Capacity-Based

Revenue Management
Necati Tereyagoglu
Scheller College of Business
Phone: (404) 894-9812
Email: [email protected]
Price Segmentation

 A strategy aimed at differentiating customers in order

to charge different prices to different customers based
primarily on differences in willingness to pay.
Pure Customer Segmentation...

 ... is hard to find in a pure dynamic price setting.

 Difficult to implement
 Unpopular with consumers
 Sometimes illegal (discrimination in the legal sense!)

 Exceptions
 Student and senior citizen discounts
 Group membership discounts (AAA discount, AARP,
Conference rate)
 Geographical
 Zone Pricing
 Couponing
Product Based Segmentation Strategies

 Create an “inferior product” to sell to more price-

sensitive customers.

 Early booking restricted versus unrestricted

 Branded gasoline versus “generic”
 National brand versus store label versus generic
An Extendable Strategy
Airline Ticket Products

Full Fare Products Full Price Unrestricted

Leisure Products Some Restrictions

Deep Discount Products More Restricted

Priceline/Hotwire Carrier unknown,

(Opaque Products) Departure time unknown
When Does it Work?

 Assume that we are currently offering a single product at

price ph, with associated demand dh and unit cost c. We are
considering offering an inferior version with the same unit
cost. The new product will have a price pL < ph and induce
demand dL but reduce standard product demand by δ.

The new product is a winner only if:

dL(pL – c) > δ (ph – pL)

Induced Cannibalized
Margin Margin
Example

 Glen blarney Scotch: $12/fifth to produce. Sells

100,000 units per year at $45/fifth.

 Create Old Overshoe and sell for $22/fifth? Expectation

to sell 80,000 additional units, but cannibalize 20,000.

 Induced Margin: 80,000 x ($22 - $12) = $800,000

 Cannibalized Margin: 20,000 x ($45 - $22) = $460,000
Net Impact = $340,000
 A profitable strategy assuming that cannibalization
can be controlled.
Product Based Segmentation Strategies

 Geographical
 Zone Prices: NY versus Midwest versus CA
 International Pricing
 Computer Chips
 “The Big Mac” Index
 Student textbooks
 Neighborhood pricing for services
Product Based Segmentation Strategies

 Time Based Segmentation

 Time of Purchase
 Airlines, Hotels, ...

 Delivery Time
Problem with Segmenting Customers

Customers do not show up with their

willingness-to-pay stamped on their foreheads

$?
Capacity-Based Revenue Management

Capacity-Based Revenue Management

algorithms provide accept/reject advice

If a customer arrives and offers to buy your

product at price $P, do you accept (and lose the
opportunity to possibly sell the product to a latter
arriving customer at a higher price) or do you
reject (in which case you may have to sell the
product a price < $P or not at all)
Capacity-Based Rev Mgmt Tools

 Overbooking

 Booking Limits
Overbooking

 Overbooking becomes necessity if …

 … order cancellations occur frequently,
 … and the good is perishable.

 Level of overbooking is set based on the trade-off

between…
 the cost of wasting the good if too many cancellations and
 the cost of arranging a backup if too few cancellations

Seinfeld Overbooking
Ugly reality: Cancellations and no-shows

 Approximately 50% of reservations get cancelled at

some point in time.

 Solution: Sell more seats (rooms, cars) than capacity.

 Danger: Some customers may have to be denied a

seat even though they have a confirmed reservation.
 Practical problem
 The Hyatt Regency in Atlanta.

 100 King/Queen rooms.

 Hyatt offers a r = $150 fare that is fully refundable.

 Demand for rooms is abundant and generally

exceeds capacity.
 The Hyatt’s Problem
 The forecast for the number of customers that do not
show up ( X ) is Normal with mean = 8 and standard
deviation = 2

 The cost of denying a room to the customer with a

confirmed reservation is $175 in penalties. (Room rate
is $150 + $25 meal voucher)

 How many rooms ( Y ) should be overbooked (sold in

excess of capacity)?
Hyatt’s Overbooking Problem

 How many rooms (Y) should be overbooked (sold in

excess of capacity)?

a) 7
b) 8
c) 9
d) 10
e) >10
 Overbooking Cost Graph

Total cost

Spoilage
(overage)

Cost

Spill
(stockout)
0 A lot
Number of Overbooks
 Probabilities for overbooking by
the mean + 2 standard deviation
Number of customers that do not show up ( X ) is
Normal with mean = 8 and standard deviation = 2

