Online Cashback Pricing: A New Affiliate Strategy for E-Business

Transcription

1 Online Cashback Pricing: A New Affiliate Strategy for E-Business Comleted Research Paer Yi-Chun Chad Ho University of Washington Seattle, WA chadho@uw.edu Yi-Jen Ian Ho University of California, Irvine Irvine, CA 9697 hoy1@uci.edu Yong Tan University of Washington Seattle, WA ytan@uw.edu Abstract This aer examines the imact of the cashback mechanism on online merchants affiliate and ricing strategies. Through reimbursing a ortion of the transactional amount to consumers in a form of cashback, merchants are able to ractice seconddegree rice discrimination. We develo an analytical framework which exlicitly considers the imlementation cost associated with the underlying romotional vehicle. We first identify the conditions under which affiliate strategy is rofitable. Surrisingly, the romotional low rice could be actually high, relative to the uniform rice when cashback is absent. We also roose channel coordination as a remedy to mitigate market inefficiency caused by double marginalization. Finally, we extend our model to a duooly setting and find that a merchant can benefit from its rival s move into the cashback market. Under some conditions both merchants have no incentive to move alone but refer its rival to do so. Keywords: cashback, online advertising, romotion, rice discrimination, duooly Thirty Third International Conference on Information Systems, Orlando 01 1

2 E-Business and Cometitive Strategy Introduction The rise of the Internet and the surging oularity of online shoing have offered raid growth in e- commerce and garnished comanies interest around adating best digital marketing strategies. Secifically, affiliate marketing, an Internet-based advertising where a business ays the affiliates for every visitor or sale brought in by the affiliates own effort, has become a revalent strategy for online businesses to boost sales volume at low costs (Swan 011b). In the early days of e-commerce, comanies relied on web traffic to establish oularity; now the attention has turned to converting such traffic into actual urchases. Table 1 shows the breakdown of affiliate tye among to 0 sales-generating websites in the United Kingdom from 006 to 01 (Swan 011a). The statistics highlight a dynamic shift in online advertising ractice, moving from ordinary methods such as ay-er-click (PPC) to the novel cashback affiliate. Over the years, cashback affiliate model which incentivizes consumers to urchase by reimbursing them with a ortion of transactional amount has received substantial accetance among online merchants 1 due to its caability of converting traffic into sales. Such higher conversion rates stem from an interesting online shoing behavior noted in Forbes: One good way to find deals is to find the cheaest rice and then check at Ebates.com to see if there s a cashback offered for the merchant you found (Rand 005). Table 1. Breakdown of Affiliate Tye among To 0 Sales-generating Websites Affiliate Method PPC Couon code Cashback Content/Others Websites built simly uon the cashback concet, such as Ebates.com and MrRebates.com, are extremely successful. Ebates, the leading cashback site in the U.S. with 1 million registered users, has reimbursed over 85 million dollars to its members since In 011 it brokered 900 million dollars in merchandise sales for its 1,600 affiliated merchants. Its revenue growth has trended 50 ercent higher for the second year in a row since 010 (Hoge 011). Interestingly, cashback sites are not the only ones trying to exloit this new marketing concet. Software giant Microsoft in 008 imlemented the cashback feature that allows its search engine Bing to act as a cashback ublisher. One year later, Google also introduced its Google Checkout as a latform on rewarding customers. There is every reason to believe the cashback concet will continue to grow its oularity. Major consumer banks in the U.S. gradually roll out cashback feature to their online shoing channels, such as Ultimate Reward Mall (Chase Bank), ThankYou Bonus Center (Citibank), and Add It U Program (Bank of America). Bank of America in August 01 further leveraged the cashback concet by launching BankAmeriDeals, an innovative rogram that allows consumers to earn cashback from shoing at hysical stores. It has become clear that cashback concet is a catalyst for collaboration between merchants and their affiliate artners, and the enabler for a level of interaction between customers and merchants that has not been ossible until now. 1 By online merchants we mean manufacturers and retailers in both click-and-mortar and online-only business models. Credit card issuers commonly reward cash back to card holders when they make ayments by cards; however it is different from the cashback as discussed in this aer in two asects. First, the former merely incentivizes card holders to use the cards whereas the latter further allows merchants to rice-discriminate among consumers. Second, card issuers run the reward rograms and dictate cash back ercentages without taking merchants interest into account. In our cashback model, however, merchants will decide whether to affiliate with intermediaries, and that given an affiliation is formed, cashback rates are then determined through a rocess in which both merchants and intermediaries are involved. Thirty Fourth International Conference on Information Systems, Milan 013

