ASP On-Demn versus In-House Softwre Solutons Dn M School of Informton Systems Sngpore Mngement Unversty Abrhm Semnn Xerox Professor of Computer n Informton Systems n Opertons Mngement W. E. Smon Grute School of Busness Amnstrton Unversty of Rochester Revse: Aprl 9, 007 The uthors grtefully cknowlege useful comments from Professors Rjv Dewn, Roy Jones, Rvnr Mnten, Eel Pnker, Mchel Rth, n Anjn Susrl, n from semnr prtcpnts t the Unverstes of Mchgn, Rochester, n Mryln, n the Workshops on Informton Systems n Economcs (WISE).
ASP On-Demn versus In-House Softwre Solutons Dn M School of Informton Systems Sngpore Mngement Unversty Abrhm Semnn Xerox Professor of Computer n Informton Systems n Opertons Mngement W. E. Smon Grute School of Busness Amnstrton Unversty of Rochester Abstrct Applcton Servce Provers (ASPs) elver on-emn nformton processng servces to user frms v the Internet. Rp technologcl evelopments n telecommunctons n recent yers hve me ASPs n ttrctve lterntve to purchsng, nstllng, n mntnng mofble off-the-shelf () softwre solutons. We stuy severl crtcl spects of user frm s choce between n ASP n softwre. The compettve moel consers heterogeneous users tht ffer n terms of ther expecte trnscton volumes n voltlty, whle ASPs n venors ffer n terms of ther prcng structure, setup cost, system customzton blty, ntegrton n servce level rrngement. Our results entfy n chrcterze the equlbrum contons uner whch ASPs n venors cn coexst n compettve mrket, n they expln whch frms coul be the prmry benefcres of ech venor type. Interestngly, we fn tht the ASP' entry to the mrket benefts not only ts own clents but lso the users, n we expln why espte tht, the compettve vntge of the pproch ecreses sgnfcntly s users trnscton voltlty, or busness uncertnty, ncreses. Hence, the vlue e by ASPs comes s much from the effcent poolng of trnscton voltlty rsks s from the reucton n IT mplementton costs. We lso look t the competton between n ASO venors n longer tme horzon wth possble future chnges n softwre qulty, n we show when the ASP cn tke over the whole mrket, even f ts qulty s stll lower thn the softwre. Our fnngs suggest tht to sty compettve, ASPs shoul nvest n both mprovng softwre qulty n n reucng servce prces over tme. Surprsngly, the opposte lso hols, n long wth ny reltve qulty ecreses, the ASP shoul rse ts prces, provng reltvely n nferor servce t hgher chrge. Ths hppens becuse t tht cse the ASP wll serve mnly those frms who re prce out of the mrket, n cn behve monopolstclly t tht segment. Fnlly, our moel explns when softwre users coul be better off espte n ncrese n trnscton voltlty n the mrket, n when they wll be less lkely to swtch to n ASP even s the reltve qulty of the ASP s pplctons ncreses over tme. We prove tht f they o swtch, ther prmry ncentve s btng hgher trnscton or busness voltlty rsks rther thn reucng trnsctons costs per se. The uthors grtefully cknowlege useful comments from Professors Rjv Dewn, Roy Jones, Rvnr Mnten, Eel Pnker, Mchel Rth, n Anjn Susrl, n from semnr prtcpnts t the Unversty of Rochester, the Unversty of Mryln, the Workshop on Informton Systems n Economcs (WISE) 004, n WISE 005.
Introucton Usng sophstcte enterprse softwre such s Enterprse Resource Plnnng (ERP), Customer Reltonshp Mngement (CRM), or Electronc Mecl Recors (EMR) systems hs become compettve necessty. Trtonlly, such softwre hs been elvere n the form of mofble off-the-shelf () proucts. The venor sells the softwre pplcton to users n helps to customze n nstll t on users stes. The users must prove IT nfrstructure, hrwre, n support servces n orer to enble contnuous use of the softwre. In the pst few yers, the Internet hs gven rse to Applcton Servce Provers (ASPs). ASPs offer bunle of pplctons, n IT nfrstructure, n ll necessry support servces to users cross network. Uner the ASP busness moel, the pplcton n users t re store off-ste n centrl locton run by the ASP. The ASP s n chrge of ll IT support servces, nclung ly softwre mntennce, t bckups, softwre upgres, n securty. ASPs represent n on-emn busness moel, where softwre s elvere s servce. Unlke trtonl perpetul lcensng, the softwre s prce s servce, n typclly users py fee per trnscton. Users pyments re closely te to the ctul utlty obtne they py only when they hve emn for the softwre. Such prcng shfts expenses from the cptl buget to the opertng buget n sgnfcntly cuts users strt-up costs; t lso helps users to mnge pek-n-vlley trnscton-lo problems well whle keepng costs own. In mny cses, usng n ASP my prove cheper thn ownng n mntnng n n-house IT system. Users expect to sve money on support n upgre costs, IT nfrstructure, IT personnel, n mplementton (Lcy 006). A recent survey shows tht the top reson frms choose n ASP over trtonl softwre s the belef tht the former s cheper to run (The Economst 006). Moreover, users A prouct s commercl softwre pplcton whose source coe cn be customze to meet customer s prtculr requrements... [It] s esgne to be esly nstlle n to nteroperte wth exstng system components. See http://whts.techtrget.com for more nformton. In ths pper, we use the terms ASP, on-emn softwre, n softwre s servce nterchngebly. 3
cn rep the benefts of the ASP s economy of scle, ccess IT expertse, n gn system flexblty (Dunn 005; Murphy 005; Sngh et l. 004). ASPs re enjoyng prosperous tmes. Accorng to AMR Reserch, the on-emn softwre mrket s growng more thn 0% yer, compre wth sngle-gt growth n trtonl softwre (Lcy 006). ASPs re expecte to generte $0 bllon n nnul revenue by 009, up from $.5 bllon n 006 (Plltto 006). An ncresng number of softwre venors, nclung nustry gnts such s IBM, SAP, Orcle, n Mcrosoft, re movng to such busness moel. For exmple, IBM s offerng IBM-on-emn, whch llows corporte users to cqure IBM s computng power n softwre pplctons s servce. 3 Menwhle, IBM s lso provng pckge of servces n ncentves to help other softwre compnes eploy ther proucts s hoste pplctons (Plltto 006). Orcle s rnke s one of the top ten ASPs. Its onlne offerng, E-busness Sute, hs successfully ttrcte mny smll compnes by chrgng low trnscton fee. 4 In Jnury 006, Orcle cqure Sebel, n t now uses ts fleglng CRM onemn softwre to compete wth Slesforce.com, the most successful ASP provng onlne CRM, whch reports 399,000 pyng subscrbers (Vr 006). Bll Gtes hs proclme tht the emergence n rse of on-emn softwre wll be the next se-chnge n computng (Nccol 005). Mny observers expect n nustry-we revoluton brought on by the ASP moel: Softwre s servce s the bggest thng to hppen n softwre n 5 yers (Knorr 004). For frms n nee of enterprse softwre, the ASP consttutes vble lterntve to the trtonl softwre soluton. As n exmple, we stue the cse of Mecl Dgnostc Imgng (MDI), n mgng center conuctng bout 0,000 stues nnully (Mrtn 004). It ws lookng for stte-of-the-rt PACS (Pcture Archvng n Communctons System) to mnge n elver ptents mecl mges. MDI looke frst t soluton from such venors s Cnnon, Onyx, n Fuson. All of them woul hve requre MDI to bul n 3 See http://www-.bm.com/servces/onemn/success.html. 4 See www.orcle.com. 4
