Revenue Management Saves National Car Rental

Revenue Management Saves National Car Rental
Author(s): M. K. Geraghty and Ernest Johnson

Source: Interfaces, Vol. 27, No. 1, Franz Edelman Award Papers (Jan. - Feb., 1997), pp. 107-
127
Published by: INFORMS
Stable URL: https://www.jstor.org/stable/25062214
Accessed: 12-12-2019 02:34 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Interfaces
This content downloaded from 155.207.206.147 on Thu, 12 Dec 2019 02:34:34 UTC
All use subject to https://about.jstor.org/terms
Revenue Management Saves National Car Rental
M. K. GERAGHTY 2188 Deep Woods Way

Marietta, Georgia 30062
ERNEST JOHNSON National Car Rental

7700 France Avenue
Bloomington, Minnesota 55435
In 1993, National Car Rental faced liquidation. General Motors

Corporation (National's parent) took a $744 million charge
against earnings related to its ownership of National Car Rental
Systems. National faced liquidation, with the loss of 7,500 jobs,
unless it could show a profit in the short term. National initi
ated a comprehensive revenue management program whose
core is a suite of analytic models developed to manage capacity,
pricing, and reservation. As it improved management of these
functions, National dramatically increased its revenue. The ini
tial implementation in July 1993 produced immediate results
and returned National Car Rental to profitability. In July 1994,
National implemented a state-of-the-art revenue management
system, improving revenues by $56 million in the first year. In
April 1995, General Motors sold National Car Rental Systems
for an estimated $1.2 billion.
In the late 1980s, the car rental industry were eroded. Automobile manufacturers
was in turmoil. Low profit margins were purchased almost all of the major car rental
subsidized by tax credits. When these tax companies and, in the early 1990s, flooded
credits disappeared, the low profit margins them with cheap fleet deals. These cars
Copyright ? 1997, Institute for Operations Research INDUSTRIES?TRANSPORTATION/SHIPPING

and the Management Sciences DECISION ANALYSIS?APPLICATIONS
0092-2102/97/2701 /0107$05.00
INTERFACES 27:1 January-February 1997 (pp. 107-127)
GERAGHTY, IOHNSON
came with large manufacturer cash incen willing to accept booking and travel restric
tives and could be disposed of quickly and tions, such as advance payment, Saturday
easily, as often as every four months, by night stopovers, and penalties for no-shows
simply returning them to designated auc and cancellations.
tions. The car manufacturers placed more The major car rental companies depend
emphasis on using their car rental subsid largely on corporate customers. They con
iaries to soak up excess production than to tract at fixed rates with companies who
produce profits. have numbers of employees who travel.
This excess supply in the marketplace led Demand peaks for rental cars midweek,
to low pricing. Several major competitors, forcing all companies to regularly turn
the price leaders, paid undue attention to down customers. The business customer,
market share and made emotion a variable who typically travels on these days, pays a
in the pricing equation. Companies still use fixed corporate rate. This leads to a large
very low pricing during periods of low de excess fleet that is idle on weekends. The
mand. These are the rates quoted in Sabre car rental industry allows price-sensitive
on February 9,1996 for a weekend rental of leisure customers to book multiple reserva
a subcompact car at Greensboro, North tions with no prepayment required. There
Carolina for pickup on February 29,1996: are rarely penalties for cancellations or no
Alamo $14.99 shows. Customers arriving as much as 12
National $15.99 hours after the specified time of reservation
Budget $16.95 are given the reserved car at the reserved
Avis $16.99 rate. These policies result in no-shows that
Hertz $16.99 sometimes exceed 50 percent of reserva
These prices include unlimited mileage.tions. This is a major problem for the in
dustry, which must maintain high
A comparison of these rates with the cost
of renting a tuxedo underscores the freutilization to make a profit.
National Car Rental Background
quent irrationality of pricing. In the early
1990s, economic conditions and improveBefore National began using revenue
ments in design and production quality im
management, it struggled with the same
proved demand for American-made cars.challenges as its competitors. But other fac
The manufacturers dramatically raised the tors made it critical for National to change
quickly. National's business was predomi
costs of cars to their car rental companies.
These market pressures, combined with nantly composed of corporate customers,
the fact that the car rental industry waswho rented cars midweek. National's strat
egy focused on these business renters and
slow to apply technology, precipitated an
neglected the leisure customers. For several
industry in crisis. By comparison, the air
years, starting in 1987, National had no sig
line industry has successfully demonstrated
how to apply the technology of revenuenificant advertising campaign. It planned
management in a service industry with its fleet in one-year cycles, and made very
high equipment and labor costs. Airlinesfew changes in fleet deployment to meet
regularly sell cheaper seats to customerschanging customer demand.
INTERFACES 27:1 108
NATIONAL CAR RENTAL
National had three legacy systems to eroding customer satisfaction. Roll-out of