3 Ho et al. / Online Cashback Pricing Practice and Problem The cashback affiliate is a novel marketing aroach featuring both online advertising and digital romotion. Once affiliating with a cashback intermediary, a merchant (advertiser) 3 can ost an affiliate link on the ublisher s site. This hyerlink re-directs consumers to the merchant s online storefront where urchasing transactions are taken lace. The cashback intermediary (ublisher), as a transaction broker, collects a commission from a merchant when consumers make transactions through the referral link. It then entices consumers into urchasing by rewarding them with a redetermined ortion of the transactional amount, also known as cashback. 4 This monetary incentive to consumers makes the cashback affiliate one of the most effective advertising aroaches. The cashback model is also a romotional vehicle which allows the merchant to ursue segmentation by rice-discriminating among consumers. Products can be listed for one rice for non-cashback users and a lower rice for the cashback users at the same time. In ractice, cashback holds advantages over others alternatives such as couons and mail-in rebates in two asects. First, from a consumer s ersective, this concet is straightforward. Searching for couons or couon codes could be time-consuming and the romotions are redeemable under some terms and conditions only (e.g., rior to exiration date). The redemtion cost associated with mail-in rebates is state-deendent and is thus uncertain to consumers (Lu and Moorthy 007). Cashback on the contrary has the caability of roviding constant and certain discount. Second, from a merchant s ersective, the cashback ractice is simle yet efficient. Comanies seek outside marketing solution like affiliate marketing because they are struggling to reach desired consumer segments by themselves. Take couon for examle. According to NCH Marketing Services, in 01, couon redemtion rate is as low as 3.5% in the United States. This low figure exhibits the inefficiency concern about traditional mass-media romotion. Cashback sites, with a huge loyal user base, rovide an efficient solution for targeting romotion. Desite its revalence in the ractice, the cashback affiliate model has received almost no attention from researchers. Motivated by the lack of theoretical examination on this still-nascent marketing aroach, we develo an analytical framework to fully understand the strategic use of the cashback affiliate model. In this aer, we are interested in answering the following questions: When should a merchant adot the cashback model? Given a merchant affiliates with the ublisher, how should they set their resective ricing terms? How would the introduction of cashback imact consumer surlus and social welfare? Literature Review Literature on rice discrimination rovides the theoretical fundamental for our research framework. Price discrimination is only feasible when three conditions are met: (i) sellers have market ower, (ii) arbitrage is infeasible, and (iii) buyers differ in their demands for a good or service and they can be segmented either directly or indirectly (Stole 007; Varian 1989). Several analytical models are develoed to examine the rationale of rice disersion (Salo and Stiglitz 198; Shilony 1977; Varian 1980). Price discrimination is commonly thought as a weaon for firms to make extra rofit. 5 However, based on exeriences with several major firms, Neslin and Shoemaker (1983) argue that most managers do not have good methods 3 While ractitioners call businesses who have adoted advertising models advertisers, it is more aroriate to use a more general term merchants instead of advertisers since in this aer we examine whether businesses should adot the underlying model. 4 To earn cashback, a consumer needs to: (1) register for membershi at a cashback site, () click on a merchant s referral link which will direct the shoer s browser to the merchant s own website, and (3) make a urchase. Accumulated cashback can be claimed via checks or Payal online transfers. Some sites (e.g., MrRebates.com) rovide rice-comarison feature and merchant-secific couon codes which allow consumers to obtain a bigger discount. 5 Strictly seaking, two more remises are required to make this argument valid. First, the merchant s segmentation strategy should be effective, i.e. charging right rices to right tyes of consumers. Second, and erhas the more imortantly, the revenue gain of doing such ractice should be greater than the cost incurred. Most of the extant analyses on rice discrimination neglect these two imortant factors. Thirty Fourth International Conference on Information Systems, Milan 013 3

4 E-Business and Cometitive Strategy for comuting the overall rofitability of a roosed romotion rogram. Anderson and Dana (009) develo a monooly ricing model and characterize the conditions under which rice discrimination is rofitable. The stream of research investigating romotional vehicles in monooly is most related to this work. Narasimhan (1984) demonstrates that couon users are different from nonusers. Since then, a variety of romotional methods are develoed by marketers and investigated by researchers. For examle, couons take various formats such as direct mail couons (e.g., Bawa and Shoemaker 1987), newsaer couons (e.g., Neslin 1990), ackage couons (e.g., Raju et al. 1994), cross-ruff couons (e.g., Dhar and Raju 1998), and digitalized couons (Oliver and Shor 003). Researchers are also interested in key factors that influence a firm s choice among different alternatives. For examle, a romotional vehicle s exiration date (long- vs. short-duration) and redemtion timing (front-loaded vs. rear-loaded) also determines its robability (Inman and McAlister 1994; Krishna and Zhang 1999; Zhang et al. 000). Promotional vehicles are also used to meet different business objectives. Gerstner and Hess (1991) argue that the manufacturer can motivate retail articiation by using rebates. Firms can also incororate romotion into ricing strategy for better customer retention (Shin and Sudhir 010). This aer is also related to ones focusing on the role of romotional methods in an oligoolistic setting. Shaffer and Zhang (1995) consider a market in which two cometing firms can distribute couons either to targeted consumers or via mass media. They demonstrate that couon targeting leads to a risoner s dilemma which makes both firms worse off. Corts (1998) find similar result by demonstrating that cometing firms may wish to refrain from rice discrimination. A relevant search question is also exlored in a more general setting with two asymmetric firms. Lal (1990) and Rao (1991) conclude that in equilibrium national brands should romote to mitigate encroachment by a rivate label. Dogan et al. (010), on the other hand, find that if one firm has absolute cometitive advantage over its rival, in the equilibrium the disadvantaged firm would offer rebates alone. Although the volume of extant research on romotion is vast, our work contributes to the literature in the following asects. First, to the best of our knowledge, this aer is the first one directly studying the stillnascent cashback model and examining the strategic use of such unique romotional method. Second, we exlicitly consider the imlementation costs associated with the underlying model, which are treated exogenous in most, if not all, of rior literature on romotion methods. Third, our stylized model endogenizes the rices of advertised roducts as a decision variable of merchants rofit function, which are commonly assumed to be given exogenously in the advertising literature. 6 These analytical advantages rovide a sharer and more conclusive insight into the merchant s best affiliate and ricing strategy. Basic Model In this section, we first introduce the consumer resonse, consumer segments, and the merchant s ricing alternatives, when the cashback affiliate is absent. The reliminaries develoed from non-cashback ricing serve as a benchmark for our analysis of the cashback mechanism. Next, we set u the cashback ricing model and derive otimal ricing decisions at equilibrium. Non-Cashback Pricing A Benchmark Consumer resonse. In the sirit of Salo s model (1979), we assume that consumers are uniformly distributed on a reference line. 7 The location of a consumer identifies the ideal bundle of roduct attributes she refers. The reservation rice a consumer would like to ay for her ideal roduct is V. A roduct at distance x away from the consumers generates a utility of V tx, where t measures consumers sensitivity to horizontal roduct reference. After accounting for the rice effect, consumers with location x away from the roduct derive a transaction utility of U(x) = V tx. Each consumer has a unitary demand and will buy the roduct if the transaction utility is non-negative. As a result, the demand can be 6 Few excetions are Dellarocas (01), Chen and He (011), and Feng and Xie (007) 7 Since our research interest is in merchants ricing strategy rather than roduct ositioning, Salo s model allows us to ignore firm s location decisions, which is a common concern in IS satial model literature (see Dewan et al 003 for examle). 4 Thirty Fourth International Conference on Information Systems, Milan 013

