o-d control abuse by distribution systems: pods simulation results dr. peter p. belobaba...
TRANSCRIPT
O-D Control Abuse by Distribution Systems: PODS Simulation Results
Dr. Peter P. Belobaba
International Center for Air Transportation
Massachusetts Institute of Technology
AGIFORS Reservations and YM Study Group Meeting
Berlin, Germany
April 16-19, 2002
2
Outline
• PODS RM Research at MIT Simulated Revenue Benefits of Network RM
• O-D Control “Abuse” by Distribution Systems Example of Fare Search Abuse
• Simulated Revenue Impacts of Abuse Methodology for PODS Simulations Proportion of Passengers Committing Abuse
• Potential Threat to O-D Control Revenue Gains Options for Dealing with Abuse
3
PODS RM Research at MIT
• Passenger Origin Destination Simulator simulates impacts of RM in competitive airline networks
Airlines must forecast demand and optimize RM controls Assumes passengers choose among fare types and airlines,
based on schedules, prices and seat availability
• Recognized as “state of the art” in RM simulation Realistic environment for testing RM methodologies, impacts
on traffic and revenues in competitive markets Research funded by consortium of seven large airlines Findings used to help guide RM system development
4
PODS Simulation Flow
PASSENGERDECISION
MODEL
REVENUEMANAGEMENT
OPTIMIZER
FORECASTER
HISTORICALBOOKING
DATA BASE
CURRENTBOOKINGS
HISTORICALBOOKINGS
FUTUREBOOKINGS
PATH/CLASS AVAILABILITY
PATH/CLASS BOOKINGS/CANCELLATIONS
UPDATE
5
PODS Network D Description
• Two airlines competing in realistic network: 40 spoke cities with 2 hubs, one for each airline 20 spoke cities on each side located at actual US cities Unidirectional : West to east flow of traffic Each airline operates 3 connecting banks per day at its own hub Connecting markets have choice of 6 scheduled paths per day O-D fares based on actual city-pair published fare structures 252 flight legs, 482 O-D markets
• Airlines use same or different RM methods to manage seat availability and traffic flows.
6
Geographical Layout
H1(41)
H2(42)
4321
109
87
65
1517
161412
11 22
21
2019
18
2827 26
252423
33 32313029
39
38
37
36
3534
40
13
7
Revenue Management Schemes
BASE: Fare Class Yield Management (FCYM) Demand forecasting by flight leg and fare class EMSRb booking limits by leg/fare class
“Vanilla” O-D Control Schemes: Representative of most commonly used approaches Heuristic Bid Price (HBP) Displacement Adjusted Virtual Nesting (DAVN) Nested Probabilistic Network Bid Price (PROBP)
8
RM System AlternativesRM System Data and
Forecasts Optimization Model
Control Mechanism
FCYM Base Leg/class Leg EMSR Leg/class Limits
Heuristic Bid Price
Leg/bucket Leg EMSR Bid Price for Connex only
Disp. Adjust. Virt. Nesting
ODF Network + Leg EMSR
Leg/bucket Limits
Prob. Netwk Bid Price
ODF Network Converg.
