WhatisRevenueManagement?
Revenuemanagement(RM)isarelativelynewfielddevotedtothestudyofideas,procedures,andmodels
that increase a firms revenue given fixed and perishable capacity. But as a new field still in the
development stage, it has no universally accepted definition. The terms revenue optimization, yield
management,demandmanagement,andrevenueprocessoptimizationaretermsusedtoaddresssimilar
techniquesandideas.Atitsheart,allofthesetechniquesactivelymanagequantityand/orpricetocapture
greaterrevenuesforthefirm.ThegeneralideabehindRMisremarkablysimple:identifydifferentclasses
ofcustomers(marketsegments)andthenexploitdifferencesin theirwillingnessto pay.Forexample,
airlinesandhotelsroutinelyidentifydifferentcustomerclasses(e.g.,businessvs.leisure)andsetdifferent
faresforeachclass.AnyonewhohasbeentoldImsorry,wedonthaveanymore[rooms,seats,etc.]at
thatprice,hasexperiencedRMinaction.
Thefocusofthisclasswillbeonrevenuemanagementmodels.Virtuallyallrevenuemanagementmodels
fallintooneoftwocategories:modelsthatmanagepricesgivencapacity(pricebasedRMmodels),and
modelsthatallocatecapacitygivenprices(quantitybasedRMmodels).Inbothtypes,thefirmmaximizes
totalcontribution,revenue,orsomecombinationthereof.Inpricebasedmodels,thedecisionvariables
arethepriceschargedto variousmarketsegments.In quantitybased (or capacitybased)models,the
marketissegmentedbypredeterminedandfixedpricepoints(calledfares),andthedecisionvariables
arethesalequantitiesassignedtothepricepoints(i.e.,howmanyunitscanbesoldateachpricepoint).
Wewillconsiderbothtypesofmodelsinthiscourse,althoughgreateremphasiswillbeplacedonthe
quantitybasedmodels.Inasense,thetwotypesofmodelsarenotasdifferentastheyseem.Afterall,if
aparticularpricepointisnotsufficientlyprofitableinaquantitybasedmodel,itwillbeassignedaquantity
ofzero,inwhichcasethepriceeffectivelydisappears.
ThePromiseofRevenueManagement:PriceOptimization
InordertounderstandwhyRMhasbecomepopular,weintroduceaproblemfromclassicaleconomics.
Letussupposethatthedemandforaparticularfirmsserviceorproduct(baseballticket,airlineseat,
hotelroom,etc.)isafunctionofprice, d ( p ) .Thefunction d ( p ) istermedtheresponsefunction.We
wouldexpectthegraphof d ( p ) tobedownwardsloping,i.e.,thedemanddropsaspriceincreases.For
example,wemightrunanexperimentwherewesteadilyincreasepriceandobservethefollowingprice
responsefunction:
ResponseFunction
Demand(Units)
12,000
10,000
8,000
6,000
Response
4,000
2,000
0
UnitPrice
0
10
20
30
40
Copyright20142015,JohnSemple
A response function like that illustrated in the previous graph is hard to obtain in practice (the data
requirementsaresignificant).Moreover,itisnoteasytoworkwithfromamodelingperspectivebecause
therelationshipbetweenpriceanddemandlacksasimpleform.Forthisreason,wetypicallyuseastylized
response function that serves as a reasonable approximation to the true response function. Three
commonresponsefunctionsaregiveninthetablebelow:
CommonResponseFunctions
Linear
d ( p ) D mp
D 0 , m 0
Exponential
d ( p ) Ce p
C, 0
d ( p)
Logit
Ce a bp
1 e a bp
C , b 0 .( a 0 istypical)
(Commonlyusedinpractice)
Theresponsefunctioncaptures,inacompactway,theimpactofpriceonthefirmsdemand.Ivegraphed
alinearresponsefunctionwithD=10,000andm=800,alongwithanexponential(C=10000,=.15)and
alogit(C=10000,b=.15,a=.15)responsefunctionbelow:
ResponseFunctions:Linear,Exponentialand Logit
12000
10000
8000
Exponential
6000
Logit
4000
Linear
2000
0
0
10
20
30
40
Nowletussupposethatthemarginalcostis$2andtheresponsefunctionislinear.Ifyoucanonlysetone
price,whatisit?Fromclassicaleconomics/optimizationtheory,weoptimizethetotalcontribution1.In
otherwords:
Max p 2 10000 800 p
s.t p 0 .
