rm ahmedabad 2005

8/10/2019 RM Ahmedabad 2005

1/83

2005 by Peter C. Bellw1

Revenue Management

Vision 2020: Ahmedabad 2005

Peter C. BellWs [email protected]


2/83


My Objective

To introduce you to the practice and theory ofRevenue Management

Since this is a large and fast growing field, this will be a

broad-brush survey.


3/83


AGENDA:

Introduction What is revenue management, who uses it

and what has been the impact?

The five pillars of RM Pricing, discount allocation, overbooking, trading up

and re-planing

Integrating the tools Conclusions


4/83


AGENDA:




and re-planing

Integrating the tools Conclusions


5/83


REVENUE MANAGEMENT: Definition

Revenue management (RM) is the

science and art of enhancing firmrevenues while selling essentiallythesame amount of product.


6/83 2005 by Peter C. Bellw6

REVENUE MANAGEMENT: History

The first reference to RM is:

Taylor, C.J. (1962) The determination of passenger

booking levels. Proceedings of the Second AGIFORSSymposium, American Airlines, New York.

This work recognizes the value of selling moreairline seats than capacity in anticipation ofno-shows. We now call this overbooking.



REVENUE MANAGEMENT: History

The first major users ofRM wereAmerican Airlines and Delta Airlinesstarting about 1985.

Both Tom Cook (American) and Robert

Cross (Delta) have been cited as thefathers of corporate RM.



OR developments in revenue

enhancement since 1985 have ledto innovative new methods of pricing

and delivering products.

We call these methods revenuemanagement tools.



MANAGEMENT SCIENTISTS

HAVE WRECKED THE AIRLINEINDUSTRY

Joseph F. Coates(California futurist)



However the impact of revenue

management has been dramatic

and the use of RM continues toexpand to new products.


11/83


We estimate that yield management hasgenerated $1.4 billion in incremental revenue in

the last three yearsby

creating a pricing structure that responds todemand on a flight-by-flight basis

R.L. Crandall, Chairman and CEO of AMR, 1992


12/83


We have estimated that the yield managementsystem at American Airlines generates almost$1 billion in annual incremental revenue

Tom Cook, President

SABRE Decision Technologies,

June 1998


13/83


"Ford Motor Co. has quietly been enjoying ahuge surge in profitability... 1995 and 1999,

U.S. vehicle sales rose just 6 percent, from3.9 million units to 4.1 million units. Butrevenue was up 25 percent, and pretax

profits soared 250 percent, from about $3billion to $7.5 billion. Of that $4.5 billiongrowth, Ford's Lloyd Hansen, controllerfor global marketing and sales, estimatesthat about $3 billion came from a series ofrevenue management initiatives.

CFO Magazine, August, 2000


14/83


(revenue management) basically saved NationalCar Rental. And you can go from the CEO of

National on down, and they will all say: justapplying these OR models made the life or death

difference for this company

Kevin Geraghty, Aeronomics Inc.


15/83


and .. on the other side:

RM used by the competition has bankruptedseveral corporations .. the clearest examplebeing Peoples Express Airlines.


16/83


Donald Burr, Founder and CEO of PeopleExpress

believes that major carriers use ofsophisticated computer programsto immediately match or undercuthis prices ultimately killed PeopleExpress


17/83


WHO USES RM?

AIRLINES (All?)

HOTELS (Hyatt, Marriott, Hilton, Sheraton, Forte, Disney ..)

VACATIONS (Club Med, Princess Cruises, Norwegian ..)

CAR RENTAL (National, Hertz, Avis, Europcar ..)

WASHINGTON OPERA

FREIGHT (Sea-Land, Yellow Freight, Cons. Freightways ..)

TELEVISION ADS (CBC, ABC, NBC, TVNZ, Aus7 ..)

