value creators are BDOs with high potential that currently have
limited implementation feasibility. Raw data on demographic profiling can
be accessed from Facebook, but there is no automated data transmission
and the queries need to be set manually. Furthermore, airlines can ap-
proach LinkedIn to drive the commercialization of business network data.
Event extractions from Facebook are complicated due to the discontinued
API, but third-party developers with full API access can potentially provide
a customized solution for airline network planners. Airlines should actively
lobby the IATA to include a mandatory real O&D field in the e-AWB. Simi-
larly, airlines should approach mobile network providers to commercialize

174
Source: Own illustration.
258 Big data opportunities for airline network planning

cargo location data in the same way as mobile location data, once Lo-
RaWANs are rolled out at a large scale. Finally, airlines can lobby for a
globally harmonized deregulation of slot trading rules to facilitate the crea-
tion of real-time slot trading platforms. Similarly, real-time traffic right plat-
forms will need access to bilateral agreements on a global scale.

BDOs with high feasibility but limited potential can serve as data add-ons,
which can be implemented quickly if a specific information need arises.
Geo-spatial image analysis and real-time inflation forecasts can enhance
economic growth forecasts, and travel trends can be identified by analyz-
ing clickstream data and reviews from travel platforms. Competitor moni-
toring with transponder data or media monitoring may be valuable in case
of increased volatility in the airlines industry. The services of global risk
assessment platforms may become more important if political instability
increases.

Four BDOs are potential value traps, since they are difficult to implement
and have a low potential to satisfy information needs in airline NP. Con-
sumer confidence measurement with social media is possible, but it is only
a very minor input assumption in the economic growth models that are
used by airlines. Furthermore, the data does not dramatically improve the
traditional survey-based consumer confidence indices. Moreover, social
media mining to determine price elasticity is a major technical challenge,
since many input data need to be connected to specific customer profiles.
Data privacy legislation further complicates this analysis. In reality, this
BDO may be more valuable for the P&RM department of an airline, as it
allows for individual pricing (if legally permitted). While data on industry
fairs and exhibitors is easily available, it is problematic to identify the cor-
responding manufacturing sites and predict trade flows. Airline will need to
match the data with their own bookings and invest heavily in data mining
to extract relevant trade flows. Finally, meteorological warning platforms
are the most relevant for flight operations but not for airline NP. Airlines
Evaluation of BDO potential 259

should restrain from investing in value traps, unless their primary purpose
is to serve a different function (e.g., flight operations).

The next chapter develops a cost and benefit evaluation scheme for the
top priorities and long-term value creator clusters. The BDOs in the data
add-on and value trap clusters are not included in the cost–benefit analy-
sis, as those data sources are most likely not the primary investment can-
didates for airline NP. This gives us 12 BDOs for further analysis, including
four cargo-specific BDOs (A8, A10, D2, D3), four passenger-specific BDOs
(A4, B3, B4, D1), and four general BDOs (A5, A6, H1, H2).
6 Financial impact of big data for airline network planning

The selection of the appropriate BDOs to support the NP process is a key

challenge for airline network planners. This chapter aims to develop a log-
ical process that can help airline network planners to choose the most ap-
propriate BDO and provide a framework to estimate the cost and benefits
of a BDO investment. Figure 6.1 displays this logical process and links it
to the relevant sections of this chapter.

Figure 6.1 – Logical process to derive cost and benefit estimate for BDOs175

Since NP departments and processes differ significantly across airlines,

the choice of the right BDO should always be based on specific KPIs of an
airline NP department. Sub-chapter 6 presents a qualitative evaluation of
the 12 selected BDOs based on the KPIs utilized in the NP departments of
the case study group. Subsequently, the best metrics to measure the KPIs
in the BDO context are chosen. Sub-chapter 6.2 develops a benefit calcu-
lation framework based on these metrics. Sub-chapter 6.3 operationalizes

175
Source: Own illustration.

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2020

M. Schosser, Big Data to Improve Strategic Network Planning in Airlines,
Schriftenreihe der HHL Leipzig Graduate School of Management,
https://doi.org/10.1007/978-3-658-27582-2_6
262 Financial impact of big data for airline network planning

the benefit calculation framework through a fictive airline example before

sub-chapter 6.4 finally adds the cost perspective for the 12 selected BDOs.