-3SD -2SD -1SD 8 + 1SD +2SD +3SD -3SD -2SD -1SD 8 + 1SD +2SD +3SD

No of customers who no show <= 10 No of customers who no show > 10

 Expressing z Values for Normal Dist

Probability

D +1 SD +2 SD +3 SD

Cumulative normal distribution from left side of distribution (X + Z)

z .0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0.0 .5000 .5398 .5793 .6179 .6554 .6915 .7257 .7580 .7881 .8159
1.0 .8413 .8643 .8849 .9032 .9192 .9332 .9452 .9554 .9641 .9713
2.0 .9773 .9821 .9861 .9893 .9918 .9938 .9953 .9965 .9974 .9981
3.0 .9987 .9990 .9930 .9995 .9997 .9998 .9998 .9999 .9999 .9999
 Overbooking solution
 Underage penalty:
 if X > Y then we could have sold X-Y more rooms…
 … we could have sold those rooms at Cu = r = 150.

 Overage penalty:
 if X < Y then we bumped Y - X customers …
 … and incur an overage penalty Co = $175 - $150 = $25 on each
bumped customer.
𝐶𝑢
 Optimal overbooking level: 𝐹 𝑌 =
𝐶𝑢 +𝐶𝑜
𝐶𝑢 150
 Critical ratio: = = 0.86
𝐶𝑢 +𝐶𝑜 25+150
Optimal overbooking level
Number of customers that do not show up ( X ) is
Normal with mean = 8 and standard deviation = 2

𝐶𝑢 150
= = 0.86
𝐶𝑢 + 𝐶𝑜 25 + 150
z .0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0.0 .5000 .5398 .5793 .6179 .6554 .6915 .7257 .7580 .7881 .8159
1.0 .8413 .8643 .8849 .9032 .9192 .9332 .9452 .9554 .9641 .9713
2.0 .9773 .9821 .9861 .9893 .9918 .9938 .9953 .9965 .9974 .9981
3.0 .9987 .9990 .9930 .9995 .9997 .9998 .9998 .9999 .9999 .9999

 How many rooms should the Hyatt overbook?

Optimal overbooking level
Number of customers that do not show up ( X ) is
Normal with mean = 8 and standard deviation = 2
𝐶𝑢 150
= = 0.86
𝐶𝑢 + 𝐶𝑜 25 + 150
z .0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0.0 .5000 .5398 .5793 .6179 .6554 .6915 .7257 .7580 .7881 .8159
1.0 .8413 .8643 .8849 .9032 .9192 .9332 .9452 .9554 .9641 .9713
2.0 .9773 .9821 .9861 .9893 .9918 .9938 .9953 .9965 .9974 .9981
3.0 .9987 .9990 .9930 .9995 .9997 .9998 .9998 .9999 .9999 .9999
 Optimal number of overbooked rooms is Y= 8 + 1.1*2 = 10.2 = 10.
 The Hyatt should allow up to 100 + 10 reservations.
 There is about a 86% chance that the Hyatt will find itself turning down
travelers with reservations.
The Newsvendor Model and Overbooking
 A single decision is made before uncertain demand is
realized.

 There is an overage penalty:

 Overage penalty if X < Y (too many seats overbooked).

 There is an underage penalty:

 Underage penalty if X > Y (insufficient number of seats
overbooked).

 Choose Y to balance the overage and underage

penalties.
Actual Overbooking Cost Curve

Loss of revenue
Revenue from
from unhappy
regular bookings
customers
linear decline

$
non-linear decline

0 20 40 60 80 100 120 140

Percentage of Capacity Claimed
Dynamic Overbooking
Overbooking

Time to Event
Event Occurs Reservations Start
 Cost for being over/under demand
 Underage Cost
 Airline: lost fare (but which fare?)
 Hotel/Casino: lost room rate + incidental profits

 Overage Cost
 Compensation ( often free future ticket or stay )
 Provision Cost ( meals, drinks, gifts )
 Reaccomodation ( sometimes list price at competitor )
 Goodwill ( ways to reduce this? )

 Applications
 Obvious: Airlines, Hotels, Car Rental, Restaurants
 Less Obvious: Sporting Venues, Manufacturing, Professional
Services
Booking Limits

 Setting booking limits is valuable if…

 Customers can be segmented based on their w.t.p
 Low value customers purchase before the high value
customers.