build on: The newly-developed Vehicle In these processes in the organization was met
formation System (VIS) accurately tracked with stiff resistance. "We can't do this;
the fleet, National's Reservation System we'll confuse our customers," and "If we
(RES) was efficient at booking reservations, do this, we'll confuse ourselves" were typi
and the Expressway System (NEX) pro cal comments. Deciding that it needed help
vided the most rapid rental-and-return pro breaking through these barriers, National
cess in the industry. By contrast, pricing asked Aeronomics Incorporated, a revenue
changes were manual and extremely time management consulting company, to evalu
consuming. Changes were keyed into Na ate its unrealized revenue opportunities.
tional's rates system and then keyed in National was at a critical juncture. Gen
again to the airlines' computerized reserva eral Motors had mandated that the com
tion systems (CRS). Setting pricing was a pany either become immediately profitable,
shared responsibility. City managers, mar so that it could be sold, or be liquidated.
keting, regional VPs, senior management, National had already undertaken cost-cut
and the pricing group all shared input, ting measures. It had to make more money
with no single person ultimately responsi with the existing operation. Larry Ramaekers,
ble for a location's pricing. Inventories in assigned by turnaround specialists Jay Alix
the CRS were controlled by field managers and Associates, led the turnaround and
with no sophisticated system advising acted as president. "We decided to go for a
them when to increase or restrict availabil revenue-based turnaround as opposed to a
ity. No demand forecasts existed at either cost-cutting turnaround" said Ramaekers
the city or the corporate level. [1995].
Revenue Management at National Car Senior management agreed to conduct a
Rental needs assessment with Aeronomics Incor
In February 1992, several National execu porated between January and April 1993.
tives identified two key issues: (1) National The mission was to understand National's
was turning down large numbers of cus business, quantify revenue potential, rec
tomers when cars were available to meet ommend organizational structure and staff
their needs; (2) competitors were raising ing requirements, define automation re
their leisure prices as the date of rental ap quirements, estimate costs, provide cost/
proached, while National's pricing re benefit analysis, and prioritize an imple
mained stagnant. National formed the rates mentation plan. The assessment identified
automation team (RAT) with a limited mis opportunities for increasing revenue and
sion to determine whether, during periods was presented to senior management in
of high demand, National could raise its April 1993. National's owner, General Mo
prices in the seven booking days before tors, agreed that implementing a revenue
rental. It selected a limited number of cities management program would be the key
for a pilot test, for which all processes were impetus to National's turnaround and com
manual. It quickly determined that it could mitted over $10 million to design and build
raise prices and increase revenues without a revenue management system (RMS), ac
January-February 1997 109
GERAGHTY, JOHNSON
quire the necessary hardware, make customer's rental car. Overbooking and res
changes to legacy systems and bridge them ervation control are common to any indus
to RMS, and build a dedicated revenue try that allows advanced bookings. There
management department. The department are important differences. Airlines can tar
would comprise 30 specialists, focusing all get low rates precisely at underutilized ca
of their talents and energies on generating pacity. In the car rental situation, the prob
revenue. The application would be phased lem is more complex. Rate cuts designed to
in beginning July 1993 and would be rolled stimulate demand on low utilization days
out to all locations by early 1995. The ex may increase demand on a day when ca
tremely rapid development of RMS and its pacity is constrained and compound the
immediate implementation to control Na problem. Managing the problem of days
tional's largest demand centers was the sin when supply is constrained by controlling
gle most important factor in keeping Na the length of rentals is a more effective so
tional alive. lution, but it requires surgical precision.
Revenue management achieves its reve Conversely, it is reasonably straightforward
nue gains by applying analytic models and to increase RPD by increasing rental rates.
methodologies to a planning horizon. By However, the rental car market is ex
consistently managing capacity, price, and tremely competitive. A price move that
booking requests in a manner that im makes the company more expensive than
proves revenue per car (RPC), revenue per its competitors can damage utilization lev
day (RPD), and utilization levels, a com els. A high RPD is not worth much if most
pany can make and sustain revenue im of the fleet is sitting on the lot.
provements. There are a couple of basic Information Systems
prerequisites for applying revenue manage The revenue management system was
ment in a new industry. Perishability is one developed jointly by Aeronomics Incorpo
of the most important prerequisites as rated, a revenue management firm; EDS,
Weatherford and Bodily [1992] discuss in National's information services provider;
their paper on perishable-asset revenue and National. It is central to the flow of in
management. The unit of inventory at Na formation at National Car Rental. EDS im
tional is the car rental day, which is lost if plemented a comprehensive set of data
it is not utilized. Another prerequisite is a links with existing information manage
segmentable market. (For a discussion of ment systems. In particular, the link be
how revenue management capitalizes on tween RES and RMS is unparalleled in the
consumers' differential willingness to pay, industry. It is a continuous transaction
see Cross [1986].) level data feed of all advanced booking ac
The rental car problem exhibits a similar tivity, including availability and booking
structure to the airline problem, and tech restrictions. The continuous feed approach
niques developed to solve the airline prob provides up-to-the-minute booking levels,
lem have been particularly useful. For ex forecasts, and system recommendations to
ample, allocating airline seats at discount revenue managers through the RMS graph
prices translates to planning to upgrade a ical user interface. Transactions include
INTERFACES 27:1 110
NATIONAL CAR RENTAL
bookings, cancellations, turndowns, and decentralization would cause a number of

shoppers. Turndowns are booking requests difficulties:
that the company did not accept because of ?City managers would not make revenue
availability controls or booking restrictions. generation the highest priority, because
Shoppers are booking inquiries that did not their most immediate problems are cus
convert to booking requests (Figure 1). tomer service and vehicle maintenance.
Revenue Management Organization ?Recruiting and training personnel and
A major issue during the design phase of equipping city offices would be very ex
RMS was to determine whether a decen pensive with a long lead time.
tralized or centralized solution would have ?The revenue manager would be a gener
the largest revenue impact, short and long alist and would be assigned to "burning
term. A decentralized organization would problems" not related to revenue genera
have been the least painful solution cultur tion.
ally, because city operations managers con ?Pricing practices would not be consistent
trolled inventories (that is, reservations sys across locations.
tem inventories) and leisure pricing. But ?Managers might be parochial concerning
Inventory Control
Figure 1: RMS synthesizes information from four principal information systems. Expressway pro
vides current fleet levels to support the capacity-management model and post-arrival data, such as
no-shows and walk-ups, for the forecast of day-of-arrival activity. RES provides transaction-level
information on booking activity. RMS availability and length-of-rent controls are communicated
to RES after review and action by a revenue manager. The Rates system maintains current rate
level information. RMS recommends rate adjustments and provides an interactive rate update
interface to the Rates system. Availability and rates are available on a number of airline CRS
(central reservation systems). RES updates the CRS whenever availability controls or booking
restrictions change. Rates from RMS update CRS rates at regular intervals throughout the day.
GERAGHTY, JOHNSON
fleet disbursement. by accepting or rejecting booking requests