6.1 Choice of appropriate BDOs based on KPIs

The status quo analysis of the current performance measurement prac-

tices reflected that airlines use a broad set of KPIs to measure their NP
activities. Since the organizational setup and role of NP departments dif-
fers widely, there is no one-size-fits-all solution for a BDO benefit estima-
tion. An expedient framework should consider the individual specificities of
airline NP departments and be based on KPIs that are relevant to the re-
spective airline. However, not all BDOs impact the same KPIs. First, the
BDOs that are most impactful for each individual KPI need to be deter-
mined (section 6.1.1). Based on this analysis, airline network planners can
choose the most promising BDOs for their specific KPI set. Subsequently,
the right metrics to measure the KPI impact of BDOs have to be defined in
order to operationalize a benefit calculation framework. Section 6.1.2 pre-
sents a set of possible metrics for each KPI and then chooses the most
suitable measure to calculate financial benefits resulting from BDOs.

6.1.1 Impact of BDOs on airline NP KPIs

The case study group primarily uses indirect KPIs as well as aggregated
route and network profitability to measure NP performance. The direct KPI
load factor, direct operating cost (DOC), and indirect operating cost (IOC)
are also used as secondary KPIs by some airlines. IOC and DOC are com-
bined in a direct cost KPI, since all airlines use them in this form only. Sim-
ilarly, network and route profitability are joined for the analysis. Figure 6.2
recapitulates the currently used KPIs from chapter 4.
Choice of appropriate BDOs based on KPIs 263

Figure 6.2 – Direct and indirect KPIs currently used by the case study group airlines176

In the last interview round, the participating airline network planners were
asked to assess the impact of the 12 BDOs selected in chapter 5. The
rating scale ranged from 0 (no impact) to 5 (very high impact), and each
primary or secondary KPI that the respective airline used was assessed.
The average scores across the case study group were calculated using an
equally weighted arithmetic mean of the answers177. The average scores
were used to produce spider graphs for each KPI, which visually indicate
the estimated impact by BDO (Figure 6.3 to Figure 6.9). Cargo-specific
BDOs were evaluated only by cargo airlines, while passenger-specific
BDOs were evaluated only by the FSCs, LCCs, and SCAs. General BDOs
were evaluated by all the applicable airlines.

176
Source: Own illustration.
177
Only answers from the airlines which use the respective KPI are included.
264 Financial impact of big data for airline network planning

Route and network profitability

Figure 6.3 – Estimated impact of BDOs on route and network profitability178

Seven of the case study group airlines use route or network profitability as
primary or secondary KPI179. All of the business models are represented.
The four cargo-specific BDOs achieved the highest scores of 4.0, which
indicates that cargo airlines expect a very high impact of all BDOs on prof-
itability. Passenger airlines expect the highest profitability impact from ho-
tel price analysis (4.0) and mobile location data (3.7). General search en-
gine data received a high impact score on profitability across all airlines
(3.5). The two “planning constraint” BDOs were rated the lowest, with low

178
Source: Own illustration.
179
LCC 2 and CAR 3 do not use route or network profitability.
Choice of appropriate BDOs based on KPIs 265

and medium impact scores of 1.5 and 2.3 respectively. Besides the low
ranking of the “planning constraint” BDOs, there was no clear pattern
across the need for information clusters.

Load factor

Figure 6.4 – Estimated impact of BDOs on load factor180

Only CAR 1 uses load factor as a secondary KPI for the NP departments.
Thus, Figure 6.4 represents the answers of only this airline. Shipping data
from e-commerce platforms is the only BDO with an expected very high
impact on the load factor. E-commerce usually entails small merchandise
in boxes, which can be assimilated on pallets. Thus, a detailed forecast of
e-commerce shipments can improve load planning, thereby improving the

180
Source: Own illustration.
266 Financial impact of big data for airline network planning

load factor. Similarly, cargo airlines can use real-time location data for
cargo (impact score of 3.0) to predict late arrivals and fill the load space
with alternative cargo, if possible. None of the other BDOs are expected to
have significant impact on the load factor.