mntn the PACS n-house. The bs from these venors typclly exceee one mllon ollrs just for the up-front nvestment n the softwre. Atonlly, MDI entfe sgnfcnt costs ssocte wth nstllton, customzton, ongong support, upgres, t securty n rounthe-clock mntennce. MDI etermne tht n on-ste system ws too expensve. To reuce the requre ntl costs, MDI chose n ASP, MePACS Dsply, whch proves vewer softwre for rologsts n physcns, s well s rchves, hrwre, softwre, n mntennce. 5 Ths ASP chrges flt fee of $3.89 per stuy, nclung ll the necessry ongong support (M 005). 6 MDI thus ws lso releve of the nee to hre n mnge n nternl IT stff for the rology eprtment. Moreover, the ASP offers hghly sclble processng servces. When MDI fces rnom fluctutons n ptent volume, MePACS respons ccorngly from the remote ste tht hnles most of the clncl t. Mny busnesses fce smlr ecson problem. In web survey by ThnkStrteges, fully one-thr of 8 responents were lrey usng ASPs, n nother thr were conserng usng one wthn the followng months (Kpln 005). A reserch report from Summt Strteges nctes tht smll n meum busnesses wth lmte IT resources n constrne bugets re more lkely to use such fee-per-trnscton pplctons (Grner 004). Some lrge orgnztons re ttrcte to ASPs s well. Amzon.com, Csco, Sprnt, Morgn Stnley, Nok, n Trget re ll clents of Slesforce.com (The Economst 006; Rcel 006). It s cler tht ASPs re stelng mrket shre from venors of softwre. On-emn softwre s powerful tren, one tht s becomng n mportnt sruptve force n the softwre nustry: It s not the en of softwre. It s just nother wy of eployng t (Sh Agss, present of SAP s prouct n technology group). However, the long-term success of ASPs remns uncertn. In ts formtve stge, the ASP mrket s experencng sgnfcnt consoltons. The fer tht ASPs coul fl s mn obstcle 5 See www.mepcs.com. 6 Most ASPs n the helthcre nustry chrge smlrly for ther PACS systems. Stentor, for exmple, offers the mge elvery softwre Ste on per stuy bss n chrges, on verge, $3.50 per elvery. 5
to the lrge-scle opton of ASPs. Dt securty n relblty s well s pplcton control re lwys mong users top concerns (Benrz 006). In ton, lthough there s no oubt tht the ASP moel oes hve numerous vntges, the trtonl n-house soluton cn elver superb results. The source coe of the softwre cn be mofe to meet user s prtculr requrements, so the softwre s well customze n cn esly nteroperte wth the user s exstng system components. In contrst, the ASP offers one-to-mny soluton, wth lmte customzton. 7 Its prouct s not tlore to ft n nvul user s specfc requrements or unque busness envronment. As result, the user my nee to py extr to mke the ASP s stnr softwre pplcton work smoothly wth ts exstng IT systems. For nstnce, n the MDI cse, gettng the vrous systems to communcte n work together requre some effort. MDI h to run n tonl moule tht took the output from ts own n-house reserch nformton system n converte the output nto n mge fle for storge n the rchve (Mrtn 004). MDI ncurs converson costs for every trnscton. Some reserchers hve lrey looke t the ASP busness moel. For exmple, Susrl et l. (003) use questonnres n orer to entfy the etermnnts of goo ASP contrct. Ther work oes not, however, touch on the theoretcl bse of the ASP busness moel. Cheng n Koehler (00) stuy monopoly ASP s prcng problem. They conser the nterreltonshp between prce n queung ely n conclue tht vrous prcng polces, such s fxe fee, two-prt trff, or metere prce, re equvlent. These n other prevous works, however, only nlyze ASPs n n solte frmework. They o not stuy mrketplce n whch venors lso prtcpte. Hence, they overlook the n-house IT opton. Lttle work hs been one on the competton between the ASP n solutons. It s not cler wht fctors ply mportnt roles n users IT choces. We stuy softwre mrket n whch corporte users hve these two IT optons for externl sourcng. The softwre venors re fferentte n the followng wys. Frst, they elver fferent proucts: customze softwre pplcton (from the ) versus bunle of stnr softwre n servces (from the ASP). 7 SAP beleves tht ths lck of ntegrton wll eventully le most lrge corportons tht currently re users of onlne CRM to move bck to n n-house system (Hmm 006). 6
Secon, they opt stnct prcng moes: n outrght purchse (the ) versus per trnscton fee structure (the ASP). Thr, they employ fferent elvery methos: softwre nstlle on user s n-house server (the ) versus n nterfce elvere over the Internet remotely (the ASP). Users therefore hve fferent cost structures n ber fferent rsks n elng wth ech venor type. In prctce, choosng n mplementng n enterprse softwre system re hghly complex ecsons. CIOs nee to compre the two optons on multple ponts, such s setup costs, prcng structure, blty to ntegrte t, sclblty, n servce level rrngements. Ther ecsons nvolve not only economc but lso technologcl n securty ssues. We o not ttempt to nclue ll the relte fctors; nste, we stuy frms choces by focusng on the economcs ssues only. In prtculr, we re ntereste n estmtng the reltve economc vntges of usng ech venor type. We ssume tht the users (they re frms) hve me the ecson to nstll computerze nformton system, mng to mprove nternl busness processes or to ncrese vlue for en customers. The ecson nvestgte n ths pper s the choce of elvery chnnel: obtnng the system through n ASP or. Strtegc opton behvors such s herng or elyng opton for better system or lower prce re gnore. ASPs n venors prce ther offerngs fferently. The venor sells the softwre (one-tme, fxe-fee prcng), n the ASP rents t (trnscton-bse prcng). Our work thus s relte to reserch on sellng n rentng nformton goos. Most prevous work n ths lterture exmnes one-frm settng. Vrn (000), for exmple, consers monopoly frm s choce when both sellng n rentng re fesble chnnels. Chouhry et l. (998) nlyze the possblty of monopoly frm sellng n rentng pckge of softwre smultneously, n Sunrrjn (004) compres fxe-fee prcng n usge-bse prcng for nformton goos for monopoly frm. Fshburn et l. (997) re mong the few who compre sellng n rentng n compettve mrket. They moel repete competton gme n whch one frm chrges fxe subscrpton fee n the other chrges bse on usge. In ther pper, the two sellers offerngs re homogeneous; the only fference between the two sellers s the prcng metho. 7
We, however, stuy the competton between two venors tht re fferentte n both offerngs n prcng structures. We entfy severl nterestng fetures of such mrket. Frst, the vlue e by ASPs comes s much from the effcent poolng of emn uncertnty rsks s from the reucton n sgnfcnt IT mplementton costs. Specfclly, when the voltlty of trnsctons ncreses, more users opt for the ASP, n the reltve vntge of the ASP pproch ncreses sgnfcntly. Secon, we fn tht the softwre mrket my show three fferent structures. Uner frly bro contons, the ASP n solutons wll coexst. The mrket wll be segmente n such wy tht frms wth low trnscton volumes opt for the ASP moel becuse of the chepness n sclblty, n frms wth hgh trnscton volumes prefer the moel to enjoy softwre tht fts ther specfc busness nees well. Uner certn contons, however, one of the provers s unble to survve, n the softwre mrket wll be omnte by one busness moel lone. Thr, our fnngs suggest tht fferent busness moels re sute for fferent types of softwre pplctons. In prtculr, the ASP moel works well for softwre pplctons such s fnncl trng, electronc mecl recors, n pyroll systems. These pplctons typclly serve functons wth hghly voltle emn but stnr process. For complex busness softwre, such s enterprse resource plnnng (ERP) systems, we stll suggest the trtonl nhouse moel. We lso stuy the ASP- competton n longer tme wnow wth potentl rel-tme chnges. We obtn surprsng yet mportnt fnngs. If users re concerne bout future chnges n the busness envronment, such s hrwre upgres on ther se, whch ecreses the ASP softwre qulty, ASPs shoul ncrese ther prces. They shoul gve up the competton wth the venor for hgh-volume users n nste focus on ttrctng smll n meum frms. If users expect the unft costs of usng stnr softwre to ecrese ue to the vnce of web technologes, whch ncreses the qulty of ASP softwre, ASPs shoul reuce ther prces to 8