?Revenue managers would vary in levels based on length-of-rent controls. A sophis
of skills in the field. ticated set of forecasts of demand and con
The decision to create a centrally located sumption patterns supports the analytic
team addressed all of these issues and al model (Figure 2).
lowed for the rapid change needed for im RMS functions as both an automated de
mediate impact. Corporate revenue manag cision management system and an interac
ers were given responsibility for pricing tive decision support tool. Overnight pro
and inventory management at the location cesses execute the forecasting and analytic
level. This allowed the revenue manage models to generate recommendations con
ment department to share its expertise and cerning availability, rate, and length-of-rent
information directly with marketing, corpo control. Revenue managers review
rate sales, fleet, strategic analysis, and sen thousands of recommendations each day
ior management. and make, accept, reject, or override deci
Traditional revenue management organi sions based on their knowledge of current
zations in the airline, hotel, and car rental market conditions and forecasted demand.
industries have left a dichotomy between Revenue priority indicators assist work
the inventory and pricing functions. This is flow management by pointing out the
a result of the problem structure. Inventory greatest revenue opportunity.
is usually controlled at the level of a non
stop flight leg, a room night, or a rental
day. Pricing is focused at market segments
Demand
O On-rent Fest
that are often multi-leg, multi-night, or

multi-day. However, pricing decisions LOR Fest
made independently of fleet availability Capacity Mgt
and customer demand retard revenue gen
eration. The decision to vest National's rev Res Control
enue managers with control of both pricing

Pricing
and inventories represented a big change in
culture for National and pioneered new
territory for a revenue management Figure 2: Historic and current demand from the
marketplace is used to produce the demand
organization. forecasts that support RMS's analytic models.
Revenue Management System The capacity-management process determines
The National RMS supports three pri availability levels from the on-rent forecast.
Pricing recommendations complement the
mary business functions: capacity manage
availability settings. The reservations control
ment, pricing, and reservations control. The process uses more detailed length-of-rent
capacity management function targets (LOR) forecasts to generate booking restric
high-valued fleet utilization. Pricing en tions. The combination of availability controls,
price adjustments, and booking restrictions
hances corporate revenues through sensi
changes booking pace. The new booking pace
tivity to consumer price tolerance. results in new forecasts and system
Reservations control maximizes revenues recommendations.
INTERFACES 27:1 112
NATIONAL CAR RENTAL
Revenue managers have a wealth of sup Forecasting

port tools at their fingertips during the re A comprehensive set of demand and rev
view process. The multiple document inter enue forecasts supports the analytic mod
face allows them access to reports with els. Demand levels are forecasted at two
historical data and demand and revenue primary levels of aggregation: length-of
forecasts at all levels of aggregation. They rent and on-rent. Both forecasts represent
can cut and paste information into spread unconstrained demand, which is the num
sheets, word processors, or electronic fax. ber of cars that can be rented if there are no
One of the more striking features of RMS is capacity restrictions. The length-of-rent
the degree of interaction it permits be forecast is a forecast of unconstrained de
tween the revenue manager and the sys mand for each potential length of rent, for
tem models. A refresh feature accesses the each arrival day in the planning horizon.
most recent data on the reservations sys National intensively manages the booking
tem and recomputes forecasts based on process at least 60 days in advance of day
intraday booking-pace profiles. Recom of pickup. The system generates forecasts
puted forecasts generate new system rec for all days within this horizon. It uses de
ommendations, which the revenue man mand levels to derive optimal reservations
ager can compare to the overnight controls. The term on-rent refers to the
recommendations as part of the review number of cars in use on a specific date. It

process. An analyst who disagrees with combines cars that are picked up on that
the forecast can override values. The re day with cars that are already in use. Ca
fresh feature will revise the RMS recom pacity management and pricing models
mendations according to the user-sup rely on the on-rent forecast.
plied forecasts. The demand forecasting methodology for
The logical flow of the models within all levels of aggregation is based on a com
RMS matches the logical flow of marketing bination of long-term and short-term fore
activities within the company. Results from casting. The long-term forecast is a time
each model depend on the output of logi series model with seasonality factors
cally precedent models. Users can adjust derived from spectral analysis of historic
output of any of the models for what-if seasonality. Demand during special events
analysis. All downline models will then does not distort seasonality. The analyst
produce new recommendations. For exam with the best knowledge of the market de
ple, the analyst can experiment with differ fines special events and can override the
ent rate levels. New length-of-rent controls system-generated factors (Figure 4).
will reflect the change in the relative valua Curry [1993] pioneered the use of Kaiman
tion of different demand elements. If the filtering for revenue management. The vari
analyst wants to try different availability able gain approach has several benefits. Sys
levels, the pricing model will generate new tem initialization uses the same processes as
rates. New rates will cause the system to the nightly update. The initial gain is set to
recommend new length-of-rent controls unit value and the initialization process ad
(Figure 3). justs the gain level as it works its way
GERAGHTY, JOHNSON
Inventory
Optimization
Rates
RES
Figure 3: Business processes, analytic models, and revenue manager expertise are tightly inte
grated in RMS. Each of the ovals represents a point where analysts can adjust system forecast
and recommendations. If the analyst requests a refresh, RMS updates down-line results to ref
the user adjustments. Underlying the decision-making process are expectations of consumer b
havior. RMS provides demand forecasts at the on-rent and length-of-rent levels. The analyst c
refresh the forecast with real-time booking levels from the reservations system. The capacity
management models produce availability recommendations, which can also be adjusted. A re
fresh at this stage produces new rates and length-of-rent controls. Finally, rate revisions can re
in new MLRs (minimum length-of-rent controls). The analyst reviews system recommendatio
before sending them to the electronic distribution channels.
which bookings are expected to accumulate