YoucandothiswithExcel(orcalculus)toarriveatthepricep*=$7.25.Atthisprice,4200customersbuy
ataunitmarginof$7.25$2=$5.25.Thetotalcontributionis$4200*($7.25$2)=$22,050.Theshaded
Totalcontributionequalsthesumofallincrementalrevenueslessallincrementalcosts.Anyfixedcostthatis
consideredsunkisnotincluded.
Copyright20142015,JohnSemple
rectanglebelowrepresentsthiscontribution;onesideoftherectangleisthenumberofbuyers,theother
istheunitmargin.
PriceResponse
12000
10000
8000
6000
PriceResponse
4000
2000
0
0
10
20
Nowsupposetherearetwomarketsegments,onethatwillpayalowerprice,andonethatpaysahigher
price.Canwedobetter?
Supposewedividethemarketinto2segments:(segment1)thosethatareunwillingtopaymorethan$6;
and(segment2)thosethatarewillingtopaymorethan$6.Ifwecoulddividethemandkeepthemapart,
thenwhatpricewouldyouchargeeachsegment?
Again,optimizationcanbeusedtofindthecorrectpriceforeachsegment.Therearenowtworesponse
functions,oneforeachsegment:
Responsefunctionforsegment1: d1 ( p1 ) 4800 800 p1 for 0 p1 6
Responsefunctionforsegment2: d 2 ( p2 ) 10000 800 p2 for 6 p2 12.5
(Notethepricerangesprohibitnegativedemands.)Thefirmsgoalistomaximizetotalcontribution:
Max
p1 , p 2
s.t.
p1 24800 800 p1 p2 210000 800 p2
0 p1 6
6 p2 12.5
Theoptimizationrevealsoptimalprices p1* 4.00 and p2* 7.25 ,whichequatestoatotalcontribution
of$25250.Segment1thereforeadds1600customerswhoeachcontribute$2($4$2)tothefirmstotal
contribution,representinga14%gainoverthecontributionachievedbyoneprice.
Copyright20142015,JohnSemple
PriceResponse
12000
10000
8000
6000
PriceResponse
4000
2000
0
0
10
20
Thisexamplerevealswhyfirmsareincreasinglytryingtosegmentcustomersandchargedifferentprices:
Thepotentialis huge.In theory,ifafirm could charge everyone thepricetheyarewilling topay,the
contributionwouldequalthetriangularareashownbelow2.Whilethisisimpossibletoachieve,youcan
see(orimagine)thataddingadditionalsegmentswouldimprovetotalcontributions.
PriceResponse
12000
10000
8000
6000
PriceResponse
4000
2000
0
0
10
20
Assumptions
Whatwouldweneedtodofromapracticalstandpointtomaketheprecedingtwosegmentanalysisa
reality?Thereareclearlymanyassumptions.Wewilldiscusstheseinclassandtalkabouthowcompanies
aretryingtomakethemstickinpractice.
PricingwithLimitedCapacity
The previous price optimization problem did not include a capacity constraint, but most revenue
managementapplicationsdo.Forexample,anairplanehasafixednumberofseatsavailable;ahotelhas
afixednumberofrooms;etc.Tounderstandtheimpactofasupplyconstraint,considertheStanford
StadiumexamplefromRobertPhillipsbookPricingandRevenueOptimization(StanfordPress,2005).
TheStanfordStadiumProblem
Whateconomistsliketocallthirddegreepricediscrimination.
Copyright20142015,JohnSemple
StanfordStadiumhas60,000seats.Ticketsaresoldtotwosegments:Generalpublic,andStudents.The
responsefunctionsforeacharegivenby
GeneralPublic: d g ( p g ) 120000 3000 p g
Student:
The notation
d s ( ps ) 20000 1250 ps
means the positive part; e.g., 20000 1250 ps max20000 1250 ps ,0 . This
simply ensures the response function does not yield negative demands. The same outcome could be
ensuredbyimposingappropriaterangesoneachprice(Isometimesdothisinstead).