5 Ho et al. / Online Cashback Pricing exressed by Q( ) ( V ) / t. Consumer segments. Consider a market with total number of consumers normalized to one. The market is comosed of two tyes of consumers, l and h, with fraction θ and 1θ, resectively. For a given roduct, tye-h consumers have a reservation rice v, whereas tye-l consumers have a lower reservation rice δv, where δ (0, 1) 8. When tye-l s relative valuation is low, (i.e., δ is small) we say the valuation difference between two consumer tyes is salient. Given the model setting, the demand generated from tye-l and tye-h segments are Q ( ) ( v ) / t and Q ( ) (1 ) ( v ) / t, resectively. Throughout this l h aer we consider the market diverse enough and hence exclude the case where market is fully covered. 9 The market configuration, measured by (θ, δ), lays an imortant role in determining firms ricing strategy, as we shall see shortly. Figure 1. Consumer Segments and Resonse Merchant s ricing roblem. Consider a market served by a monoolist. With the simlest ricing scheme, the merchant may charge a uniform rice u, to the entire market and face a simle ricing roblem given by: U max Q ( ) Q ( ). (1) u m u l u u h u Maximizing Equation (1) we have the otimal uniform rice, u v, if (, ) RH ; v 1, otherwise, 1 1 where R H (, ) 0,. (1 ) The otimal uniform rice is deendent of market configuration. When market arameters (, ) fall in the region (, ) RH R H, the merchant would set the rice at v / and serve tye-h consumers only. When, it has incentive to serve both segments and charge one half of the total market valuation. Now assume the monoolist can accurately identify consumers tye with no cost. The merchant may want to rice-discriminate consumers by charging two asymmetric rices: a low rice, to the low tye and a high rice h, to the high tye. As a result, the merchant s rofit maximization roblem with discriminating ricing is given by: D max Q ( ) Q ( ), () l, h m l l l h h h The asymmetric rices allow the merchant to searately maximize its rofit in two different segments. l 8 The setting of asymmetric reservation rice is widely adoted in marketing and economic literature on rice discrimination. See Lu and Moorthy (007) and Desai (001) for examle. 9 We have such model restriction for two reasons. First, since the main focus of this aer is the effect of rice on roduct sales, we find it unrealistic to assume that roduct sales remains the same (market is covered) when rice changes. Second, when market is covered, the ublisher s decision could be restrained to the corner solution. Thirty Fourth International Conference on Information Systems, Milan 013 5

6 E-Business and Cometitive Strategy Maximizing Equation () gives the otimal asymmetric rices l v/ and h v/. Comaring ricing terms under two different schemes yields the following observations. LEMMA 1. (i) The otimal ricing terms follow the attern: ; (ii) The ability to rice l u h D U discriminate makes the merchant better off, i.e. m m. All inequalities hold regardless of market configuration. In the absence of cashback, when a firm switches from uniform to discriminating ricing, it lowers the rice for tye-l segment while raises the rice for the other. In other words, the uniform rice is bound between two asymmetric rices. In addition, discriminating ricing is always referred if a merchant can identify consumers tye without any cost. Given these observations, we ask: Do these arguments still hold in the resence of cashback model? Cashback Model Since reservation rice is asymmetric across segments, the merchant has incentive to charge different rices to different segments. However, it cannot not directly distinguish different tyes of consumers. Cashback, a mechanism which allows consumers to self-select themselves into correct tyes, can serve as a rice discrimination device. Although the majority of online advertising literature assumes the rice of advertised roducts to be exogenous, we think it is more realistic to endogenize the rice decision into firms rofit maximization roblem. In our analytical framework, the merchant first decides whether or not to adot cashback affiliate and subsequently sets the rices. For exosition urose, our analysis roceeds in the reverse direction. We first analyze firm s otimal rices given the affiliation is formed. Then we examine whether the merchant should adot the affiliate model at all. A merchant who adots cashback affiliate is actually oerating an electronic dual-channel. When a consumer desires to buy a certain roduct, she could urchase it directly from the merchant s e-commerce website (direct channel 10 ) or via the affiliate link on a cashback site (cashback channel). Of course, cashback shoing is not costless to consumers. Cashback shoers incur transaction costs that include the disutility derived from extra works throughout cashback shoing rocess such as registering on the ublisher, searching for the merchants, clicking through affiliate links, and waiting for rewards to be redeemable 11. Nevertheless, a consumer would still choose to sho through the cashback channel if the monetary incentive obtained from cashback is greater than the transaction costs. In the case where transaction costs are erceived to be higher, the consumer is assumed to make the urchase via the direct channel. Denoting the transaction costs incurred from cashback shoing by c i (i = l, h), and since the values of time are different across segments, we assume that the costs are lower for tye-l consumers, i.e. cl ch. The assumtion that transaction cost is ositively correlated with the reservation rice is widely acceted in the rior literature (Coughlan and Soberman 005). Without the loss of generality, we normalize c l to zero. This normalization is justifiable in the following senses. From a consumers ersective, cashback shoers are a grou of consumers who value saving beyond the time associated with getting discounts (Swan 010). From an analytical ersective, the incentive for rice discrimination still holds even if c 0, as long as cl ch (Gerstner et al. 1994). l Consumers self-select whether to ay a regular rice r, or a lower ost-cashback rice 1, deending on whether they articiate on the cashback site. The relative magnitude between one s transaction cost c i, and the incentive obtained from cashback shoing, (defined as r ), determine the selfselection outcome (Figure ). 10 By the direct channel we mean a merchant s e-commerce storefront, while the term can be generalized to include the merchant s hysical stores. 11 It usually takes days for the rewards to be available for redemtion. 1 The ost-cashback rice is an ex ost one erceived by consumers after factoring in with the cashback rewards. 6 Thirty Fourth International Conference on Information Systems, Milan 013