compete ggressvely wth venors for lrge corporte users. These fnngs help us to suggest useful compettve strteges to provers of on-emn softwre. Fnlly, we show tht competton n longer wnow gves vntges to the soluton. Once the ntl mplementton costs re sunk, users only nee to py ongong mntennce costs. Hence, users of exstng n-house systems hve lttle ncentve to swtch to onemn softwre, even f the ASP cn promse qulty mprovement over tme. If these users o swtch, they o so n orer to bte hgh cpcty rsks cuse by hgh trnscton voltlty rther thn to reuce opertng costs. The rest of the pper s orgnze s follows. In Secton, we escrbe the moel. We nlyze the ASP- competton n entfy the key etermnnts of the future of the ASP busness moel n Secton 3. Secton 4 extens the competton nlyss over longer wnow. Secton 5 conclues the pper. The Moel We frst moel the mrket before the ASP ppers, n then the mrket wth both venors.. The Pre-ASP Mrket Softwre pplcton users n our moel re frms. Before the ASP, frms get ccess to softwre systems by nstllng them n-house. The venor sells the pckge softwre to ts user. The venor mofes the source coe to ft the user s specfc busness nees, whch ssures goo ntegrton wth the user s exstng IT system. The venor bers n opertng cost C to serve one user n receves one-tme pyment P from the user. For brevty, we ssume tht the venor proves only one verson of the prouct n tht the lcensng fee s not functon of users trnscton volumes. The venor cn montor the number of users or nstlltons, but t s ffcult to montor the ctul trnscton volume, whch vres over tme. The user must nstll hrwre n IT nfrstructures, hre IT stff, n orgnze n nternl IT group to prove softwre mntennce, t bckups, n securty n cpcty mngement. 9
In orer to keep these IT resources n-house, the user ncurs vrety of costs. Let c enote the servce costs ssocte wth ech use of the softwre (.e., the servce costs per trnscton). Users fce stochstc worklo n terms of the number of trnsctons usng the softwre. Ex nte, users cnnot precsely estmte the ctul trnscton volume. They therefore etermne n nstll the proper servce cpcty n-house bse on the ntcpte trnscton volume. If the ctul volume s more thn the nstlle cpcty, the user fces n nsuffcent cpcty problem n loses some potentl utlty. If the ctul volume s less thn the nstlle cpcty, the user fces n overcpcty problem, n t stll pys the costs of holng the excess IT cpcty. In both cses, the user ncurs losses. It bers cpcty rsks cuse by emn uncertnty. Fgure. The Softwre Soluton customzes n nstlls the softwre wth n opertng cost C per user stochstc trnscton volume The softwre venor A user frm wth n-house servce cpcty (servce cost c per trnscton) pys one-tme purchsng prce P vlue of u per trnscton stochstc trnscton volume Fgure. The ASP Soluton elvers the bunle of stnr softwre n support servces The ASP (servce cost c per trnscton) A user frm pys prce p per trnscton Note: The blck ot nctes the locton of the softwre system. Fgure shows the soluton. The venor sells the softwre t prce P n bers n opertng cost C per user. The user ncurs servce costs c per trnscton, n ech trnscton cretes vlue of u.. The Dul Mrket The ASP Competes wth the Venor In the ul mrket, users cn ccess softwre pplctons through n ASP. To strt ong busness wth user, the ASP frst moves the user s t to ts own server, whch costs the ASP one-tme setup cost S. The ASP proves IT servces to enble the softwre 0
use, whch costs c per trnscton. We ssume tht the ASP s servce cpcty s plenty. In relty, most ASPs hve estblshe lrge t centers n strong network nfrstructures. They esgn the servce systems wth mple cpcty s wy of reucng the rsk of congeston. ASPs therefore resemble power sttons, n users buy softwre s servce by pluggng nto the ASP gr. The ASP employs one to mny busness structure. To ny nvul user, the pplcton offere by the ASP s not well-customze. Ech trnscton gves the user totl vlue of u-t. The prmeter t mesures the user s sutlty from not usng ts el prouct. In mny cses, t lso represents the cost of extr effort to mke the ASP s prouct work wth the user s exstng IT components smoothly. In the rest of the pper, we cll t user s unft costs. The user pys prce p per trnscton to the ASP. Fgure shows the ASP soluton..3 Stochstc Trnscton Volume When moelng users, we cpture two chrcterstcs. Frst, they hve fferent IT nees. Users IT nees re mesure by the expecte volume of ther use of the softwre. Some frms my use the softwre pplcton more frequently (n expectton) thn others, n these frms hve lrger IT nees. Secon, ech user s ctul trnscton volume s rnom. As seen n Fgure 3, users re unformly strbute on unt-length lne normlze from 0 to. The locton of user on ths lne represents ts expecte trnscton volume. The ctul trnscton volume of user t locton [ θ + θ ],, where the user s locton s rnom vrble tht s unformly strbute on represents ts expecte use of the softwre (n terms of the number of trnsctons), n the prmeter θ mesures the voltlty of the trnscton volume. To vo puttng ny postve probblty on negtve trnscton volume, we o not ssume the sme strbuton for frms wth < θ. In fct, we mke no ssumpton on the strbuton of ther trnscton volumes. It cn be shown tht frms wth low trnscton
volumes re nsgnfcnt becuse they wll be prce out of the mrket n the pre-asp scenro (see the Appenx). Fgure 3. Dstrbuton of Users Trnscton Volumes + θ Expecte trnscton volume 0 θ We mke the followng generl prmeter ssumptons: Assumpton. 0 C < + 4 4( θ ; Assumpton. ( C cθ u (. Assumpton mposes n upper boun on the trnscton voltlty level. Note tht the mxmum expecte trnscton volume mong ll frms s, so ths upper boun s loose. Assumpton mposes n upper boun on the venor s costs for softwre mplementtons. If C excees ths upper boun, no user wll be serve by the venor n the pre-asp mrket, whch s cse of no nterest. 3 Anlyss n Results 3. The Pre-ASP Mrket (the Benchmrk Cse) The venor etermnes ts optml prce P. Gven the prce, users mke two ecsons: whether to buy the softwre, n, f so, the proper IT servce cpcty to nstll n-house. We solve the problem bckwr. Denote the ctul trnscton volume of user (locte t θ ) s D. Ths user s optml cpcty level, gven the softwre prce P n the unt servce cpcty cost c (.e., the servce cost per trnscton c), s obtne by solvng Mx q [ { D, q} P cq ] = uq [ F( q )] + uf( q ) E[ D / D < q ] P cq E u mn, where F (.) s the cumultve ensty functon of the trnscton volume for user. Wth probblty F( ), the ctul trnscton volume wll be lrger thn the user s servce q
cpcty. The user runs the softwre up to ts servce cpcty level n loses excess emn. Wth probblty F ( q ), the ctul trnscton volume wll be smller thn the user s servce cpcty. The user processes ll trnsctons n crres the costs of ownng excess cpcty. The frst two terms n the expresson bove represent user s expecte utlty from usng the softwre; the lst two terms represent the user s costs of purchsng the softwre (P) n costs of estblshng n-house IT servce cpcty ( cq ). Followng Dewn n Menelson (990), we ssume lner functon for cpcty costs. In prctce, cpcty costs coul be step functon, but lner functon s the best frst-orer pproch for step functon, s Zmmermn (979) shows. Solvng the optmzton problem gves q c = + θ. () u The optml cpcty q evtes from the expecte trnscton volume by fctor of c θ, whch my be postve or negtve, epenng on the reltve mgntue of servce u costs (c) n utlty (u). When the trnscton becomes voltle (θ ncreses), the evton grows. The expecte totl utlty for user s Eu ( ) = ( P θ. () u The frst term s the expecte vlue from usng the softwre; the secon term s the user s one-tme pyment to the venor. The lst term represents the user s rect loss ue to trnscton uncertnty; ts mgntue ncreses wth the trnscton voltlty level (θ ). Users wth non-negtve utlty choose to purchse the softwre. The mrgnl user s t P c M = + θ. (3) u c u M Thus, the venor s mrket shre s, n t obtns proft of P C per user. The sellng prce P s chosen to mxmze the venor s proft: 3