HowshouldStanfordsetpricesforeachsegment?(Youmayassumethevariablecostis0.)
Afteralittlethought,Isuspectyoudarriveattheoptimizationproblem:
Max pg 120000 3000 p g ps 20000 1250 ps
pg , ps
s.t. 120000 3000 pg
20000 1250 p
60000
p g , p s 0.
IvesolvedthisprobleminanExcelSpreadsheet(wewillalsobuildoneinclass),andobtainedtheanswer
p*g $22.35 and ps* $10.35 ,yieldingatotalcontributionof$1256471.Inthissolution,52941tickets
areboughtbythegeneralpublic,and7059arepurchasedbystudents.Bycomparison,ifStanfordcould
onlyselectoneprice(addtheconstraint p g ps totheprecedingoptimizationproblem),theywould
selectasingleprice p* 20 andthetotalcontributionwouldbe$1200000.Intheonepricesolution,all
60,000ticketsarepurchasedbythegeneralpublic;nostudentswouldbuyaticket.Thismeanstheone
pricemodelisinferiorinseveraldimensions.
TheThemeParkProblem
AnotherexamplerelatedtoentertainmentistheThemeParkexample(again,fromRobertPhillipsbook
PricingandRevenueOptimization,StanfordPress,2005).Inthisexample,businessrepeatseveryday
justlikeathemeparkandeachmarketsegmentcorrespondstoaparticularday.Boththenumberof
people and their willingness to pay (price sensitivity) depend on the day of the week. This type of
segmentation seems plausible for theme parks and other entertainment businesses that offer a
repeatableexperienceeachdayoftheweek(e.g.,skiareas,boattours,etc.).Peopleoftenselfselectinto
aparticulardaytofindapricethatmatchestheirbudget.
Capacity at the park is 1000 customers per day. The linear response functions on each day have an
intercept(D)andslope(m)asgiveninthetablebelow.
DayofWeek Intercept(Di) Slope(mi)
Copyright20142015,JohnSemple
Sunday
Monday
Tuesday
Wednesday
3100
1500
1400
1510
62
50
40
42
Thursday
2000
52.6
Friday
2500
55.6
Saturday
3300
60
PartI.(SinglePrice)Iftheparkcanonlychargeoneprice,whatisit?(Assumetheincrementalcostis0.)
Let p representtheprice.Wemustbecarefulherebecausetheparkneednotsatisfyallofthedemand
oneveryday.Forexample,theymightsetthepricealittlelowonweekendslessthanwouldbeoptimal
iftheywereonlyopenonweekendsbecausetheymakeupforitwithincreasedrevenueonweekdays.
Therefore,let xi representhowmuchdemandtheycantakeondayi;thisvaluerepresentsattendance.
Theappropriateoptimizationproblemis
Max p x1 x2 ...... x7
p , x1 ,...., x 7
s.t.
xi Di p mi
for i 1,2,...,7 (cannottakemorethandemandondayi)
xi 1000
for i 1,2,...,7 (cannottakemorethan1000peopleperday)
p, xi 0
for i 1,2,...,7
Thisisanothernonlinearoptimizationproblem.ItcanbesolvedinExceltoobtaintheoptimalpriceof
$25.50andatotalcontributionof$119,901.
Part II. (Daily Price) What happens if the park can charge a different price each day? (Assume the
incrementalcostis0.)
Now,ineffect,eachdayisadifferentproblem.Wecouldsolveeachdayandsumtheresults,oralterthe
previousproblemslightlyanddoeverythingallatonce.
Let pi bethepricechargedondayi(1=Sunday,2=Monday,etc.),andlet xi bethenumberoftickets
sold(attendance)onthatdayatthatprice.