O-D Bid Prices
9
Revenue Gains of O-D Control
• Airlines are moving toward O-D control after having mastered basic leg/class RM fundamentals
Effective fare class control and overbooking alone can increase total system revenues by 4 to 6%
• Effective O-D control can further increase total network revenues by 1 to 2%
Range of incremental revenue gains simulated in PODS Depends on network structure and connecting flows O-D control gains increase with average load factor But implementation is more difficult than leg-based RM
10
Network D Revenue Gain ComparisonAirline A, O-D Control vs. FCYM
0.00%
0.50%
1.00%
1.50%
2.00%
2.50%
70% 78% 83% 87%
Network Load Factor
HBP
DAVN
PROBP
11
Benefits of O-D Control
• Simulation research and actual airline experience clearly demonstrate revenue gains of O-D control
Return on investment huge; payback period short Even 1% in additional revenue goes directly to bottom line
• O-D control provides strategic and competitive benefits beyond network revenue gains
Real possibility of revenue loss without O-D control Improved protection against low-fare competitors Enhanced capabilities for e-commerce and distribution Ability to better coordinate RM with alliance partners
12
O-D Control System Development
• Based on estimates of network revenue gains, airlines have pursued development of O-D controls:
Up-front investments of millions, even tens of millions of dollars in hardware, software and business process changes
Potential revenue benefits of tens or even hundreds of millions of dollars per year
• At the same time, GDS and website technology has evolved to provide “improved” fare searches:
Objective is to consistently deliver lowest possible fare to passengers and/or travel agents in a complicated and competitive pricing environment
13
“Abuse” of O-D Controls
• Example 1: Booking connecting flights to secure availability, then canceling 2nd leg and keeping low fare seat on 1st leg.
Most airlines with O-D control are well aware of this practice, usually done manually by travel agents
Can be addressed with “Married Segment” logic in CRS
• Example 2: Booking two local flights when connecting flights not available, then pricing at the through O-D fare in the same booking class.
Appears to be occurring more frequently, as web site and GDS pricing search engines look for lowest fare itineraries
14
Requested Itinerary SEA-(HUB)-BOS
SEAMLESS O-D AVAILABILITY
SEA-BOS YBM (connecting flights)
SEA-HUB YBMQ (local flight)
HUB-BOS YBMQ (local flight)• O-D control optimizer wishes to reject connecting path and accept 2 locals with higher total revenue
HUB
BOSSEA
Q=$100Q=$150
Q=$200
15
O-D Abuse by Fare Search Engines
• In our example, a passenger wishes to travel from SEA to BOS (via HUB):
Airline’s O-D control system has determined that $200 Q fare SEA-BOS should be rejected
However, Q fare remains open on SEA-HUB and HUB-BOS legs, with expectation of ($100+$150) $250 in total revenue
• Travel agent or search engine finds that two local legs are still available in Q-class:
PNR created by booking two local legs separately But, GDS then prices the complete BOS-SEA itinerary at
$200, leading to $50 network revenue loss for airline
16
Revenue Impacts of O-D Abuse
• This type of abuse affects only O-D RM methods: Fare class control with EMSR does not distinguish between
different O-D itineraries in same booking class No revenue impact on EMSR control
• How big is the revenue impact on O-D methods? Clearly, abuse bookings can reduce the incremental
revenue gains of O-D methods over EMSR leg fare class methods
Depends on how widespread abuse booking practices are (i.e., proportion of eligible booking requests that actually commit abuse)
17
Simulation of Abuse in PODS
• For every O-D/fare in the network, we generated two path alternatives:
The connecting path priced at the published O-D fare A path comprised of the two local legs, also priced at the
connecting (through) O-D fare
• When the connecting path is closed by the O-D RM system, passengers look for the “local” alternative:
Only the passenger choice process is affected Airlines still perform RM assuming that sale of two local
seats will generate revenue equal to sum of two local fares
18
Simulation Set-Up
• PODS Domestic Network D• Average Network Load Factors = 77%, 83%, 88%• Probability of Abuse from 0% to 50%, in 10% increments for both leisure and business travelers.