Forecast accuracy

The two FSCs and SCA 2 use forecast accuracy as a secondary KPI for
NP. These airlines have long planning cycles induced by their business
model and thus rely on the precision of mid- and long-term forecasts. Ac-
cordingly, general search engine data (4.0) and meta-search website data
(3.7) are expected to have the greatest impact on forecast accuracy. Local
hotel prices (3.3), events from social networks (3.0), and mobile location
data (3.0) may still have a high impact on forecast accuracy. The “planning
constraint” BDOs were ranked the lowest, whereas cargo-specific BDOs
were not ranked at all.

The pattern is not surprising since airlines focus on BDOs that contribute
primarily towards demand forecast and trend identification, thus strength-
ening the predictive power of forecasting.
Choice of appropriate BDOs based on KPIs 267

Figure 6.5 – Estimated impact of BDOs on forecast accuracy181

Schedule robustness

LCC 2 and CAR 1 measure schedule robustness as a secondary KPI for

their NP departments. The late deliveries of cargo shipments are a major
driver of delays in the air cargo business. These can be mitigated by ob-
taining data from real-time location transmitting cargo sensors, which are
assigned a very high potential (4.0) by CAR 1. LCC 2 expects information
on the age structure of passengers from demographic profiling. This infor-
mation can help anticipate the likelihood of passengers requiring special
assistance, which is a major driver of ground operations handling delays

181
Source: Own illustration.
268 Financial impact of big data for airline network planning

for LCC 2. All other BDOs bear no to medium impact for schedule robust-
ness.

Figure 6.6 – Estimated impact of BDOs on schedule robustness182

It is surprising that none of the “planning constraint” BDOs are expected to

have a high impact, since information on slots and traffic rights can help
calculate the impact of delays at specific airports (e.g., due to unavailability
of later slots). In contrast, the low expected potential of “demand forecast”
and “trend identification” BDOs seems logical, since the influence of de-
mand on schedule robustness is limited.

182
Source: Own illustration.
Choice of appropriate BDOs based on KPIs 269

Asset utilization

Six airlines across all business models consider asset utilization to be a

KPI for NP183. Asset utilization can be improved by changing the route mix,
i.e., closing schedule gaps by adding additional flights or re-scheduling ex-
isting flights. For cargo airlines, idea generation for the next best route is a
crucial lever, which is supported by their high impact rating for air cargo
rates (4.0) and e-commerce shipping data (4.0). Both the BDOs contribute
towards identifying value pockets in the form of unserved or underserved
routes, which air cargo network planners can use to optimize their route
mix.

Figure 6.7 – Estimated impact of BDOs on asset utilization184

183
LCC 2 and SCA 2 use it as a primary KPI; FSC 2, LCC 1, SCA 1, and CAR 2 use it as a
secondary KPI.
184
Source: Own illustration.
270 Financial impact of big data for airline network planning

Passenger airlines rated local hotel price analysis the highest (3.5), as that
may enable the discovery of untapped revenue pockets and adjustment for
micro-seasonality. Real-time BDOs are not deemed relevant for asset uti-
lization across passenger and cargo airlines. The airline network planners
attested that several BDOs have a medium impact on asset utilization (A4,
A5, A6, B4, and H1), but there is no clear recognizable pattern. However,
cargo airlines expect more impact potential on asset utilization than pas-
senger airlines.

FTE efficiency

Three airlines (FSC 1, LCC 2, and CAR 2) calculate FTE efficiency as a

secondary KPI for the NP department. All three airlines attested to real-
time slot trading platform having a high or even a very high impact on FTE
efficiency.

It seems that slot management is a time-consuming process step, which

binds significant FTE capacity. CAR 2 also perceives a high potential for
FTE efficiency improvement in the use of extended e-AWBs and location
transmitting cargo sensors. While the latter would mostly benefit tactical
NP, slot management may save time across all NP process steps.

Five BDOs (A4, A5, A6, D1, and H2) received a medium impact score by
the airline network planners. Those BDOs can contribute to reducing time
for information gathering and processing, thereby making NP employees
more efficient in their daily routine.
Choice of appropriate BDOs based on KPIs 271

Figure 6.8 – Estimated impact of BDOs on FTE efficiency185

Direct and indirect operating cost

SCA 1, LCC 1, and CAR 3 use DOC and IOC as cost KPI for NP. In con-
trast to other KPIs, the “planning constraint” BDOs received high impact
scores. A real-time slot trading platform has very high potential for cost
reductions (4.0), and a real-time traffic right information platform has a high
potential (3.5). Real-time slot management and traffic right information can
help reduce delay-induced cost and fines for exceeding traffic rights or
missing assigned slots. Mobile location data achieved a medium impact
score of 2.0, whereas all the other BDOs were considered to have low or
no impact.

