A Decision Analytic Approach to Revenue Management Robert L. Phillips Nomis Solutions Stanford University March 13, 2006 Copyright Robert L. Phillips. 2006. All Rights Reserved. Copyright Robert L. Phillips 2006. All Rights Reserved Agenda • Introduction to Revenue Management • Elements of Revenue Management • Capacity Control • Overbooking • Network Management • An RM System Example Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 2 What is Revenue Management? The allocation of fixed capacity to different customer classes with different fares in order to maximize profitability. A special case of Pricing and Revenue Optimization, applicable in situations with: • Fixed and perishable capacity (and therefore opportunity costs) • Advanced bookings • Fixed fare classes • Uncertain demand and customer behavior (no-show, cancellation) Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 3 Basic Revenue Management Business Assumptions • Fixed, immediately perishable capacity -- airline seat, hotel room night, gas pipeline capacity, etc. • Units of capacity are identical - “a seat is a seat” • Reservations (bookings) for future capacity accepted prior to its use • Marginal costs per unit sale are fixed and small relative to price per unit • A finite number of fixed prices are set ahead of time. • The seller can control the number of units that he will provide for sale at each price at each time before departure. • The seller has opportunities to change availabilities at intervals prior to departure based on bookings received. • The goal of the seller is to control availability by fare class in each time before departure in order to maximize revenue (= contribution) This is the “basic revenue management” business problem. Each of the assumptions can be relaxed to create a variation. Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 4 History of Revenue (Yield) Management • Early 60’s - Overbooking Analysis, News-vendor Problem • Late 70’s - Airline deregulation • 1981 - Rise of Peoples Express (Discount Airline) • 1982 - Adoption of Controlled Super-Saver fares (American Airlines) • 1983 - 1990 Development of first leg-based Airline Revenue Management Systems by major airlines. Development of Commercial Revenue Management Systems (Aeronomics, DFI, PROS, SABRE) • 1985 - 1990 Development of early hotel (Marriott, Hyatt) Systems. Commercial hotel RM systems (Aeronomics, DFI, OPUS) • 1989 First O&D Airline Revenue Management System (SAS) • 1990 First Rental Car System (Hertz) • 1990’s E-Commerce, distribution control, lifetime customer value issues • 1995 - today. Pricing and Revenue Optimization Systems (Talus, Manugistics, Khimetrics, ProfitLogic, …) Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 5 Revenue Management Industries Industry Passenger Airline Hotel Rental Car Broadcasting Events Cruise Line Concert/Sporting Event Air Freight Capacity Unit Seat Capacity Types 1-3 Room Night Rental Day Time unit Seat Berth Seat 1-5 3 - 10 many 1 - 10 1 - 15 1 - 10 Capacity Fixed? Network Largely Origin and Destination Yes Length of Stay No Length of Rental Yes Commerical Block Yes None Yes None Yes None Weight, Volume 1-3 Largely Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 6 Origin and Destination The Basic Revenue Management Question A supplier (airline, hotel, made-to-order manufacturer) takes reservation for some stock of fixed capacity. A customer is “on the phone” requesting a particular fare. Do we say “yes” and sell him at the requested fare or do we say “no”? Why we would say yes: • To get his revenue Why we might say no: • Because we don't have sufficient capacity to accommodate him • Because he might displace a future, more profitable booking Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 7 Revenue Management Elements • Capacity Control: How to allocate limited capacity to different classes of customer? • Overbooking: How many total bookings to accept? • Network Management: How to manage bookings across a complex service network? • Additional Topics: Customer value management; group management; integration with pricing, scheduling, etc We will discuss capacity control, overbooking, and network management. Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 8 Agenda • Introduction to Revenue Management • Elements of Revenue Management • Capacity Control • Overbooking • Network Management • An RM System Example Copyright Robert L. Phillips. 