7 Ho et al. / Online Cashback Pricing Figure. Consumer s Self-selecting Mechanism via an Electronic Dual-channel To best model the current cashback ractice, we consider a three-stage Stackelberg game with two layers: a merchant and a cashback site. Unless otherwise indicated, we use the site and the ublisher interchangeably. In the first stage, the merchant decides whether or not to affiliate with the cashback site. If the affiliation is formed, the merchant then chooses a regular rice r and a commission rate b ( b [0,1] ) in the second stage. From the merchant s ersective, this commission can be considered the cost to ay for being able to rice-discriminate. In the last stage, the intermediary site makes b times the sales revenue it brokers. Meanwhile it chooses a cashback rate a ( a [0, b] ), and rewards ublisher users with a times transactional amount in a form of cashback. The site incurs a zero marginal cost since it merely oerates as an intermediary and does not directly deal with transactions (see Section 1.1 for industry ractice). The merchant s marginal cost is assumed constant and can be normalized to zero by interreting consumers reservation rice as net of marginal cost. Two firms work indeendently 13 and maximize their resective rofits: C max (1 b) Q ( ) Q ( ), (3) r, b m r l r h r max ( b a) Q ( ), (4) a r l where = r (1-a). The suerscrit C on π indicates cashback ricing and subscritions m and denote the merchant and the ublisher (or, the cashback site), resectively. For the cashback model to work as a rice discrimination device, consumers incentive comatibility (IC) constraints must be satisfied. 14 Critical readers may argue that the merchant s desire to sort out consumers is not necessarily aligned with the site s best interest, in that the site may have incentive entice also tye-h consumers by setting a higher cashback rate. In fact, this would never haen in ractice. Once both consumer tyes become cashback shoers, all consumers would ick the cashback channel and ay the lower rice and therefore nominal asymmetric rices actually work as a uniform rice. The merchant s net rofit, defined as sales revenue subtracted by commission, is strictly less than the level that can be achieved by the otimal uniform rice. If this were the case, the merchant would rather simly set a uniform rice and leave the site, which ends u making a zero rofit without commission. Such credible threat tightly aligns the interests of two affiliate artners under the cashback ricing model. Pricing Decision Maximizing () and solving (3) and (4) using backwards induction gives two firms otimal ricing terms. LEMMA. When adoting cashback ricing model, the merchant sets regular rice v r, and 13 To understand the cashback industry we conducted an interview with the resident at a major cashback site in the U.S. Per interview minutes, merchants in ractice have no control over the ublisher s choice on cashback rate. 14 Consumers IC constraint is 0 c. h ch If is too small such that the interior solution doesn t exist, then we would end u with the corner solution a c /. The merchant s rofit-maximization roblem degenerates to a h r simle case with only one decision variable. This scenario deviates from the main interest of this study, and hence we focus on the interior solutions for the rest of this aer. Thirty Fourth International Conference on Information Systems, Milan 013 7

8 E-Business and Cometitive Strategy commission rate b 1. The cashback site chooses cashback rate a 1. 3 With cashback mechanism oerating as a rice-discrimination device, the merchant is able to charge asymmetric rices across segments. It now can extract the highest surlus from tye-h consumers by increasing the rice from to. Such rice hike is consistent with the second inequality resented in Lemma 1 (i.e. u h u h ). Following the first inequality of the same attern (i.e. ), one may exect u the ost-cashback rice for tye-l consumers, to be lower than the uniform level. However, our common intuition is not always the case in the resence of cashback mechanism. PROPOSITION 1. The cashback aradox, where ost-cashback rice is higher than the uniform rice, 1 will haen as long as market configuration falls in the region RX (, ),, 1 1, if 0 ; (1 ) 3 where and (1 ), if, 3 From consumers ersective Proosition 1 indicates a shocking result. Cashback shoing rovides an attractive saving oortunity for bargain hunters as the rices they ay are erceived lower. Surrisingly, under some conditions the seemingly low ost-cashback rice is actually more exensive, relative to the uniform rice in settings where rice discrimination is absent. As a consequence, after the cashback romotion is imlemented consumers regardless of their tyes end u facing a higher rice (i.e. ). r u 1 Relative Price to The Uniform Price θ = δ Figure 3. Region for Promotional Price Paradox Figure 4. Asymmetric Prices vs. Uniform Price Figure 3 lots the market configuration region R X (on a θ-δ coordinate) in which the cashback aradox will occur. What is the driving force behind such counterintuitive result? Recall From Lemma 1 that the otimal uniform rice, is bound between and. Figure 4 illustrates various otimal rices as a u l function of θ. 15 Clearly, when θ=0, u h as all consumers are high tye; when θ=1, u l as all consumers are low tye. In the current industry ractice, the merchant exerts no control over the site s choice of commission rate. The ability to set commission rate a allows the site to seek its own highest rofit margin. This ricing structure is an analogy to a traditional suly chain setting (Dellarocas 01) where the manufacturer and the retailer indeendently choose their resective rices. The uward rice h θ 15 For resentation urose, and are truncated at since beyond this oint the merchant will switch back to the uniform ricing. Details on affiliate decision will be discussed in the next subsection. The figure is lotted with δ= Thirty Fourth International Conference on Information Systems, Milan 013