185
Source: Own illustration.
272 Financial impact of big data for airline network planning

Figure 6.9 – Estimated impact of BDOs on DOC and IOC186

If the impact of BDOs is compared across the seven different KPIs, it is

noteworthy that no BDO has a high impact on all the KPIs, but every BDO
influences at least one KPI highly or very highly (see Table 6.1). Further-
more, every BDO has “blind spots,” meaning that it has no significant im-
pact on one or more KPIs. Therefore, there is no single BDO that suits all
NP KPIs. Thus, airline network planners should select the suitable BDOs
based on their specific KPI mix to maximize the impact.

186
Source: Own illustration.
Choice of appropriate BDOs based on KPIs 273

Table 6.1 – Impact of BDOs on airline network planning KPIs187

Schedule robustness
Forecast accuracy

Aircraft utilization
Route & network

FTE efficiency
Load factor
profitability

Direct cost
BDOs
Detailed demographic profiling with social
3 0 2 3 2 2 0
media mining
General search engine data for detailed
3 2 4 0 2 2 0
O&D demand
Clickstream data of meta-search websites
3 2 3 0 2 2 0
for flights
Scraping of published rates of cargo ship-
4 2 0 0 4 1 0
ping portals
Shipping data from e-commerce platforms 4 4 0 0 4 1 0

Scraping of regional hotel prices 4 0 3 0 3 0 1

Analysis of events on social media plat-
2 0 3 0 2 2 2
forms
Mobile location data 3 0 3 0 2 2 0
Extended electronic AWB with shipper ad-
4 2 0 2 0 3 0
dress
Location-transmitting sensors for cargo
4 3 0 4 0 3 1
shipments
Real-time slot trading platform 2 0 1 2 2 4 4

Real-time traffic right information platform 1 0 0 1 1 2 3

Legend: 4 very high - 3 high - 2 medium - 1 low - 0 very low/none (rating derived from
Figures 6.3-6.9)

187
Source: Own illustration.
274 Financial impact of big data for airline network planning

It is worth noting that the results of the impact assessment are not statisti-
cally significant, since the number of individual assessments is low due to
the limited size of the case study group. Including more airlines in a broader
survey will increase the external validity of the analysis and bridge the as-
sessment gaps, such as that of the load factor, which were assessed by
one cargo airline only. However, the result patterns confirmed the earlier
hypothesis that there is no one-size-fits-all solution for airlines when it
comes to choosing the most suitable BDO for airline NP.

6.1.2 Suitable KPI metrics for BDOs in airline NP

Airlines must select the right KPI metrics to calculate the expected benefits
from BDO deployment. For most KPIs, there are multiple potential
measures, including static and dynamic metrics as well as metrics focusing
only on a KPI aspect that is relevant for NP. Dynamic metrics measure the
change in a specific KPI over time and are preferable to static measures,
which measure a KPI at a specific point of time188. Furthermore, the metrics
should fall in the domain of airline NP. External factors, such as fuel prices
or economic downturns, should be excluded from the NP-specific benefit
analysis to represent the actual impact of BDOs on NP.

Figure 6.10 collates a range of possible metrics for each of the seven KPIs
used by the case study group. Route profit, for example, can be calculated
using static metrics, such as the total or average route profitability for a
given period. This global static metric is influenced by several internal and
external factors (e.g., economic cycle, fuel prices) with limited influence by
the NP department. Alternatively, only the routes with long-term negative
performance can be used as an indicator for unsuccessful route decisions.
This metric is static but focuses on a remit that is more controllable by a

188
BDO impact should be assessed over an investment time period (e.g., 5 years) to inform
a business case calculation.
Choice of appropriate BDOs based on KPIs 275

NP department if the objective is to minimize loss-making routes. A dy-

namic metric can measure the performance of new routes, which are fully
under the responsibility of the NP department. This metric is a concrete
measure for decision-making quality, which should also be improved by
the deployment of BDOs. Including dropped routes would tune the metric
to become more holistic in terms of NP responsibility, as it also reflects the
decisions on terminated services189.