Mx P M = ( P C )( ). ( M C ( c The optml prce s P = + θ. (4) u Consequently, the nfferent user hs n expecte trnscton volume of M = C + cθ + >. (5) ( u The venor leves most users out of the mrket. Ths fnng cptures the fct tht smll n meum frms typclly hve h no ccess to n-house enterprse softwre systems, becuse of the hgh fxe costs of purchsng, nstllng, n mntnng such softwre. In ton, equton (4) nctes tht the venor wll reuce ts prce when trnscton voltlty (θ ) ncreses. Pluggng ths optml prce nto equton () shows us tht the reuce prce cn only prtlly compenste for user s rect loss. The user stll ncurs totl loss of θ ue to uncertn trnscton volumes. Moreover, the mrgnl user s u locton ( M ) wll move towr the rght s trnscton voltlty ncreses, whch mens tht the venor s mrket shre ecreses lthough ts sellng prce hs been reuce. Consequently, the venor s proft ecreses. In the pre-asp mrket, then, the venor lso ncurs losses ue to the emn uncertnty on the user s se. Hence, when trnscton uncertnty (θ ) ncreses, every mrket prtcpnt fces worse stuton. Proposton In the pre-asp mrket, frms wth low trnscton volumes re not ble to ffor the system; when the emn for softwre becomes more voltle, both users n the venor re worse off. Regrless of the sze of the users pplcton, the funmentl chllenge they fce s lwys cpcty hegng. Users of softwre must set the servce cpcty n vnce bse on the ntcpte trnscton volume; they therefore re unble to quckly scle ther cpcty up or 4
own ccorng to chngng busness nees. We show n the next secton tht ASPs cn better ress ths cpcty mngement problem by poolng the rsks of mny users. 3. The Dul Mrket wth Competton 3.. Equlbrum nlyss In the ul mrket, the ASP n the venor smultneously ece ther prces, n P. Gven the prces, users ece to use one or the other or to sty out of the mrket. p Conser user wth expecte trnscton volume. Gven (, P), user gns n p expecte utlty of Eu ( ) = ( u p t) from the ASP, n ASP Eu Eu EuASP ( ) = ( P θ from the venor. Snce 0, u n equlbrum the mrket s segmente n such wy tht users wth low trnscton volumes choose the ASP n users wth hgh trnscton volumes choose the softwre. The nfferent user s locton, enote s, s gven by Eu ( ) Eu ( ) ASP =. = p P θ +. (6) c + t u( p + t c) The venor serves users n [,] the users n [ 0, ], wth mrket shre of., wth mrket shre of, n the ASP serves The venor chooses sellng prce P : Mx = ( P C )( ) P. It s esy to show tht ts proft functon s concve n prce. The frst-orer conton s p + t c C θ P = +. (7) u The ASP sets prce p to mxmze ts proft: ( )( ) Mx p ASP = p c S. The frst-orer conton s B( t + c) + Su( t c) p =, where B = Pu + θ. (8) B Su 5
It cn be shown tht the prce equlbrum lwys exsts. 8 We re ntereste n explorng the propertes of the prce equlbrum, n n exmnng how the venor n users re ffecte by the ASP s entry nto the mrket. Proposton When the unft cost t from usng the ASP s prouct s smller thn threshol vlue t, where t u n c t s ecresng functon of the ASP s setup costs S, the ASP s entry nto the mrket results n lower prce, smller mrket shre, n less proft for the venor. Menwhle, ech user s better off, n the totl consumer surplus ncreses. <Proof> See the Appenx. The unft cost t, whch mesures users sutlty (per trnscton) from usng lesscustomze softwre system, plys n mportnt role n the competton. When t s smller thn the gven threshol vlue, common competton outcome s reche: the ASP s entry nto the mrket hs negtve effects on ts compettor (.e., the venor) n postve effects on users. The ASP benefts users n fferent wys: users wth low trnscton volumes, whch otherwse wll be out of the mrket, re ble to ffor the softwre through the py-s-you-go busness scheme; users wth meum trnscton volumes, whch otherwse wll nstll n nhouse system, opt for the ASP becuse the ASP represents cheper soluton; users wth hgh trnscton volumes stck to the n-house soluton but py less to the venor becuse of the ASP s competton. Therefore, ech user s better off, n the totl consumer surplus s strctly mprove. However, such effects pper only when n excees the threshol vlue t t. As Proposton 3 shows, when t ncreses t, the competton exhbts totlly fferent fetures. 8 To see ths, note tht equton (8) s the only soluton to the ASP s frst-orer conton. If we plug t nto the secon-orer conton, we get n expresson tht s lwys negtve. Thus equton (8) gves us the unque proft-mxmzng prce. Beses, there s no better corner soluton: for very low prces, the ASP wll tke over the entre mrket, s shown n Proposton 5, n for very hgh prces, the ASP s proft eventully flls to zero, snce the ASP wll be rven out of the mrket, s shown n Proposton 6. 6
Proposton 3 When t t, the ASP wll chrge prce p u t =. The ASP s entry nto the mrket hs no effect on ether the venor or users. <Proof> See the Appenx. When t < t < u c, the entry of n ASP oes not ffect the venor s prce, mrket shre, or proft. The venor stll behves s f t were the only soluton prover n the mrket, s the ASP n fct s not competng wth t. Although full mrket coverge s cheve, the ASP extrcts ll consumer surplus, so users gn zero utlty. In such cses, the ASP enters the M mrket, serves resul customers (n [, ] 0 ), n gns non-zero proft, but t otherwse hs no effect. If the unft cost t ncreses further, t > u c, the ASP busness moel s not ble to survve. The ASP s revenue from servng resul users s not enough to offset ts ntl setup costs. As result, the ASP cn not mke proft. We nlyze such cses n etl n Secton 3... Next we show how the reltve compettve power of the two venors chnges wth the trnscton voltlty level θ. Lemm sttes tht both venors wll reuce prces s θ ncreses, n tht the lowere prces nfluence ther respectve customers n fferent wys. All nequltes hol when <Proof> See the Appenx. t < t, n ll equltes hol when t t. Frst, we show tht the ASP s users re better off when the trnscton voltlty level ncreses. The on-emn feture of the ASP busness moel llows users to enjoy full cpcty sclblty n to hnle possble emn fluctuton t no rsk. Hence, the users of ASPs re not ffecte negtvely by emn voltlty. Moreover, the ASP wll reuce ts prce ue to the competton from the (.e., p θ 0 ), whch benefts the users. Secon, the venor, when t < t, s wllng to shre more uncertnty rsks thn t woul n the pre-asp mrket;.e., P M u c) θ > P θ =. In other wors, the venor s prce elstcty u wth regr to emn voltlty ncreses. As result, the users of the softwre re better 7
off n reltve sense: lthough they stll ncur totl losses ue to trnscton voltlty (mple by P u c) θ < ), they re better off becuse they now ber less uncertnty rsk. The ASP u not only removes ts own users uncertnty rsks by ong busness on-emn, but lso helps to reuce the other users uncertnty rsks through competton. Fnlly, we fn tht the nfferent customer s poston ( ) wll shft towr the rght when θ ncreses. When trnscton voltlty s hgh, more users ten to choose the ASP. We rech the followng concluson: Proposton 4 When the ASP exsts, ll users re ble to bte the rsks brought on by uncertn trnscton volumes. When trnscton volumes become more voltle, the ASP gns more mrket shre. 3.. One busness moel omntes Here, we nvestgte whether sngle softwre busness moel, ASP or, cn omnte the whole mrket. We prove nswers for questons such s: Is t possble tht on-emn softwre provers wll completely replce venors n become the only elvery chnnel? Or wll ASPs fnlly fll out of the mrket n become only hstorcl footnote? Propostons 5 n 6 stte our fnngs. Proposton 5 There s threshol vlue for the unft cost prmeter t, θ C t = + u Cu + θ p = + c + u <Proof> See the Appenx. S. When t t, the ASP serves the whole mrket usng the prce S. When t t, the ASP rves the venor out of the mrket. In ths cse, ll users wll ccess the softwre through the ASP. Snce S <<, t 0 lwys. Hence, n the extreme C > cse of t=0, the ASP s lwys ble to omnte the whole mrket. 8