9 Ho et al. / Online Cashback Pricing distortion stems from double marginalization raises the rice targeted at tye-l consumers from l to Interestingly enough, when θ lies in the interval [, ], the merchant can still make a higher net rofit by simultaneously increasing two asymmetric rices, comared to the highest rofit that can be achieved by a uniform rice. The exlanation to this counterintuitive result is given as follows. When the merchant has incentive to serve both segments, the otimal uniform rice u, is decreasing with the fraction of tye-l consumers θ. If θ is sufficiently large (i.e. ), the uniform rice is relatively low, comared to the otimal rice when only tye-h segment is served. Under this condition cashback affiliation not only allows the merchant to extract the highest surlus from high-tye consumers but also generates a ositive net rofit from low-tye ones. It is worth noting that while we assume linear demand for simlicity, none of our analyses are deendent on this assumtion. 16 Affiliate Decision When should the merchant affiliate with the cashback site? The subgame erfect equilibrium gives the answer. affiliating, if (, ) R, PROPOSITION. The merchant s best affiliate decision D P not affiliating, otherwise, where R P (, ) 1,. (1 ) We have shown (in Lemma 1) that discriminating ricing is always referred to uniform ricing when rice setter can discriminate consumers costlessly. Proosition suggests that the merchant should adot the cashback ricing model as long as market configuration falls in the rofitable region, R. From a merchant s ersective, the rofitability of the underlying mechanism is determined by two comonents: (1) the revenue gain resulted from being able to rice-discriminate and () the cost incurred in order to obtain such ricing weaon (i.e. the commission ays to the ublisher). In what follows, we examine the effect of market arameters on the merchant s affiliate decision. When the fraction of tye-l consumer θ is fixed, the advantage of discriminating ricing over uniform ricing is diminishing in tye-l s valuation coefficient δ. If the valuation ga between two segments vanishes (δ=1), the otimal uniform rice yields the highest rofit. When δ is fixed, the fraction of tye-l consumers, θ, moderates attractiveness of the cashback affiliate. A high value of θ imlies more sales volume is generated via the cashback channel, and therefore the merchant ays a larger ortion of its total revenue to the site. Combined, these two effects suggest that the cashback model is rofitable if and only if θ and δ are sufficiently small. As (θ, δ) becomes smaller (larger), the market will shift towards (away from) the cashback model. We will use this rincial to facilitate our subsequent discussion. Current cashback ractice suffers from market efficiency loss caused by double marginalization. Arguing that two affiliate artners can slit the rofit through a bargaining rocess, we roose a otential solution to mitigate the undesirable outcome identified in our model. Cashback Coordination We have examined when and how the cashback model can be used as a rice discrimination device. What remains unclear so far is the imact of the underlying mechanism on consumers as a whole and on the entire society. In this section, we first resent a welfare analysis of the current cashback ractice. Then, we roose an alternative aroach to imrove market inefficiency which stems from double marginalization. Our discussion on welfare roceeds in the following two cases. When (θ, δ)r H, the monoolist with uniform ricing would set u=v/ and serve the high-tye only. In the resence of the cashback mechanism, the merchant can now set the same rice r=v/ for tye-h segment and meanwhile rovides 16 The imact of double marginalization, including rice distortion and welfare reduction, would be stronger if the marginal revenue curve is convex, since the ublisher would try to get a bigger ie by setting a cashback rate further from the channel otimum.. Thirty Fourth International Conference on Information Systems, Milan 013 9