through the available historical data. Another
in the reservation system. This offset pro
benefit of the variable gain approach is that
the revenue manager can adjust the respon vides information about the current book
siveness of the forecast temporarily to re ing pace, that is, the rate at which bookings
spond to expected changes in the marketare accumulating. The short-term forecast is
place. Once the change has taken place, the
the expected change in the number of
bookings. The final demand forecast is a
Kaiman filter gains tend toward the steady
state smoothing constant. combination of the long-term and short
In generating the short-term forecast, term
the forecasts. The long-term forecast pro
system considers the offset of the actualvides stable predictions early in the plan
reservations level from a booking curve.ning horizon. As more information
becomes available from the actual booking
The booking curve represents the rate at
INTERFACES 27:1 114
NATIONAL CAR RENTAL
Q-J-1-1-1-1-1-h?H-1-1-1-1-1-1-1-f?H-1-h
? ? 10 12 14 16 18 20 22 24
Time
Figure 4: One attractive feature of variable gains is proactive response. The analyst often has ad
vance knowledge that a significant change in the market is about to occur. By proactively increas
ing the gain, the analyst can ensure that the demand forecast responds to the new demand level
quickly.
behavior, the short-term model dominates turning away walk-up demand. Walk-ups
the forecast of final unconstrained demand. represent a significant revenue opportu
The system also generates additional nity during periods of high demand and
forecasts of day-zero activity, such as walk low product availability. By predicting
ups and no-shows. Day-zero is the day a walk-up activity, managers can set aside
car is picked up by the customer. Day-zero inventory. Walk-ups during these periods
activity includes reservations that do not represent an opportunity to achieve high
materialize, that is, no-shows, and requests revenue per day. Conversely, managers
for cars that do not come through the reser can stimulate walk-ups through aggres
vations process, that is, walk-ups. No sive pricing to compensate for underutili
shows are a big management problem. The zation identified late in the booking pro
no-show rate is a day-zero effect, but it cess. The management of day-zero
affects down-line rental days by reducing activity is the responsibility of the field
expected on-rent demand and creating ex managers. Hand-off of the booking pro
tra availability. cess from corporate revenue managers to
Walk-ups, on the other hand, introduce field staff on day-zero requires constant
significant opportunity. A manager can communication and information sharing.
avert an impending oversale situation by RMS contributes to this process with
GERAGHTY, JOHNSON
distributed electronic reports. fleet made available at the last minute by

Capacity Management the manufacturer, and move cars by truck
RMS converts available capacity into rev or train for longer distances. RMS provides
enue. Capacity management is the first step tactical forecasts, notifies the analyst of pe
in the process. It includes fleet planning, riods when capacity is dangerously low
planned upgrades, and overbooking. and automates key components of the tacti
Fleet Planning cal fleet-management process.
To plan fleet levels, National identifies The involvement of the revenue manage
how much of the available fleet should be ment department in long-term fleet plan
at each inventory location to meet expected ning has been instrumental in developing
demand. An inventory location is a geo National's annual budget for charge days
graphical area which shares a pooled fleet. and revenue per day for each market seg
It may include several different physical lo ment within each city. (A charge day is a
cations. For example, Minneapolis may be rental day that genuinely generates reve
an inventory location consisting of two city nue.) This has contributed dramatically to
stations?Minneapolis airport and a hotel National's success because the RMS fore
station. Capacity is managed at the inven cast focuses on unconstrained demand and
tory location level because fleet can easily has helped break the pre-RMS fleet-plan
be moved to meet demand at the various ning paradigm. In the past, National would
stations. In contrast, pricing is managed at plan its long-term fleets based on historical
the physical location level as a means of rental patterns, which restricted growth.
further segmenting the market. National re With accurate information illustrating un
lies on long-term forecasts of demand for constrained demand, National was able to
each inventory location in its fleet plan increase its fleet in a cost-effective manner
ning. It matches current fleet and expected to capture a much larger volume of profita
fleet adjustments to the demand forecasts ble business with resulting increases in
to determine the placement of cars. revenue and market share.
National plans its fleet in three stages: for Planned Upgrades
the short term, it looks at a five-day hori Revenue management exploits the rela
zon; for the medium term, it considers 60 tionship between segmenting the market
days in the future; and for the long term, it and generating revenue. Firms can achieve
looks over the coming 18 months. Short differential pricing for commodities by in
term planning targets fleet movements be stituting fences, such as advanced purchase
tween locations, accelerates or retards turn restrictions, that capitalize on each market
backs to manufacturers, regulates car-sales segment's willingness to pay. In this way a
activities, and resets one-way pricing to ef firm presents a variety of products, all of
fectively place cars in the proper locations which are based on a single commodity.
to satisfy demand. In mid-term fleet plan The fences discourage revenue dilution be
ning, managers consider the same actions cause the lower-valued products have re
as in the short-term process, but they also strictions that are unacceptable to the
redirect new car deliveries, acquire new higher-valued market segments. National
INTERFACES 27:1 116
NATIONAL CAR RENTAL
$100 Unrealized Revenue Potential

from Diluted Demand
Price
$50
Unrealized Revenue Potential
from Unaccommodated Demand
50 100
Demand
Unrealized Revenue Potential

from Diluted Demand
Price
Unrealized Revenue Pot

from Unaccommodated
20 40 60 80 100
Demand
Figure 5: If fences can be found that effectively segment the ma