On the other hn, the venor cn block the ASP s entry nto the mrket when t s lrge enough, s stte n the followng proposton. Proposton 6 There s threshol vlue for the unft cost prmeter t, t = u c C 4u( S u + ( u + cθ )(. When t t, the ASP s not ble to survve. <Proof> See the Appenx. When t t, t ny prce the ASP mght offer, one of the two constrnts () users obtn nonnegtve utlty n () the ASP obtns nonnegtve proft must be volte. The ASP busness moel therefore s not vble. In ths cse, only the soluton survves. Furthermore, we cn show tht both t n t ncrese n θ. As the trnscton voltlty ncreses, t becomes eser for the ASP to omnte the mrket, n more ffcult for the venor to o so. 3.3 The Key Determnnt(s) of n ASP s Compettve Ablty Ths subsecton summrzes n proves complete vew of the ASP- competton. Lemm t t t. Lemm shows the orer of crtcl vlues of t. The proof of Lemm s strghtforwr. We therefore hve four competton regmes, s shown n Fgure 4. Our moel prects tht n ech regme the prce competton wll occur s follows. Fgure 4. A Complete Dgrm of the Prce Competton 3 4 t t t unft cost t Regme. t 0, t ]: The ASP serves the whole mrket. All softwre s elvere by the ASP. [ Regme. t [ t, t ] : The ASP n busness moels coexst n the softwre mrket, n the competton between the two venors benefts ech user n the mrket. 9
Regme 3. t t, t ] : The ASP n busness moels coexst n the softwre mrket. The [ ASP s not competng wth the but just serves resul users. There s no competton, n the venor stll cts s monopolst. Regme 4. t > u c : The ASP s not ble to survve. The moel s the only softwre soluton. In Regmes,, n 3, full mrket coverge s cheve, whle n Regme 4 (when the soluton omntes), some users remn out of the mrket. In fferent regmes of t, the competton exhbts fferent fetures. When t s low, the ASP soluton omntes, revelng the possblty of ll softwre n ASPs. When t s moerte, the ASP shres the mrket n competes wth the venor. When t grows further, the ASP loses ts compettve vntges n even my not be ble to survve. Thus, we conclue: Proposton 7 As users unft costs from usng not fully customze system ecrese, the ASP s blty to compete ncreses monotonclly. It s n nterestng but not strghtforwr concluson. To some extent, our moel resembles the uopoly competton moel wth vertclly fferentte proucts, wth the ASP s the lowqulty prover n the s the hgh-qulty prover. The unft cost t mesures the qulty fference between the n ASP proucts. In the vertcl fferentton moel, when the qulty fference ecreses, the low-qulty prover ntlly gets better off becuse ts prouct becomes more ttrctve. But t gets worse off s the qulty fference ecreses further, becuse the competton becomes more ntensve n eventully hurts both venors. However, n our pper, we fn tht reucng the vlue of t benefts the low-qulty prover (.e., the ASP) monotonclly. Moreover, when the two venors proucts re close enough, the hgh-qulty prover (.e., the venor) coul be squeeze out of the mrket. Our fnngs evte from trtonl vertcl fferentton lterture becuse the ASP n the venors use fferent prcng strteges. Wht we fn here s more lke wht Vrn (000) showe n monopoly 0
settng: when the qulty fference between sellng n rentng prouct s smll enough, the rentng strtegy omntes. We show tht smlr concluson hols n uopoly competton. Ths result s very mportnt n proves useful mngerl mplctons. It tells us tht ASPs must offer softwre pplctons wth lower vlue of t n orer to survve n succee n the compettve mrketplce. Therefore, not every softwre pplcton cn be elvere through n ASP. The on-emn busness moel s not smply new strbuton chnnel for exstng softwre; nste, t requres funmentlly new set of proucts n technology rchtectures. Applctons tht re sute to the ASP chnnel re those most comptble wth other systems n progrms. On the venor s se, ASPs shoul wrte ther pplctons usng n open lnguge (for exmple, XML), n proper moulr structure, n wth loosely couple nterfce wth other pplctons (Dzubeck 004; Thboeu 004). All these strteges woul help to reuce users unft costs. In ton, nustry-we opton of softwre stnrs n protocols, once cheve, woul reuce the unft costs cross fferent pplctons n therefore ffect ASPs postvely. Some efforts lrey hve been me n ths recton. Mcrosoft, Orcle, n Sun hve jone forces wth IBM n Hewlett-Pckr to fcltte techncl stnrs to govern how commercl softwre shoul be wrtten. The ASP Inustry Consortum (ASPIC) n the Informton Technology Assocton of Amerc (ITAA) re workng closely to mke ASPs more stnrze. These strtegc moves shoul beneft ASPs compettveness n the future. Another fctor tht ffects the performnce of ASPs s the voltlty of emn, θ. A hghly t t voltle trnscton volume enhnces the ASP s compettve vntges: > 0, > 0, n θ θ t > 0. As we hve shown bove, the ASP s entry nto the mrket not only llows ts own θ clents to go on emn n therefore shft wy ther uncertnty rsks, but t lso helps users of the softwre to bte ther losses cuse by emn uncertnty. The ASP busness rrngement s pprecte by users to lrge extent becuse t offers hghly sclble systems.
Users cn promptly scle ther cpcty up or own t no tonl cost, becuse there s no fxe IT nvestment me on the users se. Moreover, when t becomes expensve to prove on-ste IT support servces ( lrge vlue of c), user frms wll fn the ASP soluton more ttrctve. As s shown by c [( ( p + t c) + ] C θ = + > 0, when users fce lrge c, they prefer to ( p + t c) u( p + t c) get ther IT support servces from n outse prover, rther thn nstll system n mntn t nternlly. In such stuton, the ASP s expecte to tke sgnfcnt mrket shre. 3.4 Robustness Check Although we mke some smplfyng ssumptons to reuce the complexty of the nlyses, our results re robust to these ssumptons. For exmple, we ssume tht the softwre s hghly customze, whle the ASP tens to come up wth much hgher unft costs becuse of ts one-to-mny structure. Such n ssumpton oes not mply tht the prouct must be perfectly ft system. In prctce, n n-house softwre system coul lso mpose lck-of-ft costs on users. All tht mtters for moelng purpose s the mrgnl sutlty for users between the soluton n the ASP. In fct, our moel resses the cse n whch the softwre hs unusully hgh unft costs,.e., t < 0. Ths les to the trvl outcome tht the ASP omntes the mrketplce. In ton, we ssume tht both venors use bsc prcng strteges: the ASP chrges fee per trnscton, n the venor requres n outrght purchse of the system. Amttely, there re numerous combntons of prcng strteges n the compettve envronment. Prctclly, nonlner prcng cn be consere. 9 Snce our objectve s to chrcterze the nture of the competton between the ASP n the venor, we focus on the most slent 9 An nlyss of the fxe fee + vrble chrge by the ASP cn be foun n the pper by Cheng n Koehler (00).
fferences between the two elvery chnnels. Assumng complex prcng strteges wll not scret our fnngs here. Our mn results cn lso be extene to the scenro n whch the user frm s trnscton voltlty s proportonl to ts expecte trnscton volume. For nstnce, we my ssume tht the ctul trnscton volume of user locte t s unformly strbute on [( θ ),( + θ ) ], wth θ <. In ths cse, most of our lemms n propostons stll hol. 0 4. Two-Stge Competton wth Uncertnty So fr, we hve nlyze mrket wth sttc unft costs (t). In prctce, however, these costs coul chnge urng n ASP s contrct pero, whch typclly lsts for 3 to 5 yers. Unft costs coul grow over tme gven softwre or hrwre chnges on the users se, or ecrese over tme ue to mprovements n the ASP s softwre n relte technology. For exmple, f the ASP uses browser nterfce tht s epenent on nonstnr spects of IE6 but busness crcumstnces fce by users rve emn for the ltest IE or Frefox, or f the ASP s nterfce nvolves moule bult on top of progrm tht only works n pre-vst MS Wnows envronment but hrwre replcement t the user s ste les to multple PCs wth the Vst OS, unft costs my ncrese. In such cses, n exstng ASP user fces hgher unft costs t n ts lter usge. On the other hn, f the ASP venor contnuously nvests n mprovng ts system ntegrton fetures, users unft costs my be ecresng over tme. For nstnce, Slesforce.com evelope n lunche AppExchnge n Jnury 006. AppExchnge s n onlne mrketplce for on-emn busness softwre. Currently t nclues over 50 pplctons, n Aobe, Skype, n Fctv re mong the vrous prtners. AppExchnge llows Slesforce.com n other ASPs to estblsh cross-pplcton ntegrton n therefore proves softwre users semless extenson of ther exstng systems (Cowley 005; Kuchnsks 006). In ths cse, users expect to hve reuce unft costs becuse unform pltform eses collborton cross pplctons. 0 The only excepton s Lemm, whch cn be proven uner the specl cse wth neglgble setup costs, but for the more generl cses, t gets too complcte mthemtclly. 3