We assume initially that probability of abuse is the same for each passenger type
• RM database records abuse bookings as path bookings accepted after path/fare was closed:
Historical abuse bookings added to detruncated estimates of path/fare booking demand, distorting future forecasts
19
O-D Control Gains Under Differing Abuse Probabilities (Base Case: EMSRb Fare Class Control, LF=83%)
-0.40%
-0.20%
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
1.40%
1.60%
0% 10% 20% 30% 40% 50%
Probability of Abuse
Reve
nue
Gai
n
HBP DAVN ProBP
20
Simulated Revenue Impacts
• Revenue gains of O-D methods drop from 1.4% with no abuse to almost zero at 50% abuse:
DAVN revenue gains are least affected, dropping to 0.55% over EMSRb base case at 50% probability of abuse
ProBP and HBP are affected more substantially, dropping to almost zero and even small negative revenue impacts
• Several factors contribute to revenue losses: Direct losses from taking lower connecting fare than
expected two local fares Distortion of demand forecasts leads to subsequent errors in
estimation of network displacement costs and bid prices
21
Impacts Depend on Abuse Probability
• Simulations show that more than 50% probability of abuse required to wipe out O-D revenue gains:
Of all opportunities where two local legs are open and the connecting path is closed in the same fare class, more than half of passengers would have to abuse the O-D controls
•Actual probability of abuse appears to be low: Anecdotal evidence suggests 10-20% or less But evolution of web site and GDS search engines raises
concerns that this probability will continue to grow For time being, more likely that leisure travelers paying lower
fares are involved in O-D abuse, not business travelers
22
Impacts Differ by Network ALF
• Negative impacts on revenue gains are more dramatic at higher network load factors:
Because revenue gains of “perfect” O-D control are higher at higher demand levels, more to lose with O-D abuse
25% probability of abuse reduces revenue gains by about 1/3 at 83% network load factor, and by 1/2 at 88%
At lower 77% network load factor, 25% probability of abuse actually leads to slightly higher O-D revenue gains (relative to EMSRb control), since additional traffic is accommodated with less displacement
23
O-D Revenue Gains with Varying Probability of Abuse (Base Case: EMSRb Fare Class Control, ALF=77%)
-0.50%
0.00%
0.50%
1.00%
1.50%
2.00%
0.00% 25.00% 50.00%
Probability of Abuse
O-D
Rev
enue
Gai
n
HBP
DAVN
ProBP
24
O-D Revenue Gains with Varying Probability of Abuse (Base Case: EMSRb Fare Class Control, LF=83%)
-0.50%
0.00%
0.50%
1.00%
1.50%
2.00%
0.00% 25.00% 50.00%
Probability of Abuse
O-D
Rev
enue
Gai
n
HBP
DAVN
ProBP
25
O-D Revenue Gains Varying Probability of Abuse (Base Case: EMSRb Fare Class Control LF=88%)
-0.50%
0.00%
0.50%
1.00%
1.50%
2.00%
0.00% 25.00% 50.00%
Probability of Abuse
O-D
Rev
enue
Gai
n
HBP
DAVN
ProBP
26
Summary of Findings
• Simulated negative revenue impacts due to “O-D” abuse by availability search and pricing engines:
Even at 10-20% probability of abuse, revenue gains of O-D methods are reduced by up to 1/3
Means actual revenue gain of ProBP is closer to 0.8% than estimates of 1.4% under perfect O-D control conditions
• Practical O-D control issues have an unexpected and substantial impact on network RM models:
Bid price methods appear to be more affected than DAVN, because forecasting distortions affect probabilistic bid prices more consistently than deterministic LP shadow prices.
27
Unanswered Questions
• How widespread is this type of O-D abuse? Certainly possible with manual action by travel agents Evidence of systematic abuse by some website and GDS
search engines
• Can these simulation results be generalized? Intuitively clear that overriding the optimized results of the
network RM system will lead to reduced revenues. Order of magnitude seems to be reasonable, dependent on
probability of abuse and network load factors
• Can airlines stop this abuse and revenue loss?
28
Possible Solutions
• Some airlines have considered (and implemented) “Journey Control” for PNR booking:
Recognize that first local leg has been booked and refuse second local leg availability if connection is not available
Also known as “shotgun wedding” controls in CRS
• Alternative is a “ticket as booked” policy: If two legs were booked based on local availability, then they
must be ticketed as two local fares Requires changes to GDS processing and/or travel agency
enforcement, which might be more difficult to implement