10 E-Business and Cometitive Strategy a lower rice for the other. In this case, the cashback ractice ossesses all the merits: it increases overall consumer surlus, sellers surlus and social welfare. When (θ, δ) R H, however, the imlementation of the current cashback ractice leads to a reduction in overall consumer surlus and social welfare, comared to the level under uniform ricing. This reduction in both welfare measurements echoes rior literature (e.g., Schmalensee 1981) since the aggregate outut decreases due to the uward rice distortion. The affiliated merchant makes (1-θ)v /t from its direct channel and θ(δv) /4t from the cashback channel. The intermediary site receives sales commission θ(δv) /8t. Since each firm is able to set its own rofit margin, two sellers joint rofit, given by 4(1 ) v 3 ( v) / 8t, is below the joint otimum (1 ) v ( v) /t if the work as an integrated firm (Jeuland and Shugan 1983). In our research context, coordination between two affiliate artners in determining cashback rate is the most straightforward solution to this roblem. In what follows, we identify conditions under which cashback coordination can be achieved through a bargaining mechanism. For distinction urose, we call our roosed mechanism Coordinated Cashback, as oosite to the current cashback ractice. Bargaining Process and Coordination We denote the cashback site s and merchant s bargaining ower by φ and 1-φ, resectively. Following the rior literature, the values of bargaining arameters are exogenous and may deend on each arty s relative market ower such as market value, brand image, etc. The merchant and the site (hereafter we call them sellers for brevity) bargain over the cashback rate. If bargaining fails, two firms resort to their outside otions which are determined by market configuration. If revenue of the current cashback model is verifiable, two firms would lay the Stackelberg ricing game. Otherwise, no affiliation would be formed and the site earns a zero rofit. The sellers resective outside otions, denoted by a suerscrit O, are given by: C ( v) 1 v m O 4t t m U 1v m 4t if (, ) R, otherwise; P C ( v) O if (, ) RP, 8t U 0 otherwise. In its simlest resentation, the Nash bargaining solution (Nash 1950) 17 is the solution ( c, a c) to maximize sellers joint rofit: I I I max (1 a ) Q (1 a ) Q ( ), c, ac m c c l c c c h c I where denotes the sellers joint rofit when two affiliate artners coordinate and work as an integrated firm. Following the slit-the-difference rule (Muthoo 1999), two firms rofits under I I I I I coordination is allocated by m (1 ) m and, resectively. The success of the bargaining rocess eliminates double marginalization by droing the site s choice of commission rate. Exogenous bargaining arameters serve to allocate the rofit generated through the cashback channel. When φ is large, the site receives a big ie of, making coordination less attractive to the merchant. When φ is small, the site would leave the bargaining table because of insufficient incentive. Therefore, we can exect that channel coordination can be achieved when the value of φ is bound between some critical numbers. PROPOSITION 3. There is an interval ˆ for the cashback site s bargaining ower, such that both firms have incentive to coordinate and slit the revenue by their relative bargaining ower, where I 17 We choose Nash bargaining solution over Rubinstein model because we assume each consumer s demand is unitary and the cashback game is a static one-shot game. 10 Thirty Fourth International Conference on Information Systems, Milan 013

11 Ho et al. / Online Cashback Pricing 1 1, if (, ) RP ; 4 ˆ (1 )(1 ) 0, otherwise. (5) Clearly, the interval ˆ deends uon the market configuration. When (, ) RP, two firms have incentive to coordinate in choosing commission rate as long as the merchant s bargaining advantage 18 is moderate. O When (, )., since the site s outside otion is zero ( 0 ), any ositive value of φ can incentivize R P the site to coordinate. As a result, our roosed mechanism is sustainable as long as the ublisher does not claim too much share of revenue generated from tye-l consumers (i.e. φ is sufficiently small). Figure 5(a) deicts the interval ˆ as a function of δ (θ=0.5). Figure 5(b) illustrates the uer bound of ˆ when (, ) RP. Since δ moderates attractiveness of rice discrimination devices 19, a large value of δ shrinks the interval ˆ. The value of θ has a similar imact, as shown in the same lot. The intuition is that it is harder to align two firms interest when market shifts away from discriminating ricing. When (, ) RP, however, neither δ nor θ has imact on ˆ. This is because the revenue of the current cashback model is verifiable and rofit-sharing scheme is indeendent of market configuration. (a) Feasible region ˆ for coordination (b) Uer-bound of interval ˆ 0.5 φ φ 0.5 φ (θ=0.1) φ (θ=0.5) φ (θ=0.9) 0 δ Figure 5. Coordination condition, ˆ, as a function of δ δ Welfare Imrovement Now, we demonstrate how our roosed mechanism rovides a more efficient aroach to ractice rice discrimination. From sellers ersective, this alternative aroach exands the cashback model s rofitable region by taking bargaining ower into decision making rocess. The bargaining rocess ensures a win-win situation for both firms. As seen in Figure 6, while the current ractice outerforms the uniform ricing when, the success in bargaining makes both firms better off as long as. 0 From buyers ersective, coordinated cashback increases consumer surlus by restoring the rice targeted at low-tye segment from to. The left anel of Figure 7 shows that consumer surlus l under coordinated cashback (black solid lines) dominates the current ractice (gray dashed lines). The cashback aradox will never occur under this rofit-sharing scheme. Social welfare (SW) has an unambiguous increase (the right anel of Figure 7). The cashback coordination discussed in this section is different from coordination achieved by a rofit-sharing contract in that the former is analyzed under a rice-discrimination and dual-channel framework. 18 By construct, the merchant s bargaining advantage is (1 φ) φ = 1 φ. 19 A detailed discussion is given in revious section. 0 We derive by arranging the bottom line of Equation (5). The figures are lotted with θ=0.5 and φ=0.3. Thirty Fourth International Conference on Information Systems, Milan