To cpture how competton tkes plce n longer wnow wth possble chnges n unft costs over tme, we bul two-stge competton moel. In the frst stge, the venors choose ther prces ( p, P) smultneously. We ssume tht ther prces remn unchnge over tme. The ASP s pplcton mposes unft costs t n the frst stge. Users coul hve certn expecttons bout future chnge n unft costs. They my expect unft costs to ncrese f they ntcpte chnges n emn or hrwre upgres, or to ecrese f they ntcpte technologcl vnces tht fvor the shre softwre busness moel. In the secon stge, such chnge s relze. Users conser swtchng from the ASP s softwre to the softwre or the reverse. We mke two smplfyng ssumptons. Frst, users n venors weght utltes or profts obtne from both stges eqully. The tme scount fctor for stge two s. Secon, the ASP s ntl setup costs to serve new clent re neglgble;.e., S = 0. These two ssumptons help to ese the nlytcl nlyss wthout chngng the results qulttvely. Our stuy on the two-stge moel wll only focus on the scenro wth both venors coexstng. Scenros n whch one venor fls re of lttle nterest n ths context. 4. The Two-Stge Moel wth Increse Unft Costs Fgure 5 shows the two-stge competton wth n (expecte) ncrese t. In the frst stge, wth unft costs t, users n [ 0, ] choose the ASP n n [, ] use the softwre. In the secon stge, users fce chngng busness crcumstnces, n the unft costs of usng stnr ASP pplcton ncrese from the ntl t to t > t H. Exstng ASP users my conser swtchng to the soluton, whch proves customze softwre wth better ft. An exstng ASP user compres the utlty from the ASP, ( u p t ), n the utlty from the softwre, θ ( P. The mrgnl swtcher, enote by S, s u gven by P θ S = +. (9) p + t c u( p + t c) H H H 4
Fgure 5. Competton wth Increse Unft Costs S Expecte trnscton volume 0 Users choose the ASP ntlly n sty wth the ASP fterwr. Users choose the ASP ntlly n swtch to the softwre n the secon stge. Users choose the venor ntlly n sty wth the softwre fterwr. As shown n Fgure 5, the user, whch s nfferent between the two solutons n the frst stge, gns the sme totl utlty. If t chooses the ASP n swtches to the softwre lter on, ts totl utlty s θ {( u p t ) } + ( P. If t chooses the u softwre ntlly, ts totl utlty s θ θ ( P + (. Note tht u u once the user nstlls the system n the frst stge, t owns the system n only nees to py ongong mntennce costs n the lter stge. By equtng these two utltes, we get θ =. (0) u( p + t c) Equtons (9) n (0) mply tht the venor s prce P only ffects S, not. When the venor s prce ncreses, ll else beng equl, the number of users tht choose the n-house system n the frst stge remns unchnge. Ths s becuse the prce P s cncelle out n the nfferent user s utlty functon (equton (0)). In the secon stge, however, the number of swtchers (from the ASP to the soluton) ecreses s P ncreses. A hgh softwre prce, representng the lrge up-front costs of n nternl system, reuces exstng ASP users ncentve to move the softwre n-house. The venor gets [,] users n the frst stge n S, ] users n the secon stge. [ Its proft comes from users one-tme pyment. The optml prce s gven by Mx P [ ] = ( P C ). () S The ASP serves 0, ] users n the frst stge n 0, ] users n the secon stge. It gns [ [ S proft from users every use of the softwre. The optml prce s gven by 5
Mx p ASP =. () 0 0 S ( p c) xx + xx = ( p c) ( + S ) If users o not ntcpte future chnges n unft costs, ths reuces to one-stge sttc gme wth nvrnt unft costs t, whch we hve nlyze n Secton 3. Here, we tke such cse s the benchmrk to nvestgte how users expectton tht unft costs wll ncrese ffects the competton equlbrum outcome. Let p t, t H n P, t t H be the prces of the ASP n the venor when users expect unft costs to ncrese from t to t H, n let t, t H be the nfferent user efne by equton (0). Let p n t P be the prces when users beleve unft costs t t wll not vry, n let t be the nfferent user efne by equton (6) n tht nstnce. Proposton 8 When users ntcpte future ncrese n unft costs, t t H, both venors wll ncrese ther prces;.e., p > p n t, t H t P > P. More users wll choose the t, t H t softwre ntlly (.e., < ), n the ASP wll lose exstng clents to the venor once the ncrese occurs. <Proof> See the Appenx. t, t t H Proposton 8 sttes three mportnt fnngs. Frst, lthough ncrese unft costs mply ecrese n prouct qulty of t, t H t u t, the ASP shoul nevertheless ncrese ts prce: p > p. By chrgng hgh prce, the ASP gves up competng for hgh-volume users wth the venor; t nste concentrtes on explotng low-volume users tht re unble to ffor the soluton nywy. Secon, the venor lso rses ts prce, whch s ntutve becuse ts prouct becomes more ttrctve. Interestngly, we fn tht the The ASP s prce cnnot excee the upper boun of u t H. Otherwse low-en users hve no ncentve to use the ASP s prouct, n they wll opt to sty out of the mrket. 6
seller s best response functon ( p + th c C θ P = +, see the proof of Proposton 8 u n the Appenx) s the sme s tht n sttc competton wth unft costs t = t (equton (7)). H Ths mens tht the venor n prctce wll opt smple prcng strtegy n prce ts softwre s f t were n one-stge competton wth nvrnt unft costs. Fnlly, we conclue tht belef tht unft costs wll ncrese benefts n-house soluton provers but hurts ASPs. 4. The Two-Stge Moel wth Decrese Unft Costs costs Fgure 5b shows the two-stge competton when t ecreses. In the frst stge, wth unft t, users n [ 0, ] choose the ASP n n [ ], buy from the venor. In the secon stge, the unft costs of usng the ASP pplcton eclne to t < t L. Such chnge coul be the result of vnces n web technology, opton of softwre stnrs n protocols, or the ASP s efforts to crete unform softwre pltform. Fgure 5b. Competton wth Decrese Unft Costs S Expecte trnscton volume 0 Users choose the ASP ntlly n sty wth the ASP fterwr. Users choose the softwre ntlly n swtch to the ASP n the secon stge. Users choose the softwre ntlly n sty wth t fterwr. When unft costs ecrese n the secon stge, n exstng softwre user compres the utlty vlble from the ASP prouct, ( u p t ), n the utlty from the L softwre, θ (. Note tht becuse the n-house system hs lrey been nstlle, u the one-tme pyment to the venor P hs been sunk. Ths user only nees to py ongong mntennce costs. The mrgnl swtcher, enote by S, s gven by S θ =. (3) u( p + t c) L 7
As Fgure 5b shows, the user s nfferent between the two solutons t the frst stge. If t chooses the n then swtches to the ASP lter on, ts totl utlty s θ ( P + L ) u {( u p t }. If t chooses the ASP ntlly, ts totl utlty s {( u p t ) } + {( u p t } L ). Hence, we get = p P θ +. (4) + t c u( p + t c) Equtons (3) n (4) show tht the prce P only ffects but not S. When the prce ncreses, ll else equl, the number of users tht choose the n-house system n the frst stge ecreses. However, the number of swtchers (from the to ASP, n the secon stge) s not ffecte by the prce ncrese. A hgh prce for the softwre oes not gve exstng users greter ncentve to swtch to on-emn softwre becuse t s consere sunk cost. The venor n the ASP choose the followng proft-mxmzng prces, respectvely: Mx P [ ] = ( P C ) ; (5) Mx p ASP =. (6) 0 0 S ( p c) xx + xx = ( p c) ( + S ) We tke the sttc competton wth nvrnt unft costs t s the benchmrk cse. Let p t, t L n P, t t L be the prces of the ASP n proucts when users expect unft costs to ecrese from t to t L, n let n t, t L be the nfferent user efne by equton (4). Let P be the prces when users beleve unft costs t t wll not vry, n let nfferent user efne by equton (6). t be the Proposton 9 When users ntcpte ecrese n unft costs, t tl, both venors wll p t reuce ther prce;.e., p < p, t, t L t P < P. More users wll choose the ASP soluton t, t L t ntlly;.e., >. Exstng clents of the venor hve lttle ncentve to swtch to t, t t L 8