of revenue opportunities that arises from differential pricing. It
prices at the willingness-to-pay level of each market segment. T
creased revenues, and the consumer benefits because the comm
market. Revenue management has been an important driver in
customer base.
117
January-February 1997
GERAGHTY, JOHNSON
Luxury
Midsize
Economy
\ 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
-10 -L
Demand
Figure 6: The planned-upgrades model uses a modification of the classical EMSR (expected mar
ginal seat revenue) heuristic that has evolved from Littlewood's [1972] and Belobaba's [1989] re
search. The cumulative demand distribution is multiplied by average revenue to get expected
marginal-revenue curves for each class. Inventory is protected for each class until its marginal
revenue falls below the revenue available from lower classes. Bookings are accepted for a class
until they exhaust all inventory not protected for higher classes.
has instituted competitively priced weekly other. There is a definite upgrade hierarchy
rates, which require the customer to keep that it needs to manage. The planned-up
the car over Saturday, for price-sensitive grades model fits demand into this hierar
leisure customers, fencing out the business chy in a way that minimizes revenue dilu
traveler who wants to be home on the tion while maximizing utilization. It
weekends (Figure 5). allocates inventory to booking classes based
Market segmentation is also inherent in on demand for the booking classes, the
the different car types. The demand for car number of vehicles that are acceptable to
types falls into definite market segments. renters in this class, and the expected reve
Business renters typically demand midsize nue associated with each class (Figure 6).
cars. Low-valued leisure customers prefer In a fully commoditized environment,
economy cars with low rates. More valu where all inventory is interchangeable,
able leisure customers want specialty vehi firms usually adopt a nesting approach.
cles, such as minivans or four-wheel-drive Nesting allows the most valuable booking
vehicles. The drawback of accomplishing class access to all available inventory. Sub
market segmentation with different prod sequent booking classes are allowed access
ucts is that a car rental company cannot to inventory nominally set aside for the
directly substitute one vehicle type for an next lower classes. The planned-upgrades
INTERFACES 27:1 118
NATIONAL CAR RENTAL
hierarchy interferes with a simple nesting planned-upgrades model takes account of

approach. It is inappropriate to put higher forecast variability and expected revenues
class customers in economy cars. However, to compute the marginal revenue for each
if the company restricts economy bookings class (appendix). The availability calcula
to economy cars only, it would miss a large tion is based on the relative marginal reve
nues.
revenue opportunity. Typical fleeting pol
icy is to acquire more large and midsize Planned-upgrades activity produces a
cars than expected demand requires and change in availability in the reservations
fewer economy level cars. The difference in system because customers are driving car
the costs of different sized cars is small in a class different from that which they
enough that the advantages derived from booked. When an economy customer ar
having enough high-valued inventory to rives at the rental counter and drives aw
meet the high-valued, high-demand peri in a midsize car, the availability for mids
ods justify the extra cost. Also, few econ demand is decreased for each day the car
omy customers complain about getting a on rent. This is because the rental contrib
better car than they booked for the same utes to the current on-rent value for mid
rate. The planned-upgrade model decides size cars even though it is an economy
how many high-valued vehicles to make rental. This planned-upgrade effect is par
available to lower booking classes (Table 1). ticularly pronounced on booking days clo
Fleet in the higher classes that is not re to the day of pickup. The model handles
quired for that class's demand enters the the change in availability by revising the
available pool. Classes with more demand on-rent allocation to allow for the number
than fleet extract fleet from the available of bookings from other classes it expects
pool of higher classes to cover their excess impact the current class.
demand requirements. The result is a set of Overbooking
allocations of the number of cars available The airlines have instituted a creative so
for rent in each booking class. The actual lution to the problem of oversales, which
Forecast on
Class Fleet Rent Demand Available Pool Excess Demand Available to Sell
Fullsize 4 door 230 165 65 165