12 E-Business and Cometitive Strategy Merchant s Profit Cashback Site s Profit δ Figure 6. Coordinated Cashback vs. Current Cashback Consumer Surlus (CS) Social Welfare (SW) δ Figure 7. Comaring Consumer Surlus and Social Welfare δ Otimal Pricing Strategy in Duooly Based on a monooly setting, we have identified unique roerties of the cashback model. We now turn our attention to merchants ricing and affiliate decision in the resence of cometition. In what follows, we base our analysis on a duooly while our framework can be generalized to an oligoolistic setting. Duooly setting. Consider a market served by two merchants (M 1 and M ) who sell differentiated roducts. Motivated by Dogan et al. (010), we consider the differentiation both horizontal and vertical. On one hand, roducts offered by different merchants have various combinations of attributes. Such horizontal difference is modeled in the following way: two cometing merchants are located at two different ositions on consumers reference horizon. 1 Each merchant has its own brand valuation. Factors determining a merchant s brand valuation could be its brand image, level of customer services, reutation, etc. We assume tye-h consumers have brand valuation of v j for M j whereas tye-l consumers have brand valuation of δv j for M j (j=1,). Without the loss of generality, we assume v 1 > v and we call M 1 a suerior merchant. Denoting the distance between two merchants locations by d, we model the intensity of cometition between merchants by the recirocal of d. A smaller d reresents a higher degree of cometition. Figure 8 illustrates tye-h consumers transaction utility derived from two differentiated roducts. A consumer would refer to urchase from the merchant who gives her a higher utility. Consumers located at the rojection of the intersection of two inwards utility segments on the reference horizon are indifferent between buying from either merchant. Similarly, consumers located at the intersection of M j s utility segment and reference horizon are indifferent between buying M j s roduct and not buying at all. 1 The reasons why we choose Salo s model over Hotelling s are given as follows. First, Salo s model allows us to ignore firm s location decisions, which is a common concern in marketing and IS satial model literature (see Dewan et al. 003 for examle). Second, Salo s model is more flexible in a sense that it can be easily extended to a multilemerchant case. Lastly and erhas most imortantly, Hotelling s satial model cannot model the ublisher s decision when the cometition between merchants is resent. This is because when market is fully covered, the ublisher would have no decision to make. 1 Thirty Fourth International Conference on Information Systems, Milan 013

13 Ho et al. / Online Cashback Pricing Figure 8. Consumers Choice Over Differentiated Products (Tye-h) To simlify notations, we move some notations to suerscrits. In a general format, the merchant j s rofit maximization roblem, conditional on its affiliate decision, are given by: A r h r r l max Q (, ) (1 b ) Q (, ) if affiliating, r j, b j U u h u u l u max Q (, ) Q (, ) otherwise, u j and the ublisher s roblem is given by: j j j j j j j j j j j j j j j j j j j 1, max r l ( b a ) Q, a j j j j j j j j where j denotes the rice set by M -j and is deendent on M -j s affiliate decision. Otimal Pricing Decisions The fact that a merchant s rofit also deends on the rival s affiliate decision adds extra difficulty to our duoolistic model. Since each firm has to decide whether or not to affiliate, there are four ossible market outcomes: both merchants affiliate, M 1 affiliates alone, M affiliates alone, and neither merchant affiliates. To derive the subgame erfect solutions, we first solve the thrid-stage ublisher s rofit maximization roblem, given that two merchants affiliate and ricing decisions are known. LEMMA 3-1. The otimal cashback rates a, 1 a under the cometitive market outcome are: B B 1 b1 dt v1 1 b dt v a1, a, The second-stage solutions suggest that the site s otimal cashback rate is increasing with the commission rates chosen by the merchants. This finding is consistent with the cashback rates observed on a tyical site. For examle, merchants belong to magazine/book category usually give a high commission rate (u to 50%) to the ublisher, which in turn, assigns high cashback ercentages to consumers. Having the solutions to the ublisher s roblem, we now are able to derive the merchants best resonse in the second stage. LEMMA 3-. The merchant j s otimal rice and commission rate under the cometitive cashback market where both merchants affiliate are: B 1 1, B v v dt 1, v v b 35 17v13v 7dt ; b B 1, B v v dt, v v v 3v17 dt The comarative statics of the otimal ricing terms under the cometitive cashback market are summarized in Table. We find that the effect of brand valuation on rice and that on commission b move in the same direction. If a merchant is able to enhance its brand valuation, i.e. v j is higher, it can We only resent the solutions under the cometitive market in the main text of this aer. Comlete solutions are available uon request. Thirty Fourth International Conference on Information Systems, Milan

14 E-Business and Cometitive Strategy charge a higher regular rice and assign a higher commission rate to the ublisher. The increase in commission rate will in turn lead to a higher cashback ercentage. Interreting the magnitude of b j as the attractiveness of cashback affiliate to a merchant j, we find that a merchant s incentive to adot cashback mechanism is increasing with its own brand valuation v j. This nature is helful in exlaining the market equilibrium derived in the next subsection. A merchant s otimal rice and incentive to affiliate decrease with the cometitor s brand valuation, v -j. Unlike brand valuation, the effects of market arameters on merchant s ricing terms and b move in the oosite directions. Recall that 1/d measures the cometition intensity between merchants. In a highly cometitive market where horizontal roduct differentiation is insignificant, i.e. 1/d is large, merchants would engage in a rice war by cutting the rice and taking a more aggressive rice discrimination strategy (i.e. a larger b j). If consumers areciate roducts attributes more than rice, as modeled by a larger t, merchants would extract larger surlus by ushing the rice uwards. If δ is larger, indicating the valuation asymmetry between two segments is small, i.e. we will exect that rice asymmetry across channels goes down. Table. Comarative Statics of the Otimal Price and Commission Rate Decision Variables j bj Parameters vj v-j 1/d t δ Equilibrium of the Cashback Market Although we have solved for firms ricing decisions, the solutions themselves do not constitute equilibrium (Dogan et al. 010). To derive equilibrium conditions we need to further verify that neither merchant has incentive to deviate from a articular solution set by changing its affiliate decisions. The rocedure of deriving subgame erfect equilibrium requires solving an inequality system with multile arameters. To circumvent the comlexity of searching through a multidimensional sace, we resort to numeric analysis. The following lemma rovides an interesting finding in the resence of cometition. LEMMA 4. When a merchant adots cashback affiliate, this move would also benefit its cometitor. For discussion urose, consider a scenario where M 1 moves to the cashback market given M s strategy is fixed. According to the nature of rice discrimination, M 1 s rice for tye-l segments would decrease in the most cases. Such rice cut boosts M 1 s sales volume but hurts M s. At the same time, M 1 would increase the regular rice; the rice to ay here is to surrender some tye-h market share to M. As a result, M would be better off if M 1 lessens the rice cometition in the more rofitable high-tye segment. Now, we turn our attention to the conditions of market equilibrium. Recall that the attractiveness of the cashback model is decreasing with market arameters δ and θ. When the market configuration is strongly in favor of cashback mechanism, (i.e. close to the origin of the θ-δ coordinate), affiliating is a dominant strategy for both merchants. A duoolist would benefit not only from its own move into cashback market but also from its rival s (Lemma 4). This win-win situation leads the market to a ure strategy equilibrium wherein both merchants affiliate and make the highest rofit, comared to the rofit they can achieve in any other equilibria. When the market configuration is against discriminating ricing (i.e. away from the origin of θ-δ coordinate), both merchants lean towards conservation in affiliate, leading the market to an equilibrium wherein neither merchant affiliates. What remains unclear so far is the outcome when the market configuration is weakly in favor of the ricing tool. Since the rofitability of the cashback model become ambiguous, one may conjecture that the asymmetric equilibrium wherein one merchant affiliates alone will emerge. We formally state our finding in the following roosition. PROPOSITION 4. When market configuration is weakly in favor of the cashback affiliate, at equilibrium market outcome deends on merchants valuation difference v, where v v v If v is small, both merchants have no incentive to affiliate alone but refer its rival to do so. This unique characteristic leads the market to a hawk-dove game. 14 Thirty Fourth International Conference on Information Systems, Milan 013