 Level of booking limit is set based on the trade-off

between…
 … the cost of not selling the capacity at all vs.
 the cost of selling capacity to low w.t.p customer and
missing the chance to serve a high w.t.p customer
 The Hyatt’s New Problem
 Hyatt offers a rL= $150 (low rate) discount rate for a mid-
week stay targeting leisure travelers.
 Regular fare is rH= $400 (high rate) targeting business
travelers.
 Demand for low rate rooms is abundant (> capacity).
 Let D be uncertain demand for high rate rooms.
 D has a Normal distribution with mean = 30 and standard dev = 3.
 Assume most of the high rate (business) demand occurs
only within a few days of the actual stay.
 Objective:
 Maximize expected revenues by controlling the number of low rate
rooms you sell.
 Booking limit decisions
 The booking limit is the number of rooms you are willing to sell in a
fare class or lower.
 The protection level is the number of rooms you reserve for a fare
class or higher.
 Let Q be the protection level for the high rate rooms.
 Q is in effect while you sell low rate rooms.
 Since there are only two rate classes, the booking limit on the low rate
class is 100 – Q:
 You will sell no more than 100-Q low rate rooms because you are
protecting (or reserving) Q rooms for high rate customers.
0 100

Sell no more than the low Q rooms protected for

rate booking limit, 100 - Q high rate customers
Hyatt’s Protection Level Problem

 How many rooms ( Q ) should be saved for the

higher fare customers?
a) 27
b) 29
c) 30
d) 31
e) 33
 The Newsvendor Model and Booking Limits
 A single decision is made before uncertain demand is
realized.

 There is an overage penalty:

 If D < Q then you protected too many rooms (you over protected) ...
 … so some rooms are empty which could have been sold to a low
fare traveler.

 There is an underage penalty:

 If D > Q then you protected too few rooms (you under protected) …
 … so some rooms could have been sold at the high fare instead of
the low fare.

 Choose Q to balance the overage and underage penalties.

 Optimal booking limits
 Overage penalty:
 If D < Q we protected too many rooms and earn nothing on Q - D
rooms while incurring Co = rL penalty on each room we did not sell.
 Underage penalty:
 If D > Q we protected too few rooms and miss earning Cu = rH - rL in
revenue on each rooms that we could have sold at a higher price.
 Optimal high fare protection level:
∗
𝐶𝑢 𝑟𝐻 − 𝑟𝐿
𝐹 𝑄 = =
𝐶𝑢 + 𝐶𝑜 𝑟𝐻

 Optimal low fare booking limit = 100 – Q*

 Choosing the optimal high fare protection level is a

Newsvendor problem with properly chosen underage and
overage penalties.
Optimal protection level

Forecasted number of high fare customers ( X ) is

Normal with mean = 30 and standard deviation = 3
𝐶𝑢 𝑟𝐻 − 𝑟𝐿 400 − 150 250
= = = = 0.625
𝐶𝑢 + 𝐶𝑜 𝑟𝐻 400 400

z .0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0.0 .5000 .5398 .5793 .6179 .6554 .6915 .7257 .7580 .7881 .8159
1.0 .8413 .8643 .8849 .9032 .9192 .9332 .9452 .9554 .9641 .9713
2.0 .9773 .9821 .9861 .9893 .9918 .9938 .9953 .9965 .9974 .9981
3.0 .9987 .9990 .9930 .9995 .9997 .9998 .9998 .9999 .9999 .9999

 How many rooms should the Hyatt protect for

the high-fare customers?
Optimal protection level

Forecasted number of high fare customers ( X ) is

Normal with mean = 30 and standard deviation = 3
𝐶𝑢 𝑟𝐻 − 𝑟𝐿 400 − 150 250
= = = = 0.625
𝐶𝑢 + 𝐶𝑜 𝑟𝐻 400 400

z .0 .1 .2 .3 .4 .5 .6 .7 .8 .9
0.0 .5000 .5398 .5793 .6179 .6554 .6915 .7257 .7580 .7881 .8159
1.0 .8413 .8643 .8849 .9032 .9192 .9332 .9452 .9554 .9641 .9713
2.0 .9773 .9821 .9861 .9893 .9918 .9938 .9953 .9965 .9974 .9981
3.0 .9987 .9990 .9930 .9995 .9997 .9998 .9998 .9999 .9999 .9999
Optimal number of rooms to protect is Q= 30 + 0.3*3 = 31 rooms.
Hyatt should only sell 100 – 31 = 69 low-fare rooms.
There is about a 62% chance that the Hyatt will end up with at least
one empty room.
 Airline Game

• How many seats should be saved for the higher fare

customers?
What if there are many fares?
 Nested buckets:
Bucket 0: highest fare
Bucket 1
Bucket 2
Bucket 3

Bucket 4:
lowest fare
Key Takeaways

 Application of Newsvendor Model in Capacity Based

RM Strategies:
 Overbooking Levels
 Booking Limits
Next Week

 EMSR Heuristics and Bid Price Controls