UPS, SNCF

RETAIL (Retek, Khimetrics) REAL ESTATE (Archstone)

NATURAL GAS

TEXAS CHILDRENS HOSPITAL


18/83


AGENDA:




and re-planing

Integrating the tools

Conclusions


19/83


THE 5 PILLARS OF RM

New approaches to pricing Discount allocation

Trading-up

Overbooking

Re-planing

+ markdown optimization, purchase loans, etc


20/83


NEW APPROACHES TO PRICINGPRODUCTS AND SERVICES


21/83


THE NEW PRICING CONCEPT

Use product price as a managementcontrol variable.

Set this price optimally to variouscustomer groups or clusters.

Be prepared to change prices often.


22/83


THE CONCEPTUAL LEAPS

Products do not have a value, rather thevalue of a product depends on the point intime of purchase.

Products have different values to differentclusters of customers.


23/83


DEMAND AT A POINT IN TIMEBY ACUSTOMER CLUSTERDRIVES RM PRICING

MARKET SEGMENTATION IS THE KEY TO

ENHANCING REVENUES THROUGH RMPRICING


24/83


Price

Quantity Sold

A MARKET

Q = F(P,..)


25/83


Price

Quantity Sold

SEGMENTING A MARKET

Market

Segmentation

Q = F(P,..)

becomes qi = Fi(pi,..)

for i = 1,N


26/83


MARKET SEGMENTATION: METHODS

Time of purchase

Customer characteristics (seniors, others)

Sales channel (clicks and bricks) Offer a discount to large customers

Offer a discount for slow delivery

+ +


27/83


Example of market segmentation

Coke price rises with heat

Vending machine that alters price with temperature

change is possible

NEW YORK (CNNfn) - It's not the New Coke, but it maybe the Smart Coke. Soft-drink giant Coca-Cola Co. isworking on a vending machine that automaticallyraises the price of a soda whenever the weather growshot.

Coke chairman Doug Ivester said the machine wasdesigned to reconcile supply and demand by raisingthe price when demand increased.

"Coca-Cola is a product whose utility varies frommoment to moment," he was quoted as saying.


28/83


CLUSTERING CUSTOMERS

Different groups of customers value a product

differently Example:

Business class air customers value

convenience, comfort, flexibility Economy class customers value low

prices, (and perhaps longer stays,

advance reservations) Clustering means assigning customers to

clusters or market segments


29/83


An example of clusters

E-Bivalent Newbies (5% of online shoppers) Newest to Net, somewhat older, spends least online, likes online

least Time Sensitive Materialists (17%)

Most interested in convenience, less likely to read reviews orcompare prices

Clicks and Mortar(23%) Browse online, prefers to buy offline, more likely female, has

privacy and security concerns, goes to malls often

Hooked, Online and Single (16%) More likely young, single males with high incomes, have beenon Net longest, most likely to play games, download, bankonline

Hunter-Gatherers (20%) Typically 30-49 years old, two kids, most likely to visit sites that

provide information and comparison Brand Loyalists (19%)

Most likely to bypass search engines to go directly to sites theyknow, most satisfied with shopping online, spend most online

-- Harris Interactive study of 3,000 Internet Shoppers, 2000

CHANGING PRODUCT VALUE


30/83


CHANGING PRODUCT VALUEOVER TIME (time segmentation)

Perishable products that age,

Seasonality,

Some customers will pay for the security ofearly purchase, or for the flexibility of latepurchase,

Suppliers may attach value to the security ofearly sales.


31/83


Demand

Time

CHANGING DEMAND OVER TIME

Example: PERISHABLE PRODUCTS

Segment 1 S2 S3


32/83


Demand

Time Event Date


Example:EVENT or TRIP TICKETS


33/83


Demand

Time Event Date


Example:EVENT or TRIP TICKETS


34/83


Common Demand Models

Q = Quantity sold

p = price/unit

Linear demand:

Q = A B p

Usually maximum and minimum prices are specified

Constant elasticity demand:

Q = A p-e

where e is the price elasticity of demand


35/83


Price

Quantity Sold

Demand Curve

Linear Approximation

LINEAR APPROXIMATION OF DEMANDCURVE


36/83


MODEL CHOICE MAY NOT BE OBVIOUS

Q = 92.1 - .37 P R2

= 0.95 Q= 2170037377 P-3.59

R2

= 0.95

DEMAND EQUATION ESTIMATION

0

10

20

30

40

50

60

70

$100.00 $150.00 $200.00 $250.00

Price

Qu

antity


37/83


General Demand Function

In general, for any time period and cluster:

Q = F(P, Ps, Pc, X1, X2, .)

where: P = price of product

Ps = price of substitute products

Pc = price of complement products Xi are exogenous variables (weather,

economic factors, advertisingetc)


38/83


Estimating Demand

Demand estimation is a craft: firms tend to keepthis part of their RM confidential.

Two steps are required:

1. Build a demand model (off-line). Commontechniques include regression, curve fitting,and cluster analysis.

2. Develop an on-line demand model updatingprocedure to respond immediately tounexpected observed demand.


39/83


LEAKAGE

Demand leakage (from a high pricedsegment to a low priced segment)

occurs when segmentation is notperfect.

Q1 = F1(p1) L(p1 p2)

Q2 = F2(p2) + L(p1 p2) where p1 > p2


40/83


FENCES

Revenue will disappear unless marketsegments are kept separate to limit leakage

from high priced segments to low pricedsegments.

Tools to maintain segment separation arecalled fences.

Examples of fences: The fee airlines charge to modify a low fare ticket

(usually $150-200).

The requirement for a Saturday night stopover for alow fare ticket.

Booking and paying 90 (or 60, or 30) days inadvance.

Look for examples of creative fences!


41/83


Pricing

Approaches to Revenue Maximization

Traditional fixed pricing

Variable pricing Optimum dynamic pricing

Computing optimum prices


42/83


Price

Quantity Sold

Single price: 2 market segments

p

Q1 Q2

Revenue = p (Q1 + Q2)

There is usually a p* thatoptimizes revenue.


43/83


Price

Quantity Sold

Two prices: 2 market segments

p1

Q1 Q2

p2

Revenue = p1Q1 + p2Q2

If p1* and p2* maximize revenue

then:

p1*Q1 + p2*Q2 p*(Q1 + Q2)

Formal statement of deterministic ODP


44/83


Formal statement of deterministic ODPproblem (time segments)

Let: pi be the product price in period i,

qi

be the quantity sold in period i,

qi= fi(pi) be the demand curve for period i,and

Ibe the inventory to be sold over Npricingperiods.

then:

Max

Subject to:

qi= fi(pi) for i = 1, ......N

i

N

ii qpZ

=

=1

Iq iN

i=

=1

Deterministic ODP problem with forecast


45/83


Deterministic ODP problem with forecasterrors

Let: pi be the product price in period i,

qi be the quantity sold in period i,

qi= fi(pi) be the demand curve for period i, and

Ibe the inventory to be sold over Npricing periods.

then:

For each period, k, k = 1,2,...N

Max

Subject to:

qi= fi(pi) + error fori = k, k+1, ......N

i

N

ki iqpZ ==

==

=1

1

k

i ii

N

ki

qIq


46/83


ACCEPT/REJECT DECISION MAKING

RM algorithms are mostly accept/reject rules.

If a customer appears and offers to buy a unitfor $P, do you accept (in which case you giveup the opportunity to sell the same unit to a

later arriving customer who may pay morethan $P), or do you reject (in which case yougive up $P in the hope of selling the unit later>$P but may not sell the unit).

The issue is one of balancing yield loss and

spoilage.


47/83


Revenue optimization drove the entry of RM intoservices.

- non-replenishable inventories,

- low (zero?) variable cost of providing product.

For restockable items, contribution optimizationcan be more difficult.

- role of inventory and replacement policies,

- need for profitable market share.

The Single Period Stochastic Optimum


48/83


The Single-Period Stochastic OptimumPricing Problem

Define: p - the price (a decision variable), c- the variable production cost (c


49/83


Inventory/SC models vs. RM models

Most inventory and supply chain modelsassume demand (Q) is given.