15 Ho et al. / Online Cashback Pricing. If v is large, the cashback model is not rofitable to the low-valuation merchant. At equilibrium only the high-valuation merchant would affiliate with the ublisher. 3. If v 0, two merchants are identical and as a result asymmetric equilibrium doesn t exist. If v is small, merchants would lay an anti-coordination game in which it is mutually beneficial for duoolists to lay oosite strategies: M j would like to affiliate only if M j does not; and M j would refer not to affiliate only if M j affiliates. Economists would use the hawk-dove game (also known as snowdrifting or chicken game) to describe this secial scenario where multile equilibira co-exist. In our research context, both merchants hoe for the other to move, but neither does. If v is sufficiently large, M would adot the simle uniform ricing because the cost of the affiliation outweighs the benefit. Under this scenario M 1 can still benefit from rice discrimination. This finding is consistent with the literature stream arguing the advantageous firms should take a more aggressive romotion strategy (e.g., Rao 1991). Notably, our work rovides a different insight from Dogan et al. (010) who concluded that the equilibrium where only advantageous rice-discriminates doesn t exist. What drives this difference is that we allow asymmetric reservation rice across segments and generalize their assumtion that all consumers have identical willingness to ay. Conclusion and Future Research The rimary objective of this aer has been to examine the strategic use of the still-nascent cashback affiliate model. By affiliating with the cashback latform, merchants are able to exercise second-degree rice discrimination to increase rofits. Desite the merits from sellers ersective, cashback mechanism actually hurts consumer surlus and social welfare. Surrisingly enough, consumers as a whole could face a higher rice after the introduction of cashback shoing. Arguing coordination can be achieved through a bargaining rocess, affiliated artners can make themselves better-off by imroving channel efficiency. This aer contributes to the literature in many asects. We exlicitly consider the imlementation cost associated with the underlying romotional vehicles. In addition, we endogenizes the roduct rices as a decision variable into merchants rofit function. These analytical advantages allows us to obtain more recise insights into the attractiveness of the cashback model. This unique feature distinguishes our work from rior literature in romotions and channel control (e.g., Chiang et al. 003). Our analytical framework is simle; yet, it can be generalized to other commission-based aroaches. This aer makes a few assumtions. First, the seller in our model face a linear marginal revenue curve based on the linearity of demand. However, none of our analyses are deendent of this assumtion. One should exect, for examle, the imact of double marginalization would be more severe if the marginal revenue curve is convex. Second, we assume that the consumers sensitivity to horizontal differentiation (as reresented by t) is the same across segments. It could be case that tye-h consumers areciate roduct attributes more than those of the other tye. In fact, our results do not change significantly if we consider such secification. Intuitively, if t h > t l, a monooly would have higher incentive to ricediscriminate, making our analysis on affiliate decision conservative. In a duooly setting, the merchant with lower brand valuation would benefit more from cashback marketing. Lastly, for the sake of model tractability we assume the transaction costs incurred in cashback shoing are discrete cross segments, since without this assumtion the sellers rofit will be a cubic function and the equilibrium solution will turn out to be messy. Our analytical framework should give a similar result even if we release this assumtion by allowing c h to be continuous, with c l still being equal to zero since cashback shoers value savings more than time cost (Swan 010). We should exect that the merchant s affiliate decision will deend on the tradeoff between additional demand generated by tye-h segment and lower margin er unit sold. Future research could investigate the change of otimal rofit-sharing scheme when multile cashback sites are involved. In reality, a merchant does affiliate with multile cashback latforms at the same time. The observation oints out a series of new questions. Is multi-homing strategy more beneficial to a merchant? If so, how does multi-homing strategy influence the cometition among merchants and how do latforms comete with each other? Moreover, network externality is one of key comonents to deict the nature of latform. Previous literature concludes that latforms fight for a enny to survive. However, we wonder whether or not the conventional intuition still holds when the cashback concet comes into lay. Thirty Fourth International Conference on Information Systems, Milan