Fullsize 2 door 50 60 10 60
Midsize 400 370 30 370
Economy 100 40 60 40
Subcompact 20 165 145 165
Table 1: This table illustrates a simple deterministic c
Available pool represents cars that are available to low
marginal revenue heuristic. Excess Demand represents
insufficient inventory to meet demand. The revenue m
system parameters that provide a throttle to the flow
the number of bookings National is willing to accept
Available to Sell by borrowing cars from the next high
GERAGHTY, JOHNSON
occur when more passengers show up than
I..
can be accommodated. When a flight is
overbooked, the airlines in effect hold an
auction, offering rewards to any passengers
willing to give up their seats and take a
later flight. Unfortunately, this is not possi
ble in the car-rental business: customers O -I-^-h?^^"-1-^-1-^-1
1992 1993 1994 1995
flow to our counters at different times, ex
pecting and receiving immediate service.
Figure 7: Overbooking compe
They would find it intolerable to be herded
lations and no-shows, which ar
into an enclosed area, forced to wait until a vance booking process. More
certain time, and then to take part in an booking policy results in fewer
the lot but increases the instan
auction to determine those willing to wait
tions-No Car." National's reco
for a vehicle until a later time. Thus, it is
ment of Res/No Car incidents l
critical that our forecast and planning be in 1992. Unfortunately, turndo
extremely accurate during peak periods and utilization was not optimal
GM's direction to turn the com
(Figure 7).
utilization improved, revenues
The overbooking model revises the re without the tools provided by
sults of the capacity management process Car incidents rose. When the
to account for the impact of no-shows and ers and field managers had the
limited locations) in July 1993
cancellations. It may be regarded as a map
not only able to continue to re
ping from demand space to reservations and improve utilization, but als
space where demand space represents ac vice problems. The graph illust
advances made in 1994 and 1995
tual materialized demand on the day of
corporate-wide implementation
pickup and reservations space is the num and department.
ber of bookings required, at a given num
ber of days prior to arrival, to achieve that discrepancies in checkout/re
demand level. The overbooking process day. (For a discussion of the
produces an adjusted, sometimes called of overbooking see Smith, L
overbooked, availability allocation for each and Darrow [1992].)
car class. The overbooking model identifies Pricing
optimal overbooking levels by balancing The car-rental customer population is
the expected cost of an oversale against the made up principally of two segments: cor
opportunity cost of an unrented car, subject porate and leisure. A corporate customer
to service-level constraints (appendix). Na generally books close to the date of rental,
tional identifies acceptable oversale risk is inflexible in rental and return dates and
levels at the corporate level. Individual lo times, does not shop competitors exten
cations set target utilization levels that ad sively, is unwilling to prepay, is not willing
just available capacity to compensate for to stay over a Saturday night, and most im
the resulting oversale rate and for such op portant, expects to pay an authorized fixed
erational issues as car turnaround time and rate, negotiated and reimbursed by his or
INTERFACES 27:1 120
NATIONAL CAR RENTAL
her company. The leisure customer, on the stimulate demand. An elasticity model re
other hand, is often willing to stay over a lates historic rate and demand variability.
Saturday night, books in advance, is some The discrepancy between the demand fore
what flexible on time of pickup, for exam cast and the target utilization indicates the
ple, willing to wait until noon on Thurs required change in booking pace. The elas
days to qualify for cheaper weekend rates, ticity model provides a rate adjustment that
is willing to prepay rental charges and will induce this change.
shops competitors extensively looking for The model relies on the revenue analyst
the best value. to provide an appropriate base rate for
The National pricing model links rate each individual product. The base-rate level
levels with availability and consumer book embodies the analyst's expertise about such
ing activity to achieve revenue and utiliza areas as the price tolerance of the market
tion objectives. Our initial analysis of Na segment and the degree of competition at
tional's rate behavior indicated that the location. When the analyst disagrees
competitive positioning was the determin with a system-generated rate, he or she will
ing factor in its pricing decisions prior to either reject it or override it in the process
the revenue management program. At of reviewing the recommendations. When
times of low demand, sensitivity to com this happens, the pricing model recalibrates
petitor behavior is crucial. Utilization levels
can suffer drastically from poor rate posi
tioning in the marketplace. During high A comparison of these rates
demand periods however, the firm can lose with the cost of renting a
large revenue opportunities by following tuxedo underscores their
competition-based pricing rules that cause
irrationality.
the firm to exhaust inventory by accepting
low-valued bookings. RMS implements a
demand-based pricing policy. We devel around a new base rate. The system infers
oped a simple, but extremely effective the new base rate by working backwards
pricing model to support this policy (ap from the actual rate the analyst sent to the
pendix). reservations system and the demand
The pricing model recommends in based rate offset recommended by the
creased or decreased rates based on on-rent elasticity model. The system maintains de
demand for a each arrival date. The rates mand and availability information at a
are designed to encourage maximum utili higher level of aggregation than price.
zation of rentable capacity. Therefore if re RMS-generated price recommendations
maining demand plus current on-rents ex cause groups of rates to move together.
ceeds rentable capacity, the model will This simplifies rate management by main
increase the rate to extract high-valued con taining consistent differentials. The analyst
sumers from the total demand. If remain adjusts rate differentials by overriding the
ing demand plus current on-rents is below current system recommendation. The re
rentable capacity, it will reduce the rates to computed base rate will maintain the new
GERAGHTY, JOHNSON
differential. The pricing model is more revenue. The pricing model supports rate
than just a price-management system that premium profiles that offset base-rate
mimics analyst behavior. By making fre levels by different amounts depending on
quent adjustments in response to market the number of days left before pickup (Fig
behavior, it extracts the maximum revenue ure 8).
potential from each market segment over The National RMS bases its rate recom
the course of the booking process. Gallego mendations on forecasts of demand. At
and van Ryzin [1994] suggest that one pos times when on-rent demand exceeds avail
sible reason for revenue management's ability, the pricing model extracts the most
success is its ability to capitalize on statis valuable customers from this mix by dis
tical fluctuations. couraging lower-valued demand. When
Traditional revenue management models competitors undercut National's price at
capitalize on a reduction in customers' times of high demand, they end up with
price sensitivity later in the booking pro the low-end customers and leave the
cess, primarily by using restrictions on ad higher-valued, later-booking customers for
vance purchases. This kind of market seg National. RMS allows National to break the
mentation has been difficult to implement "competition paradigm" by recommending
in the car rental industry, but rate premi when it can price above the competition
ums for late booking are an alternate and when it should price extremely
means of capturing at least some of this competitively.
$54
$45 Average Daily Rate
$28
0 15 30
Days before Pickup
Figure 8: Before RMS, National would set a base price, for example, $45 per day, and it would
remain in effect regardless of demand or how far in advance customers would book. Price
sensitive leisure customers would search for a lower rate. The price-insensitive late-booking
customer would have been willing to pay more, causing revenues to be diluted.
INTERFACES 27:1 122
NATIONAL CAR RENTAL
Reservations Inventory Control program to identify the length-of-rent cate

