Event‐based allocation of airline check‐in counters: a simple dynamic optimization method supported by empirical data

• Published date: 01 September 2018
• Author: Moosa Sharafali,Brian Rodrigues,Mahmut Parlar
• DOI: http://doi.org/10.1111/itor.12332
• Date: 01 September 2018

Intl. Trans. in Op. Res. 25 (2018) 1553–1582

DOI: 10.1111/itor.12332

INTERNATIONAL

TRANSACTIONS

IN OPERATIONAL

RESEARCH

Event-based allocation of airline check-in counters: a simple

dynamic optimization method supported by empirical data

Mahmut Parlara, Brian Rodriguesband Moosa Sharafalib

aDeGroote School of Business, McMaster University, Hamilton, Ontario, L8S 4M4, Canada

bLee Kong Chian School of Business, SingaporeManagement University, 50 Stamford Road, Singapore,178899, Singapore

E-mail: parlar@mcmaster.ca[Parlar]; brianr@smu.edu.sg [Rodrigues]; sharafalim@smu.edu.sg [Sharafali]

Received 13 February2016; accepted 24 June 2016

Abstract

This paper studies the real-life problem of dynamically optimizing the number of airport check-in counters

to allocate for a single flight. The main feature of our work is the use of empirical data collected at the

Singapore Changi Airport, which drives the dynamic optimization model of a parallel queues system. We

propose an event-based dynamic programming model that simplifies considerably the optimization analysis

even for large-scale problems with 700+booked passengers. We investigate the following research questions:

(a) For a particular flight, what is the optimal number of counters the system should open with and what is

the corresponding optimal total cost? (b) Given the state of the system at any event epoch, should we open

another counter or not and what is the optimal cost-to-go from this state? The empirical data we collected

at the airport are used to test the assumptions, estimate the key parameters, and run the computational

experiments. We apply our model to 14 flights at the Singapore Changi Airport and identify cases in which,

depending on the cost parameters, the model advocates the use of either a dynamic or a static policy. Although

the model concerns only an exclusive-use system, it is flexible enough to apply to other configurations such

as a common-use system or a single-queue, multicounter system.

Keywords:airport operations; queueing; dynamic programming

1. Introduction

The problem studied in this paper was motivated by a study at the Singapore Changi Airport, often

voted the “Best Airport in the World.”1The airport strives to providebetter service for its customers,

and our focus was to improve queue management at one of its international terminals. Typically, as

in many high-volume airports, the passenger check-in service provider must decide the allocation

of a limited number of check-in counters to each flight/airline to suit flight schedules and planned

1http://www.changiairport.com/en/aboutus.html (accessed 29 January 2016).

C

2016 The Authors.

International Transactionsin Operational Research C

2016 International Federation of OperationalResearch Societies

Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA02148,

USA.

1554 M. Parlar et al. / Intl. Trans. in Op. Res. 25 (2018) 1553–1582

passenger loads. The challenge is to complete check-in within a specified time window without

having to open additional counters or compromise customer service standards. The practice is to

preallocate counters and open additional counters during check-in using experience-based rules

of thumb. A major concern for the service provider (a third-party company such as Hong Kong

Airport Services Ltd., in Hong Kong; SATS Ltd. and DNATA in Singapore) is to maintain high-

quality counter allocation schedules for each day, given variations in skill and experience levels of

human schedulers. The effectiveness of the service provider’s counter management has high impact

on the image of the airport and airlines that operate through it since an efficient schedule translates

to shorter queues and faster check-in. In addition, on-time counter closure impacts downstream

operations, including baggage processing and flight gate scheduling. It is clear that the service

provider has to perform a balancing act in meeting, (a) the demands of the airlines in terms of the

number of counters, (b) the service-levelstandards set by the airport authority, and (c) the concerns

about its own bottom line.

Thus, the problem studied in this paper, namely, determining the number of counters to allocate

to a single flight with the objective of minimizing counter operating and waiting time costs, is of

critical interest to the service provider. The check-in counter system in use is an exclusive-usesystem

dedicated to a single flight. It is worth pointing out that in countries such as the United States,

airlines have predominantly moved to using common-use counters that might make our model

seem to have limited applicability. But there are many prominent airports around the world, which

still use exclusive-use counters. In particular, in Singapore’s Changi Airport,2some of the terminals

employ an exclusive-use counter system. Some years ago, Singapore Airlines decided to do away

with self check-in kiosks3citing low usage of these kiosks and the growing popularity of Internet

and mobile check-ins. Although Changi Airport is now promoting the use of these self check-in

kiosks, we are of the view that it will be a very long time for these to become popular, as research

evidence points to the fact that “many passengers are reluctant to use new technology within the

public sphere for fear of social embarrassment” (Minton, 2008).

On this subject, Parlar and Sharafali (2008) were the first to propose a stochastic dynamic

programming (SDP) model based on the analysis of the underlying queue in the transient regime.

They modeled the system as a multiserver queue with the arrival process occurring according to a

passenger departure process from the finite population of passengers holding confirmed bookings.

In queuing parlance, this is referred to as a death process (see Feller, 1968). In their model, the

decision to open additional counters was made periodically at equally spaced time instants. This

required the time-dependent transition probability functions of the underlying absorbing Markov

process. Moreover, the periodic-review dynamic programming approach involved the evaluation of

a very large number of these functions that made it difficult to use for solving realistically sized

problems. Also, in that work, a single-queue/multiple-counter system was used because for some

of the flights, the service provider uses such a system. But at the terminal that we now study in this

paper, the service provider used parallel-queue/multiple-counter system due to counter layout and

space constraints. Another important observation we made, which was also pointed to us by the

service provider, is the fact that queue lengths are almost always equal as a result of jockeying—a

2http://www.changiairport.com/en/flight/departures.html (accessed 29 January2016).

3Kaur, K., 2011. SIA pulls plug on kiosks for self check-in. The Straits Times, 13 December 2011.

C

2016 The Authors.

International Transactionsin Operational Research C

2016 International Federation of OperationalResearch Societies