on-emn softwre even f unft costs ecrese, but they my o so f trnscton voltlty s hgh. <Proof> See the Appenx. Proposton 9 sttes three mportnt fnngs. Frst, the ASP s response to users ntcpton of ecrese unft costs s to reuce ts prce, espte n ncrese n the qulty of ts prouct, n orer to compete for the more proftble hgh-volume users. Secon, the venor once gn stcks wth smple prcng strtegy. Its best response functon p t c C c u c ( + ( ) P = θ +, see the proof of Proposton 9 n the Appenx) s the sme u s tht n sttc competton wth unft costs t = t (equton (7)). The venor therefore cn just gnore the expecte chnge n unft costs on the users se n prce the softwre s f t were n one-stge competton. Thr, our nlyss shows tht exstng users of softwre re unlkely to move to on-emn softwre. These users hve two choces: sty wth the soluton, wth utlty of θ (, or swtch, wth utlty of ( u p tl ). Snce u p > c lwys, the user swtches only f ts trnscton voltlty (θ ) s hgh. Hence, we conclue tht once n n-house system hs been nstlle, users hve lttle ncentve to swtch to n ASP unless they nee to mnge hgh cpcty rsks. 5 Dscusson n Concluson We exmne frms choces between the ASP n solutons when they fce stochstc emn for the softwre. Our fnngs show tht the ASP on-emn moel s superor when the user fces low trnscton volume, hgh trnscton voltlty, or expensve n-house IT servces, or when the user s unft costs from usng the ASP s stnrze prouct re low. Uner bro mrket contons, the ASP n busness moels wll coexst, so t s lkely tht the Snce the ASP gets p bse per trnscton, users wth hgh trnscton volumes re consere more proftble. 9
softwre nustry wll not be omnte by just one. The ASP s entry, t the expense of the trtonl venor, my beneft every user frm. It offers smll n meum frms costsvng ccess to softwre, compettvely reuces lrge frms mplementton costs, n mtgtes ech frm s trnscton uncertnty rsks. Although the ASP moel hs severl vntges, ts long-run vblty lrgely epens on the mgntue of users unft costs. As such costs ncrese, the ASP s blty to compete wth the venor monotonclly ecreses. Ths mportnt fnng mples tht the ASP moel requres funmentlly new set of proucts n technology rchtectures. To cheve profts, the new generton of ASPs shoul leverge n open lnguge, proper moulr structure, n loosely couple nterfce wth other pplctons, n thus offer proucts wth low unft costs. Toy s ASPs re evelopng n elverng ther own pplctons usng pure Web rchtectures, n stnr XML lnguge, n n screte unts, whch wll gretly ncrese ther chnce of success. Our work proves theoretcl bse for these strtegc moves. ASPs re ctvely nvestng n reucng users unft costs. It s nterestng n necessry to nlyze the prce competton over longer horzon n whch users unft costs my chnge over tme. Our fnngs suggest tht nvestments n mprovng ASP softwre qulty shoul be ccompne wth ownwr prce justment. In the future, we expect to see tht ASPs offer better proucts t lower prces. On the other hn, venors shoul lwys opt smple prcng strtegy: they shoul prce ther softwre s f they were nvolve n sttc competton wth nvrnt unft costs. Hence, we prect tht no prce chnge wll be observe for trtonl pckge softwre. In ton, our nlyss shows tht exstng softwre users swtch to the ASP soluton only to elmnte cpcty, or busness volume rsks rther thn to reuce opertng costs. These results hve rect mplctons for the kn of busnesses tht ASP wll trget n the future. There re mny possbltes for further ASP stues. It woul be nterestng to exmne the role of the Servce Level Agreement (SLA). When users hve emn for fferent levels of 30
IT servce, the SLA consttutes wy to segment the mrket n mprove the ASP s proft, n mybe to ncrese users surpluses s well. Another possble reserch topc s the softwre venor s choce. Some mjor softwre compnes hve chnge ther busness moels to operte s n ASP s well. For exmple, Orcle now hs more thn 400 customers usng ts ASP proucts n growng shre of ts revenues from ts ASP unt. It s nterestng to see tht mny softwre provers re ong busness both wys: sellng sophstcte verson s prouct n lesng smplfe verson s n ASP. Future reserch my nvestgte whether softwre venor shoul the ASP s n tonl chnnel, swtch to t exclusvely, or eschew tht pproch entrely. 3
References Benrz, A., Mnufcturers Eye on On-Demn Softwre, Network Worl, Apr 4, 006 Cheng, H.K. n Koehler, G.J., Optml Prcng Polces of Web-enble Applcton Servces, workng pper, 00 Chouhry, V., Tomk, K., n Chturve, A., Economcs Benefts of Softwre Rentng, Journl of Orgnztonl Computng n Electronc Commerce 8, 77-305, 998 Cowley, S., Slesforce.com Mkes Pltform Move wth AppExchnge, InfoWorl, Sep 005 Dewn, S., n Menelson, H., User Dely Costs n Internl Prcng for Servce Fclty, Mngement Scence, Vol. 36, No., Dec. 990 Dunn, D., Systems Mngement on Cll, Informton Week, Jn 0, 005 Dzubeck, F., XML n Mngement Tem up, Network Worl, Vol., Iss. 7, Jul 004 Fshburn, P., Olyzko, A., n Sers, R., Fxe Fee versus Unt Prcng for Informton Goos: Competton, Equlbr, n Prce Wrs, Frst Mony (7), 997 Grner, R., Softwre for Hre, CRN, Nov, 004 Hmm, S., SAP Gets On-Demn Relgon, Busness Week, Feb, 006 Kpln, J., Sortng Through Softwre As A Servce, Network Worl, Nov, 005 Knorr, E., Softwre on Demn: the En of IT s We Know It? InfoWorl, Vol. 6, Iss. 48, Nov 9, 004 Kuchnsks, S., Slesforce Fnlly Shps AppExchnge, Ecommerce, Jn 7, 006 Lcy, S., The On-Demn Softwre Scrum, Busness Week, Aprl 7, 006 M, D., prvte communctons wth the representtves of MePACS (phone: 6-367-08), 005 Mrtn, C., Fee-per-Stuy Approch Cn Cut Strt-up Costs, Rology Toy, Vol.5, No. 9, Sep 004 Murphy, D., As Esy s ASP, Drect Response. Lonon, Jn 005 Nccol, J., Gtes Memo Puts Servces t the Hert of Mcrosoft, Network Worl, Nov 005 3
Proponents of Softwre s servce sy t wll wpe out trtonl softwre, The Economst, Apr, 006, rtcle vlble t http://globltechforum.eu.com/nex.sp?lyout=rch_story&oc_=8439&ttle=unversl+serv ce%3&ctegory=4&chnnel=3 Plltto, J., IBM Recrutng ISVs, Prtners to SAAS, Chnnel Inser, Feb 3, 006 Rcel, A., Does Slesforce.com Hve the Chops to Become Bllon-Dollr Compny? Informton Week, My 006 Sngh, C., Shelor, R., Jng, J., n Klen, G., Rentl Softwre Vluton n IT Investment Decsons, Decson Support Systems, Vol. 38, Iss., Oct 004 Sunrrjn, A., Nonlner Prcng of Informton Goos, Mngement Scence 50 (), 004 Susrl, A., Bru, A., n Whnston, A., Unerstnng the Servce Component of Applcton Servce Provson: An Emprcl Anlyss of Stsfcton wth ASP Servces, MIS Qurterly 7(), 003 Thboeu, P., Tech Stnrs Set Tone for On-Demn Systems, Computer Worl, Vol. 38, Iss. 44, Nov, 004 Vr, V., Web Servces Fce Relblty Chllenges, Wll Street Journl, Feb 3, 006 Vrn, H., Buyng, Sellng, n Rentng Informton Goos, Journl of Inustrl Economcs XLVIII, 473-488, 000 Zmmermn, J., The Costs n Benefts of Cost Alloctons, The Accountng Revew, Vol. 54, No. 3., pp. 504-5, Jul., 979 33