RM models replace given demand (Q) with a

given demand curve [F(p)]. The firm mustoptimizep (and hence determine Q) andsimultaneously optimize inventory.

Many research opportunities.


50/83



51/83


THE 5 PILLARS OF RM

Optimum dynamic pricing Discount allocation

Trading-up (or planned upgrades)

Overbooking

Re-planing (or short selling)

+ markdown optimization, purchase loans, etc


52/83


53/83


Discount allocation: Solver example


54/83


Discount allocation: hotel example

DYNAMICS OF MULTIPLE DISCOUNT


55/83


DYNAMICS OF MULTIPLE DISCOUNTPACKAGES AND PRICE CATEGORIES

Issue: If a customer appears and demands thelow fare or discount package, when do yousay No in order to preserve product for thehigher paying customers?

Example: How many full fare economy seats

on the plane?


56/83


Reservation rule (Littlewood 1972)

Issue: How many units (seats) to reserve forlow price sale?

Continue to sell discount product at time t until:

r

(1 - Pt) R or (1 - Pt)

r / R

Where: r = low price (marginal revenue)

R = high price (marginal revenue)Pt = probability of selling at leastthe remaining number of units.

EMSR Heuristic (Expected Marginal Seat


57/83


EMSR Heuristic (Expected Marginal SeatRevenue) Belobaba 1987

Apply Littlewoods rule sequentially to fare classes inincreasing fare order.

Let i ,i be the estimates of mean and std. deviation of demandfor product class i with price pi

Set a reservation (or protection) level of Li so that

pi+1 = Pi P(Xi > Li)

Where Pi is the weighted average fare for classes 1,2,,,,,I

And Xi is a normal random variable convoluting demand forclasses 1,2,.i


58/83


Protection levels: Expected MR curves

O b ki ( k lli )


59/83


Overbooking (aka overselling)

Issue: Should you take more orders than you

have product in the expectation that somecustomers who have ordered will not collectthe product?

If so, how many extra orders should you takeand when?

P{cancel order} declines as delivery dateapproaches

OVERBOOKING EXAMPLES


60/83


OVERBOOKING: EXAMPLES

airline seats, and passenger train seats, are usuallybooked in advance of the date of travel,

rental cars can be reserved ahead of the day of rental,

hotel rooms and campsite spaces, seats for stageshows, sports events, and concerts are sold in

advance, fresh turkeys can be ordered for delivery at

thanksgiving,

package vacations and cruises are usually booked in

advance, tuxedos, cut flowers, some baked goods, etc. are

commonly booked for future delivery

M i th b k d t


61/83


Managing the overbooked customer

A key cost in all the models

Airlines offer cash or travel coupons to

ticketed customers in order to persuade themto take an alternate flight when space isneeded for overbooked passengers.

Hotels will trade up overbooked customers torooms on the executive floor,

Rental car companies will substitute a higher

class of car at no extra charge.

In all these cases, the cost is well know.

Mi i i i th h t


62/83


Minimizing the no show rate

If all customers with reservations showed up,there would be no benefit from overbooking

Require payment at the time the booking is made,

perhaps offering an early payment price reduction. Require payment in full at some prearranged time in

advance of the delivery date. If payment is notreceived, the supplier cancels the booking.

Fees may be charged if a booking is changed.

O b ki th b i d l


63/83


Overbooking: the basic model

M = the amount of product

B = the booking limit (B M)

RN = the value of sale of unit N with RN+1 RN,

Ci = cost of satisfying the ith overbooked customer if no product isavailable, Ci+1 Ci

P[Q|B] = probability that Q customers will show up if we sell B units.

The expected cost of unsold product is:

{(RQ+1 + RQ+2 +...+ RM) P[Q|B] } summed over all values of Q < M

The expected cost of handling oversold customers is:

{(C1 + C2 +...+ CQ-M) P[Q|B] } summed over all values of Q > M

Find B that minimizes the sum of these two expected costs.