Figure 6.10 – Relevant metrics for KPI calculation in airline NP190

The metric “profit delta from flight portfolio changes” compares the financial
performance of new flights with the financial performance of terminated
services. Ideally, the new flights should show a significantly higher profita-
bility than the ended flights, thus proving the value added by NP. Choosing

189
Otherwise, highly profitable routes can be replaced with other profitable routes, which will
not affect the total route profit for the airline.
190
Source: Own illustration.
276 Financial impact of big data for airline network planning

“profit delta from flight portfolio changes” or only the “new flight perfor-
mance” depends on the capacity development of the airline. An airline with
little capacity growth needs to replace its existing services if it wants to
pursue new route opportunities, while an airline with strong capacity growth
can simply add new flights without terminating the existing ones. For air-
lines with high capacity growth, the “new flight performance” metric is more
suitable. For all other airlines, the “profit delta from flight portfolio changes”
metric is preferable to measure the BDO impact on route profitability.

Unfortunately, this logic does not consider network effects, comprising up-
line and downline revenues. If an airline uses network profitability as a KPI
instead of route profitability, the extraction of the NP-specific impact is chal-
lenging191. One option to solve the problem is using the total network profit
as a global and static metric, acknowledging that the influence of NP is
limited. Alternatively, the “route portfolio delta” metric can be calculated as
well, omitting the network effects. The total network profit is chosen as a
metric for network profitability, as it is the only metric that specifically ad-
dresses network profitability.

Forecast accuracy requires dynamic metrics by definition, since the differ-

ence between the forecasted performance and the actual performance is
at the core of this KPI. Usually, forecast accuracy is measured by compar-
ing the forecasted demand and/or profit with the actual figures for a single
period at the route level. The average relative forecast deviation then ex-
presses the accuracy of the forecast. If an airline wants to reduce extreme
outliers, the deviation variance can be calculated as well. Some airlines
accept deviations if they are “positive surprises” but want to control nega-
tive forecast deviations. However, considering only the negative deviations
creates an incentive for the NP departments to forecast demand and profit
very conservatively. This can lead to the refusal of new route opportunities,

191
Upline and downline revenues tend to display significant variance on many routes, so the
dynamic comparison is diffused (Sterzenbach et al., 2013, p. 393).
Choice of appropriate BDOs based on KPIs 277

although they would be financially beneficial. The initially presented metric

“average forecast deviation per route” is the most holistic measure of fore-
cast accuracy and is thus recommended for a BDO benefit analysis.

The load factor calculation is standardized, since it expresses the percent-

age of available capacity that has been sold. In order to balance different
route lengths and aircraft sizes, the load factor is calculated on a unit basis,
which is seat kilometers for passenger airlines and ton kilometers for cargo
airlines. Like network profitability, load factor is a global and static metric,
influenced by many factors. Low-cost carriers tend to increase the load
factor by aggressive pricing, which may lower yield and profit, despite high
load factors. Load factors are only partially in the remit of NP and require
extensive assumptions if used as a KPI for the evaluation of a BDO.

Asset utilization is typically measured with the average block hours that
aircrafts fly per day. The objective is to minimize the unproductive time on
the ground and distribute the aircraft financing cost along more flight hours,
thereby reducing the indirect unit cost. The aircraft utilization can be differ-
entiated by aircraft type, as long-haul and short-haul aircrafts have signifi-
cantly different operation patterns. However, asset utilization cannot be in-
creased indefinitely, because maintenance schedules and turnaround
times are required in every rotation plan. For this reason, some airlines use
an alternative metric, called “idle hours.” “Idle hours” measure only the un-
productive time during which an aircraft is parked idle on the ground. NP
departments are responsible for the minimization of the unproductive
ground time with schedule optimization and find suitable flights instead.
Therefore, “idle time” is fully in the remit of NP, but it is still a static measure.
The metric can be dynamized by measuring only the change in average
idle time per aircraft over two different periods. These times are aggregated
and converted to a reasonable number of flights that can be operated in
this idle time to create a monetary benefit value. This metric is both dy-
namic and NP controlled. Hence, it is preferred to assess BDO value con-
tribution.