Appenx Proofs for Lemms n Propostons Prove Proposton : users wth low trnscton volumes ( < θ ) re not serve by the venor. We prove by contrcton. Suppose tht the venor oes serve some users wth low trnscton volumes. Let, be the mrgnl users n [θ, ] n [0, θ ] respectvely. Frst, we h l exmne the extreme cse tht users n [0,θ ] fce etermnstc trnscton volume. The venor s proft-mxmzton problem s Mx P C )( + θ ), where P ( h l l P = n ( h cθ 0 C θ cθ = l +. Solvng t gves l = + + > θ, whch u 4 ( 4 4u les to contrcton. Next, we look t the generl strbuton cse. For users n [0, θ ], we mke no ssumpton on the strbuton of ther trnscton volumes. Denote user s utlty loss ue to trnscton uncertnty by f (δ ). It s esy to get P = ( l f ( δ ) n = h l cθ f ( δ ) +. The softwre venor s problem s u ( Mx l cθ f ( δ ) ( C f ( δ )) + θ + l l u (. The proft functon s concve, n the frst-orer conton s ( cθ = ( 4( l + ( θ + C + 3 f ( δ ). The u l terms n the prenthess re just the frst-orer conton when the trnscton volume s etermnstc. Hence, ths frst-orer conton s postve t 0 l. Mkng use of the concvty, we conclue tht the optml soluton > 0 > θ, whch s gn contrcton. Q.E.D. l l <Proof of Proposton > Denote by ( p ) P the prce pr t the ntersecton pont of equtons, (7) n (8). Conser the bounry contons for the venors prces: the venor s prce hs lower boun P, n the ASP s prce hs n upper boun u t. The ASP C p 34
must offer prce tht gves users enough ncentve to use t. A prce lrger thn the upper boun u t wll ttrct no users. On the other hn, the ASP s prce must be hgh enough to cover the ASP s ntl setup cost to gve the ASP n ncentve to serve the mrket. The prce pr ( P p ) s n equlbrum outcome f t stsfes ll these bounry contons. In the specl cse wth C PS = 0, p, S= 0 = mx C, + t, t + c u S=0, we get θ ( ), whch s fesble, outcome f u c p t + c u t t t. It s esy to show tht > 0 s= 0. In the generl cse S wth S 0, the ASP s prce s = c + t + ( S), where ( S ) 0, > 0 vlue t must stsfy t p u c u = ( S), n c ( S ) S t s ecresng n S.. Thus, the crtcl Compre the venor s best response functon n the ul mrket (equton (7)) wth ts optml prce M P n the pre-asp mrket (equton (4)). Snce p < u t, the venor s prce ecreses fter the ASP enters the mrket: mrket: M P < P Pluggng equton (7) nto equton (6) gves the locton of the nfferent user n the ul = + C cθ ( ( p + t c) u( p + t c) +.. (A) M Compre ths wth the mrgnl user s locton n the pre-asp mrket (, equton (5)). We fn tht M > : the venor s mrket shre ecreses. Together, lower prce n ecrese mrket shre result n less proft for the venor. Fgures A() n (b) show the pre-asp n ul mrkets respectvely. When the ASP s n opton, full mrket coverge s cheve. Smll users wth low trnscton volumes, locte n M [, ] 0, re ble to ffor the softwre through the ASP, becuse the ASP oes not mpose hgh mplementton n mntennce costs on them. Users n [, ] M wll purchse the 35
softwre n the pre-asp mrket but choose to use the ASP n the ul mrket. For these users, the ASP represents cheper soluton thn the opton. Snce M E [ u ( P P )] E[ u ( P = P )] E[ u ASP ( p = p )] users n [,] =, they re better off s well. Fnlly, re frms wth hgh volume of trnsctons. They choose the soluton n both mrkets, but ther pyments re lower n the ul mrket becuse of the ASP s competton. The ASP s entry nto the mrket benefts these frms s well, lthough they re not ts clents. Ech user therefore s better off, n the totl consumer surplus s strctly mprove. Fgure A. The Pre-ASP Mrket n the Dul Mrkets When () The pre-asp mrket (the softwre soluton lone) t t users purchse the softwre t M P 0 users re prce out of the mrket M (b) The ul mrket users purchse the M softwre t P < P 0 users choose the ASP M <Proof of Proposton 3> When Q.E.D. t > t, the ASP s unrestrcte optml prce (from equton (8)) s p > u t. Ths prce s too hgh to ttrct ny user. The ASP wll set ts prce t the restrcte optml '' p = u t. The ASP s users therefore get zero surplus. Plug '' p nto equton (7). The venor s optml sellng prce, enote by ( P M, gven by equton (4)). Pluggng ( u '' p n " P, s the sme s n the pre-asp mrket " P nto equton (6) gves the venor s " C c mrket shre, θ =, whch gn s the sme s n the pre-asp mrket 36
M (, gven by equton (5)). The concluson tht the venor s proft oes not chnge follows. When the vlue of t ncreses further, n especlly when t > u c, t s esy to see tht the ASP s not ble to operte proftbly. Q.E.D. <Proof of Lemm > Usng equton (8), we get B P = u +. θ θ p θ B = [ 4Sut], where θ ( B Su) To juge the sgn of p B θ, conser two possble cses: () < 0 <=> p θ > 0 θ, B n () > 0 <=> p θ < 0. If the frst cse s true, B < 0 θ θ mples P <. θ u However, equton (7) nctes tht P = θ p >. Hence, we elmnte θ u u B P the possblty of cse (). Conser the secon cse. If > 0, ether > 0 θ θ or P p < < 0. On the other hn, equton (8) mples < 0, whch mens tht u θ θ P <. It therefore must be the cse tht p < 0 θ u θ, P < θ 0, n c ( < P θ <. Ths proves prts () n (). Use equton (A): u u C = θ 4( p p u p θ + + t c) 4u ( p p + t c θ θ > 0. Ths proves prt (3). Q.E.D. + t c) <Proof of Proposton 5> Frst we conser the cse of t = t. We cn show tht the prce pr P, p ) = ( C, p ) s the equlbrum. Note tht wth ths prce pr, the whole mrket s ( 37
serve by the ASP lone: ( P = C, p = p ) =. Gven the venor s prce C, the ASP hs no ncentve to evte, becuse the prce response functon (equton (8)). Ths prce p s obtne usng the ASP s best p cn gve the ASP postve proft, becuse t s hgh enough to cover the ASP s setup costs. On the other hn, gven the ASP s prce p, the venor hs no ncentve to evte s well: prce ncrese wll not ttrct ny user, n prce ecrese wll not brng postve proft. Thus we conclue tht the prce pr P, p ) = ( C, p ) s the Nsh equlbrum, n the ASP uses the prce p to serve the ( whole mrket when t = t. For cses wth t < t, smlr nlyss pples. Snce ecresng ts prce wll not brng the ASP more users, the ASP wll keep ts prce t p. Q.E.D. <Proof of Proposton 6> As shown n the proof of Proposton 3, when t ncreses, the ASP s competton becomes nfr-mrgnl. The venor wll set the prce t the monopoly prce C ( u c) (the sme s n the pre-asp mrket): ( c P = + θ. The ASP s best response u prce s boune by the upper boun u-t. In ths cse, the venor wll tke the users locte n C + cθ +, ( u. When t t, t s not proftble for the ASP to tke the resul users, snce Π ASP = C + cθ + ( p c) + ( u ( C cθ + S 0, u u t. Q.E.D. p <Proof of Proposton 8> The optmzton problems for the venor n the ASP re gven by equtons () n (). The best response functons (BRFs) re p + th c C θ P = + (for the u c ( θ c + t p venor) n + ( t + ) = 0 3 H c p (for the ASP). The proof of u ( p + t c) 38
Proposton shows tht n one-stge sttc competton wth nvrnt unft costs t n neglgble ASP setup costs ( S = 0), the ASP s prce s p = c + t. It s strghtforwr to check the ASP s BRF n conclue tht ny prce smller thn c + t s not optml. In fct, we t cn show tht the optml prce must be n the rnge of c + t < p t, t H < c + t H. Snce the ASP s prce ncreses, the venor s prce wll ncrese s well (usng the venor s BRF);.e., P > P. Pluggng the two prces nto equton (0) gves t, t t H t, t H. It s esy to show tht t, t t H <, where t s gven by equton (A). Q.E.D. <Proof of Proposton 9> The optmzton problems for the venor n the ASP re p gven by (5) n (6). The best response functons (BRFs) re + t c C P = θ + u c ( θ c + t L p (for the venor) n + ( t + c p ) = 0 3 (for the ASP). u ( p + t L c) Check the ASP s BRF. It s strghtforwr to see tht the optml prce must stsfy c + t < p < c + t = p. Snce the ASP s prce ecreses, the venor s prce wll L t, t t L lso ecrese (usng the venor s BRF);.e., (4) n compre to equton (A); the concluson P < P. Plug these prces nto equton t t, t L, t t L > follows. t Q.E.D. Ths bounry conton my be reche when the unft cost t s very smll. As t ecreses, the prce from equton (8) lso ecreses. It my rech level tht s not hgh enough to mke the ASP proftble, especlly when the setup cost S s lrge. In ths cse, the ASP s best response prce wll not be the prce gven by equton (8), whch s nlyze n Proposton 6. When the ASP serves the whole mrket, the mnmum prce requre to cover the ASP s setup costs s S + c. Snce S << C (specfclly, S < C ), the ASP s prce stsfes p > S + c. 39