O b ki D t i i th ti l l l


64/83


Overbooking: Determining the optimal level

T di ( l d d )


65/83


Trading up (planned upgrades)

Issue: If a customer appears demanding aproduct that is sold out, should you trade thecustomer up to a higher valued product (at

your expense)?Issue B. If you adopt this as policy, what does

this do to your inventory management?

Re planing


66/83


Re-planing

Issue: If a plane is sold out several days beforeflight date and a customer appears preparedto pay a high price for that flight, can you

profitably incentivize a low-fare paidcustomer to change flights?

2001 McKinsey, CALEB Technologies,American Airlines

Other RM tools


67/83


Other RM tools

Markdown optimization is common in retail(Retek, Manugistics, Spotlight): This is really

price optimization with non-increasing prices. The benefits are claimed to be very

substantial

Bottleneck optimization (Maxager Tech.) isprice optimization where inventory isproduction time on the scarce production unit.

There are others but all seem to be variationson the 5 basic tools.

AGENDA:


68/83


AGENDA:

Introduction

What is revenue management, who uses itand what has been the impact?

The five pillars of RM Pricing, discount allocation, overbooking, trading upand re-planing


Conclusions

Integration


69/83


Integration

Although these tools are almost always

discussed separately, they all function withina highly integrated system


70/83


PRICES

RESERVATION LEVELS


71/83


PRICES

RESERVATION LEVELS

CAPACITY


72/83


PRICES

RESERVATION LEVELS

CAPACITYOVERBOOKING LEVEL


73/83


PRICES

RESERVATION LEVELS


PLANNED UPGRADES


74/83


PRICES

RESERVATION LEVELS


PLANNED UPGRADES

REPLANE

LEVEL


75/83


PRICES

RESERVATION LEVELS


PLANNED UPGRADES

REPLANE

LEVEL


76/83


Product SalesHistorical Data

Build

Demand Model

Update

Database

Revise Demand

Model

PricingSystem

Real time

Sales Data

Current

Demand

Forecasts

Revenue Management

System

Inventory

Levels

Off-line Support Activities

The

Market

Posted Prices

REQUIREMENTS FOR SUCCESS


77/83


REQUIREMENTS FOR SUCCESS

SUPERIOR INFORMATION TECHNOLOGY,

SUPERIOR OPERATIONS RESEARCH SKILLS,

and

THE ABILITY TO MANAGE DYNAMIC PRICES.

AGENDA:


78/83


AGENDA:

Introduction

What is revenue management, who uses itand what has been the impact?

The five pillars of RM

Pricing, discount allocation, overbooking, trading upand re-planing


Conclusions

BUSINESS OPPORTUNITIES


79/83


BUSINESS OPPORTUNITIES

A great number of products seem ripe for RM.

Some examples:

Cinemas, golf courses, electricity, taxis, publictransit, newspapers and magazines,advertising, fashion clothing, blue jeans,vegetables, fast food, supermarkets, sports

events, theatre, pop concerts,

RESEARCH OPPORTUNITIES


80/83


RESEARCH OPPORTUNITIES

Academic

Almost every inventory model can be extended

to include a demand curve (some alreadyhave been!)

Accept/reject decision heuristics (Bayesian)

ApplicationThe merger of SC optimization and RM (EPO)

Clustering

New and improved fencing

Demand modeling (how good does it need tobe?)

CONCLUSIONS


81/83


CONCLUSIONS

RM tools have become very important for bothbusiness and OR

There are opportunities for many new revenuemaximizing models and pricing/inventorymodels

Heuristics to aid implementation are a priority

Competitive models provide an opportunity

The competitive customerCompetition among RM firms

CONCLUSIONS


82/83


CONCLUSIONS

RM techniques raise many business issues:winning firms will be able to implement theseideas while keeping customers (and

regulators) happy.Everyone gains from the efficiencies that RM

produces, but some individuals lose.Successful implementation usually requirestaking good care of the few who lose.


83/83

rm ahmedabad 2005

Documents