2006. All Rights Reserved. Property of Nomis Solutions Inc. - Confidential Material Page 9 Basic Airline Segmentation Leisure Travelers •Price Sensitive •Book Early •Schedule Insensitive fd = Discount Fare Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 10 Property of Nomis Solutions Inc. - Confidential Material Business Travelers •Price Insensitive •Book Later •Schedule Sensitive ff = Full fare Two-Class Capacity Management Problem • Fixed capacity C • Two fare classes (full-fare and discount) with fares ff > fd > 0. Marginal costs are 0. • Discount fares book first. All seats not sold at discount are available for sale at full fare. • No cancellations or no-shows. • The demands at each fare are random variables, dd and df.. • Ff (x) = Probability that df < x. How many seats should we save for late-booking full-fare customers? Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 11 Property of Nomis Solutions Inc. - Confidential Material Capacity Control Problem Tradeoffs • Cannibalization - Seats were sold at fd, but some full-fare customers were turned away due to lack of seats. Cost = fd - ff for each full-fare customer turned away. • Spoilage - Discount passengers were turned away but the plane left with empty seats. Cost = fd for each “spoiled” seat. How do we set b -- the first period booking limit -- to optimally balance cannibalization and spoilage and maximize expected total revenue? Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 12 Property of Nomis Solutions Inc. - Confidential Material Capacity Control Problem: Marginal Analysis Relative Impact dd < b 0 Fd(b) df > C - b b b+1 1- Ff(C-b) 1-Fd(b) fd - ff dd > b Ff(C-b) Hold b Constant Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 13 Property of Nomis Solutions Inc. - Confidential Material df < C - b fd 0 Optimality Condition for Two-Class Problem The optimal booking limit b* solves Ff(C-b*) = 1 - fd/ff . This is known as Littlewood’s Rule. Littlewood's Rule is a simple variation on the standard critical fractile solution to the newsvendor problem. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 14 Property of Nomis Solutions Inc. - Confidential Material Two ways to implement Littlewood's rule • Set booking limit b* and hold • Use Littlewood's rule as the basis a dynamic decision rule: Accept discount bookings as long as fd > [1-Ff(C-xd)]ff, where xd is the number of discount bookings already accepted. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 15 Property of Nomis Solutions Inc. - Confidential Material Interpreting Littlewood's Rule Acceptance Criterion: fd > [1-Ff(C-xd)]ff This is expected opportunity cost! 1 Expected Opportunity Cost $ fd 0 0 10 Accepted Bookings Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 16 Property of Nomis Solutions Inc. - Confidential Material Net revenue as a function of booking limit $15,000 Revenue $14,000 $13,000 Expected DB Cost $12,000 Expected Net Revenue $11,000 $10,000 $9,000 100 105 110 Booking Limit (b) Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 17 Property of Nomis Solutions Inc. - Confidential Material 115 120 Standard Structure for Multi-Class Problem First Booking Period: Departure … n n-1 3 2 1 Fare: fn fn-1 f3 f2 f1 Bookings: xn xn-1 x3 x2 x1 Low Fare Bookings High Fare Bookings The basic assumption -- bookings occur in order of fare, that is: low to high. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 18 Property of Nomis Solutions Inc. - Confidential Material Time Nesting -- a Three-Class Example x2 = seats reserved for 2 and 3 x1 = seats reserved for 3 Aircraft Seating Capacity b1 = Booking Limit for 1 b2 = Booking Limit for 2 We assume three classes 1,2, and 3, booking in order, with f1 < f 2 < f3 Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 19 Property of Nomis Solutions Inc. - Confidential Material Capacity Control with three fare classes Relative Impact d3 < b3 0 F3(b3) Displace Class 1 Booking b3 b3+1 p3 – p1 q1 1-F3(b3) Displace Class 2 Booking dd > b3 q2 p3 – p2 1-q1-q2 No Displacement Hold b3 Constant Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 20 Property of Nomis Solutions Inc. - Confidential Material p3 0 Solving the multi-class problem • In general, the capacity allocation problem with more than two classes does not have a closed-form solution. • Two solution alternatives: • • Solve by dynamic programming -- generally too computationally intensive. EMSR heuristics -- formulate as a series of two-class problems and approximate the solution. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 21 Property of Nomis Solutions Inc. - Confidential Material Agenda • Introduction to Revenue Management • Elements of Revenue Management • Capacity Control • Overbooking • Network Management • An RM System Example Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 22 Property of Nomis Solutions Inc. - Confidential Material Overbooking Airlines and other industries historically allowed passengers to cancel or noshow without penalty. Airlines book more passengers than their capacity in order to hedge against this uncovered call, Airlines need to balance two risks when overbooking: Spoilage: Seats leave empty when a booking request was received. Lose a potential fare. Denied Boarding Risk: Accepting an additional booking leads to an additional denied-boarding. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 23 Property of Nomis Solutions Inc. - Confidential Material Denied Boarding Cost Consists of four elements: 1. Provision Cost of meals or lodging provided 2. Reaccom Cost of putting a bumped passenger on another flight 3. Direct Cost of direct compensation to the passenger -- usually a discount certificate for future travel 4. Ill-will Cost for involuntary denied boarding. Voluntary denied boardings have a higher cost than involuntary denied boardings. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 24 Property of Nomis Solutions Inc. - Confidential Material Denied Boarding Rates 1993 1997 2000 Voluntary 15 21 20 Involuntary 1 1 1 Total 16 22 21 Denied Boarding Rates per 100,000 Boardings Source: US DOT. Large US domestic Carriers Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 25 Property of Nomis Solutions Inc. - Confidential Material Marginal Analysis: Overbooking Relative Impact s does not increase (1-p) (s|b) > C D>b F(b) 0 f-d p s s+1 b b+1 (s|b) < C D<b Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 26 Property of Nomis Solutions Inc. - Confidential Material f 0 Overbooking Problem Solution We want to find the smallest b* such that: F(b*)p[Pr{(s|b*)>C}(f-d) + Pr{(s|b*)<C}f] = 0 or: Pr{(s|b*) > C}(f-d) +[1- Pr{(s|b*) > C}]f = 0 Pr{(s|b*) > C} = f/d Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 27 Property of Nomis Solutions Inc. - Confidential Material Overbooking Revenue C Revenue $11,000 b* Passenger Revenue $10,000 Overbooking Cost $9,000 Net Revenue $8,000 90 100 Booking Limit Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 28 Property of Nomis Solutions Inc. - Confidential Material 110 120 Overbooking Dynamics Bookings Booking Limit No-show “Pad” Capacity Bookings A Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 29 Property of Nomis Solutions Inc. - Confidential Material B Departure Time Agenda • Introduction to Revenue Management • Elements of Revenue Management • Capacity Control • Overbooking • Network Management • An RM System Example Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 30 Property of Nomis Solutions Inc. - Confidential Material Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 31 Property of Nomis Solutions Inc. - Confidential Material Hotel Network ... Resources: Monday Products: Tuesday Wednesday Thursday Friday Friday 1-Night Monday 3-Night Stay Tuesday 2-Night Stay Wednesday 3-Night Stay ... Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 32 Property of Nomis Solutions Inc. - Confidential Material Passenger Train Capacity and Load July 2002 9,000 Final Bookings 8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 Riders Aboard Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 33 Property of Nomis Solutions Inc. - Confidential Material Total Seats CHI OMA DEN GSC SLC RNO SAC SFO 0 Hotel Example Unconstrained Occupancy Capacity Su M T W Day of Week Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 34 Property of Nomis Solutions Inc. - Confidential Material Th F Sa Why the Problem is Hard Flight 1 San Francisco (SFO) Flight 2 Denver (DIA) St. Louis (STL) SFO – STL Fare = $400 SFO – DIA Fare = $200 DIA – STL Fare = $250 Which passengers we want to accept depends upon expected demands for all products. Sometimes we prefer SFO-STL pax, sometimes we prefer SFO-DIA or DIA-STL pax. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 35 Property of Nomis Solutions Inc. - Confidential Material When the Problem Really gets Interesting... Flight 1 San Francisco (SFO) Market Flight 2 Denver (DIA) St. Louis (STL) Y-Class M-Class B-Class G-Class SFO-STL $600 $400 $300 $250 SFO-DIA $280 $200 $150 $140 DIA-STL $350 $250 $180 $110 Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 36 Property of Nomis Solutions Inc. - Confidential Material Bid Pricing Set a bid price equal to the opportunity cost (λ) on each leg. Flight 1 Flight 2 BP = $230 BP = $190 San Francisco (SFO) Market Denver (DIA) St. Louis (STL) Y-Class M-Class B-Class G-Class SFO-STL $600 $400 $350 $250 SFO-DIA $280 $200 $150 $140 DIA-STL $350 $250 $180 $110 Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 37 Property of Nomis Solutions Inc. - Confidential Material Hotel Bid Price Calendar Mon 31 Tue 1 Wed 2 Thur 3 Fri 4 Sat 5 Sun 6 d=85 d=93 d=112 d=108 d=99 d=65 d=80 b=$84.34 b=$92.07 b=$153.12 b=$112.34 b=$92.57 b=$54.30 b=$62.33 7 8 9 10 11 12 13 d=91 d=102 d=135 d=120 d=92 d=53 d=44 b=$88.47 b=$122.00 b=$172.15 b=$142.34 b=$95.67 b=$42.34 b=32.34 14 15 16 17 18 19 20 d=67 d=85 d=110 d=97 d=93 d=72 d=66 b=$54.37 b=$72.48 b=$122.47 b=$99.97 b=$92.34 b=$55.18 b=$54.54 21 22 23 24 25 26 27 d=86 d=104 d=157 d=140 d=122 d=95 d=85 b=$89.11 b=$130.02 b=$199.93 b=$178.25 b=$122.20 b=100.69 b=$85.18 28 29 30 31 1 2 3 d=84 d=92 d=114 d=100 d=82 d=60 d=75 b=$84.33 b=$93.44 b=$155.67 b=$101.01 b=$78.77 b=$53.92 b=$62.74 Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 38 Property of Nomis Solutions Inc. - Confidential Material Agenda • Introduction to Revenue Management • Elements of Revenue Management • Capacity Control • Overbooking • Network Management • An RM System Example Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 39 Property of Nomis Solutions Inc. - Confidential Material Sporting Event Revenue Management A classic Revenue Management opportunity: • Fixed, immediately perishable inventory • Seating sections in stadiums are similar to cabins in airplanes • Section prices are fixed prior to season starting • Season tickets are offered first; all other seats are free-sell • Bookings come in over time…from a year out to the day of the game • Most baseball teams have a range of discounts that apply to a seating sections market segments • Jnr, Snr, 4H, buy-one-get-one-free Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 40 Property of Nomis Solutions Inc. - Confidential Material EventRM – Challenges • Industry resistant to change - • Very risk averse -- like quick sell-outs (especially concerts) • Fear public perception of venue trying to “gouge” the fans with higher prices -- want to increase revenue without increasing price • Bookings do not follow traditional industry approach of lowest value books first, highest value books last • value of ticket is not correlated to time of booking • Availability denials not currently captured • Data can be sparse Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 41 Property of Nomis Solutions Inc. - Confidential Material EventRM – What it Does • Forecasts demand for each market segment • historical data enables market segmentation • Maximizes expected revenue from forecasted remaining demand into remaining capacity • Recommends which market segments to sell to • venue sets rate structure for market segments • event manager determines which market segments are always open • event manager has final say on which segments to keep open for sale or to close • DOES NOT recommend changes in price Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 42 Property of Nomis Solutions Inc. - Confidential Material Event Overview Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 43 Property of Nomis Solutions Inc. - Confidential Material Pricing Screen Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 44 Property of Nomis Solutions Inc. - Confidential Material Booking Pace Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 45 Property of Nomis Solutions Inc. - Confidential Material Demand Forecast Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 46 Property of Nomis Solutions Inc. - Confidential Material Optimization Parameters Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 47 Property of Nomis Solutions Inc. - Confidential Material Re-Optimize Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 48 Property of Nomis Solutions Inc. - Confidential Material Optimization Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 49 Property of Nomis Solutions Inc. - Confidential Material Availability Controls Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 50 Property of Nomis Solutions Inc. - Confidential Material Summary • Revenue management is the science of setting availabilities for multiple fare classes in the case in which capacity is constrained and perishable. • The decision analytic approach gives lots of insight and some useable answers to basic revenue management problems without lots of annoying multiple integrals! • Many dynamic revenue management implementations involve calculating the opportunity cost of a unit of remaining capacity and accepting only those requests whose fare exceeds the opportunity cost. Copyright Robert L. Phillips. 2006. All Rights Reserved. Page 51 Property of Nomis Solutions Inc. - Confidential Material