The capacity management and pricing gories on each arrival day that provide the
models work closely together to manage greatest revenue. (Williamson [1988] dis
on-rent demand. The availability alloca cusses optimization for reservation control.)
tions from the planned upgrades model To implement the LP recommendations,
provide a degree of reservations control but National would need a reservations system
do not provide adequate protection for con with the ability to switch availability on
strained days. The pricing model tries to and off for each possible length of rent on
compensate by adjusting rate levels to each arrival day. The existing reservations
achieve utilization targets. Both of these controls at National allow specification of
models are limited because they function at minimum length-of-rent controls for each
an aggregate demand level. By controlling arrival day. Therefore, the second phase of
length of rent, National can more precisely the reservation-control model degrades the
trade off between demand elements that full-pattern solution to a set of minimum
are competing for inventory (Figure 9). length-of-rent values and constrained arri
Demand forecasts for each length-of-rent val-day indicators. The degradation
category, revenue forecasts based on sys algorithm reconciles conflicting open/close
tem rate recommendations, and remaining recommendations within the full-pattern
on-rent capacity provide input for a mathe control by weighting each recommendation
matical programming model that generates by the amount of demand it impacts and
minimum length-of-rent restrictions for its proximity to a future constrained day.
each arrival day (appendix). The first phase For example, the reservation system pro
of the model solves a deterministic linear cesses a booking request for a four-day
Available
On Rent
Figure 9: National fits demand for each length-of-rent into remaining capacity using a determinis
tic linear program. The result is a set of allocation recommendations for each demand element.
The linear program's allocations are used to determine minimum length-of-rent (MLRs) controls
that can be implemented on the reservations system. The MLRs protect arrival days with con
strained capacity while building up utilization on shoulder days (arrival days with availability
adjacent to constrained days).
GERAGHTY, IOHNSON
rental arriving on Sunday. It tests availabil strategies to estimate lost demand. There
ity for all days the booking request re are two principal types of lost demand:
quires. Once this test is passed, the length turndowns are reservations requests that
of-rent specified for the arrival day is were rejected due to revenue management
considered. The reservation system has a controls, and shoppers are customer inquir
five-day minimum length-of-rent specified ies that do not result in reservations re
for Sunday rentals, but it does not automat quests. Turndowns are key indicators of
ically reject the booking. It needs to check if the effectiveness of capacity management
the booking request demands inventory and reservations control. Shoppers provide
from a constrained day. The reservation useful insight into the effectiveness of rate
system finds that Wednesday is a con levels. National's reservation system tracks
strained day so it rejects the booking. Next turndowns and shoppers at the transaction
the reservation system processes a two-day level. By combining the reservations that
booking request with arrival on Sunday. actually occurred with appropriate turn
Because this does not impact Wednesday's down and shopper transactions, it can com
inventory, the reservation system accepts pile an accurate reservations history. Anal
the booking. The constrained-day indicator ysis of the reservations history with perfect
functions as a maximum length-of-rent hindsight provides an estimate of the total
control. A booking must have a length-of revenue potential in the marketplace. Con
rent that fits between the arrival date and versely, National can estimate the revenue
the next constrained-day indicator, or else that would be realized in the absence of
have a length-of-rent greater than the speci revenue-management controls. We subtract
fied minimum. In this way, the reservation the revenue from the no-controls scenario
system protects constrained inventory from the actual revenue realized to get the
while building up utilization on the shoul
der days (days with inventory either side of
a constrained day).
Demand peaks for rental cars
The National RMS sets the standard for midweek.
revenue management in the car rental in
dustry. It is the first implementation of in revenue impact of revenue management at
tegrated capacity management, pricing, and National. National Car Rental also uses this
length-of-rent control. The length-of-rent process to evaluate the impact of individual
optimization, which uses revenue forecasts revenue-management controls. For exam
and detailed length-of-rent forecasts by ar ple, it assesses the impact of overbooking
rival date, is also unique. by comparing the results of the reservation
Impact of Revenue Management process with the historic overbooking levels
When it comes to evaluating revenue to the results of that process with inventory
management performance, National Car levels set at the fleet level.
Rental has an advantage over many of its The integration of the revenue manage
competitors. Traditional revenue opportu ment system and the revenue management
nity models use demand untruncation department have catalyzed change in the or
INTERFACES 27:1 124
NATIONAL CAR RENTAL
ganization. It has supported more flexible X?j = the decision variable for arrival day i
fleet strategies and tactics. It has led in the and length-of-rent category ;,
fjbij = mean remaining demand for length
development and implementation of new
of-rent category ;,
products (Saturday-night keep rates, pre Ci = remaining capacity for arrival day i,
paid rates, guaranteed rates). It has helped and
National to give customers a higher level of Yij = expected revenue for length-of-rent
service, especially our late-booking corpo category ; on arrival day i.
rate clients. It has upheld our ability to serve The LP solution is degraded to maximum
customers who suffer broken reservations and minimum length-of-rent recommenda
tions by a voting scheme that weights the
by our competitors. But most important, in LOR recommendations for each arrival day
its first year of coast-to-coast deployment, according to its proximity to a future con
the operational revenue management de strained day.
partment allowed the creation and realiza Planned Upgrades
tion of $56 million in incremental revenue. The planned upgrades algorithm is based
APPENDIX: Length-of-Rent Control on a heuristic that was developed for allo
Reservations are accepted or rejected cating airline reservations inventory on a
based on length-of-rent controls. Optimal single flight leg (see Belobaba [1989]). This
length-of-rent controls are determined by a EMSR (expected marginal seat revenue)
revised simplex algorithm with upper heuristic computes protection levels for
bounds for the following formulation: each booking class. A protection level is the
number of cars that should be reserved for
maximize the demand in the current class. The opti
N L mality conditions for a constrained revenue
2u 2* YijXij, maximization problem are as follows:
? - ? -XV '
subject to dir i diTj
i L
where
X X SkjXkj < Q
R is the revenue function,
fc=max(l,??L) ;=1
\ represents the expected marginal revenue
1 if k + j > i of the last car allocated to each class, and
?jq =0 otherwise iTj is the protection level for class /.
Once the EMSR heuristic determines pro
Xij ^ IMj, tection levels for each class, the planned
upgrades algorithm determines the avail
where ability in each class. The availability
N = the number of arrival days in the number that appears on the reservations
planning horizon, system is the sum of the availability for fu
L = the number of length-of-rent catego ture bookings and the current bookings for
ries, the car class. Availability for car classes
i = {i: 1,..., N] arrival days within the with excess demand is computed as the
planning horizon, fleet in that class plus any cars available
k = {k: 1,..., N] arrival days that impact from higher classes to cover the excess de
day i availability, mand. The availability for car classes with
j = {j: 1,..., L] length-of-rent categories, excess fleet is computed as the fleet less
GERAGHTY, JOHNSON
any cars used by lower car classes. Fleet solute price relationships across programs
not used by lower car classes is returned to and city stations. Therefore the percentage
the availability of the original car class. The price adjustments returned by this model
final avail-to-sell number is the sum of are converted to absolute dollar changes
availability, current bookings, and adjust based on a median rate for the inventory
ments for planned upgrades that are ex location.
pected to pick up a car in the next 24 Overbooking
hours. The overbooking model identifies opti
Pricing mal overbooking levels subject to service
The pricing model recommends rate level constraints. The optimal overbooking
changes in response to the bookings for level is the point at which the marginal over
each arrival date, relative to the expected sale cost is balanced against the marginal
booking pace. The magnitude of these rec increase in revenue due to overbooking.
ommended changes depends on the re A = authorization, that is, maximum ac
sponsiveness of demand to rate changes, or ceptable on-rent bookings,
the demand elasticity. The basic assump C = capacity,
tion behind the model is that the historical S = number of on-rent cars,
variance in demand is correlated with the p(S | A) = probability density function of
elasticity: the greater the variance, the more the on-rent demand for a given authoriza
responsive demand is to price changes. tion level,
The rate changes are limited to a maxi U = number of empty cars,
mum range; for example, between 80 per O = number of oversales,
cent and 120 percent of the PB, base price. OS-Cost = cost of an oversale, and
The slope of the rate-response function de Spoilage-Cost = opportunity cost of an
pends on a, the variance in demand. This empty car.
relationship can be expressed as an inverse The expected number of empty cars is
demand function of the following form: given by
APn
P = l.2PB--^Q,
2(7 E(U\A)= I (C - S)p(S\A)dS.
where P and Q represent price and de The expected number of oversales is given
mand, respectively, and \x represents the by
mean demand level. This in turn implies a
poo
demand elasticity given by
E(0\A) = J (S - C)p(S\A)dS.
s?P
The optimum expected revenue for authori
zation level occurs at the minimiim value of
where eD represents the own-price elasticity
of demand. It is clear that, holding other Spoilage Cost*E(U\ A) + OS_Cosf*E(0|A).
factors constant, an increase in a will in
crease the elasticity of demand and induce Since expected oversales increase and ex
smaller price adjustments for a given de pected unused cars decrease with respect to
sired change in demand. authorization, we can find a global mini
Expected booking pace is maintained at a mum by increasing authorization from ca
higher level of aggregation (inventory loca pacity until the value of this equation starts
tion) than prices, and there is an opera increasing. The authorization is constrained
tional requirement to maintain constant ab above by the following service-level re
INTERFACES 27:1 126
NATIONAL CAR RENTAL
quirement, which is set at the corporate Flight Transportation Laboratory, MIT,