Theappropriateoptimizationproblemis
Max p1 x1 p2 x2 p3 x3 p4 x4 p5 x5 p6 x6 p7 x7
p1 ,..., p 7
x1 ,...., x 7
s.t.
xi Di pi mi
for i 1,2,...,7 (cannottakemorethandemandondayi)
Copyright20142015,JohnSemple
xi 1000
for i 1,2,...,7 (cannottakemorethan1000peopleperday)
pi , xi 0
for i 1,2,...,7
Theoptimalpricesareshownbelow.Thenewrevenueis$155266.1,whichisaconsiderableimprovement
overthesinglepricemodel.
pSUN pMON
33.87 15
OptimalDailyPrice
pTU
pWED pTHU
pFRI
pSAT
17.5 17.98 19.01 26.98 38.33
PriceOptimizationwithDiversion
InthepreviousThemeParkproblem,RobertPhillipsproposesaninterestingextensioninvolvingdiversion.
Diversionoccurswhencustomersaredivertedfromonedaytoanotherbasedonprice.Ingeneral,thisis
averyhardproblem,butPhillipsoffersaninterestingandsimplifiedmodelthatcapturestheessenceof
diversion.SupposeafixednumberPhillipsuses8intheThemeparkexamplearedivertedfromone
daytoanotherforeach$1differenceinprice.Thismeansthedemandconstraintsforeachdayinthe
ThemeParkproblembecome
xi Di pi mi 8 p j pi
7
for i 1,2,...,7
j 1
Youcanthinkofeachresponsefunctionashavingalinearportionequalto Di pi mi plusadiverted
p
7
amountequalto 8
j 1
pi .ResolvingthepreviousThemeParkproblemwithdiversionyields
OptimalPrices
Demandwithout
Diversion
Demandwith
Diversion
NetChange
pSUN
28.89458269
1308.535873
1000
308.535873
pMON
18.11544908
594.227546
889.3231553
295.0956092
pTU
19.48122728
620.7509087
839.3629386
218.6120299
pWED
19.64498331
684.9107008
894.3523928
209.441692
pTHU
21.26667364
881.3729664
1000
118.6270337
pFRI
25.17527561
1100.254676
999.9999991
100.254677
pSAT
31.11690308
1432.985815
1000
432.985815
Copyright20142015,JohnSemple
Thedemandwithoutdiversionreferstothedemandthatwouldoccurbasedonthelinearresponseatthe
given price (i.e., Di pi mi ); the demand with diversion includes the diverted demand term
8 p j pi .Thedemanddivertedisequaltothenetchangelistedinthelastcolumn.
7
j 1
TheConnectionbetweenResponseFunctionsandWillingnesstoPay
Forthosewhoarecomfortablewithalittlemath,readon.Ifnot,pleaseskiptothelastparagraph
ofthissection!
There is a connection between the response function and a populations willingness to pay.
Suppose w(x ) isthedensityfunctionforthedistributionofbuyerswillingnesstopay.Bythiswe
mean
p2
w( x)dx =Fractionofthepopulationthatwouldpaybetween p and p
1
p1
Thenthecorrespondingnumberofpeoplewhowouldbuyis
p2
D w( x)dx
p1
Where D isthenumberoftotalpotentialbuyers( D isessentiallythenumberofpeoplewhowouldwant
theproductforfree).Thisimpliestheresponsefunction,ortotalnumberofbuyersatprice p ,is
d ( p) D w( x)dx .
p
Thisalsomeans
d ( p ) D w( p ) d (0) w( p ) or w( p )
d ( p )
d ( 0)
(Notewehaveusedthefundamentaltheoremofcalculusandthefact d (0) D .)
Theimportantthinghereisnotthemathbutthefactthattheresponsefunctionandthedistributionfor
apopulationswillingnesstopayareeffectivelyequivalentnotions.Byequivalencewemeanifweknow
one,wecandeterminetheother.
AdditionalReading
Copyright20142015,JohnSemple
Forthoseinterestedinpricingoptimization,seechapters3,4,and5ofPricingandRevenueOptimization
by Robert Phillips, Stanford University Press, Stanford, California, 2005. This lecture draws extensively
fromthesethreechaptersofPhillipsbook.
Homework
Homework1(individualassignment)willbedistributedinclassorpostedonBlackboard.
Copyright20142015,JohnSemple