level: Cambridge, Massachusetts.
i - I p(S | A)dS ^ Maximum probability

'o
of one or more oversales.
References
Belobaba, Peter P. 1989, "Application of a proba
bilistic decision model to airline seat inven
tory control," Operations Research, Vol. 37, No.
2 (March-April), pp. 183-196.
Cross, Robert G. 1986, "Strategic selling: Yield
management techniques to enhance revenue,"
presentation to the Shearson Lehman Broth
ers, Inc. 1986 Airline Industry Seminar, Key
Largo, Florida, February 14.
Curry, Renwick E. 1993, "Kaiman filtering and
exponential smoothing," presentation to Air
line Group International Federation Opera
tions Research Societies Reservations and
Yield Management Study Group, May, Syd
ney, Australia.
Curry, Renwick E. 1990, "Optimal airline seat al
location with fare classes nested by origins
and destinations," Transportation Science, Vol.
24, No. 3 (August), pp. 193-204.
Gallego, Guillermo and van Ryzin, Garret 1994,
"Optimal dynamic pricing of inventories with
stochastic demand over finite horizons," Man
agement Science, Vol. 40, No. 8 (August), pp.
999-1020.
Littlewood, Kenneth 1972, "Forecasting and con
trol of passenger bookings," Airline Group In
ternational Federation Operations Research
Societies Symposium Proc. 12, pp. 95-117.
Ramaekers, Lawrence 1995, "National Car
Rental Systems, Inc.," SCORECARD , The
Revenue Management Quarterly, first quarter,
pp. 2-3.
Smith, Barry C; Leimkuhler, John F.; and
Darrow, Ross M. 1992, "Yield management at
American Airlines," Interfaces, Vol. 22, No. 1
(January-February), pp. 8-31.
Weatherford, Lawrence R. and Bodily, S. E.
1992, "A taxonomy and research overview of
perishable-asset revenue management: Yield
management, overbooking and pricing," Oper
ations Research, Vol. 40, pp. 831-844.
Williamson, Elizabeth L. 1988, "Comparison of
optimization techniques for origin-destination
seat inventory control," Report FTL-R88-2,

Revenue Management Saves National Car Rental

Uploaded by

Copyright:

Available Formats

Revenue Management Saves National Car Rental

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Revenue Management Saves National Car Rental

Uploaded by

Copyright:

Available Formats

Revenue Management Saves National Car Rental

Author(s): M. K. Geraghty and Ernest Johnson

M. K. GERAGHTY 2188 Deep Woods Way

ERNEST JOHNSON National Car Rental

In 1993, National Car Rental faced liquidation. General Motors

Copyright ? 1997, Institute for Operations Research INDUSTRIES?TRANSPORTATION/SHIPPING

INTERFACES 27:1 January-February 1997 (pp. 107-127)

INTERFACES 27:1 108

National had three legacy systems to eroding customer satisfaction. Roll-out of

January-February 1997 109

INTERFACES 27:1 110

bookings, cancellations, turndowns, and decentralization would cause a number of

January-February 1997 111

fleet disbursement. by accepting or rejecting booking requests

that are often multi-leg, multi-night, or

enue managers with control of both pricing

INTERFACES 27:1 112

Revenue managers have a wealth of sup Forecasting

recommendations as part of the review number of cars in use on a specific date. It

January-February 1997 113

which bookings are expected to accumulate

INTERFACES 27:1 114

January-February 1997 115

distributed electronic reports. fleet made available at the last minute by

INTERFACES 27:1 116

$100 Unrealized Revenue Potential

Unrealized Revenue Potential

Unrealized Revenue Pot

Figure 5: If fences can be found that effectively segment the ma

INTERFACES 27:1 118

hierarchy interferes with a simple nesting planned-upgrades model takes account of

Fullsize 4 door 230 165 65 165

January-February 1997 119

occur when more passengers show up than

INTERFACES 27:1 120

January-February 1997 121

$45 Average Daily Rate

Days before Pickup

INTERFACES 27:1 122

Reservations Inventory Control program to identify the length-of-rent cate

January-February 1997 123

INTERFACES 27:1 124

January-February 1997 125

INTERFACES 27:1 126

quirement, which is set at the corporate Flight Transportation Laboratory, MIT,

i - I p(S | A)dS ^ Maximum probability

January-February 1997 127

You might also like