Abstract
We study optimal fee setting decisions by a monopoly online platform connecting advertisers with potential buyers in two environments: a simple model that captures stylised features of advertising on search engines, social networks and advertisement-supported email; and a richer model that is more relevant for ‘directed’ search at price comparison sites. While the platform can choose to charge for both impressions and clicks, we show that the platform maximises profits by using clickthrough fees exclusively. Our model offers a rationale for the evolving practice of relying purely on clickthrough fees for revenues in many online advertising markets.
The pricing of online advertising has undergone a sea change since its early days. The dominant old media model of paying per impression, the so-called CPM (cost per impression) model, has given way to a pricing model where most payments are made contingent on the viewer taking some action, typically clicking on the advertisement.1 This is the so-called CPC (cost per click) model of pricing. In this article, we investigate why platforms such as price comparison sites have shifted from CPM to CPC.
This change is usually seen as a response to pressure from advertisers worried about the performance of the new online media. Under the CPC model, an advertiser only pays when an advertisement is effective – based on the user action of clicking. Thus, the risk of performance is shifted from the advertiser to the site displaying the advertisements. While there is little doubt that advertisers were (and still are) sceptical about the returns to online advertising, this article points out an additional motivation for the shift: the platform hosting advertising content – what Baye and Morgan (2001) have previously dubbed the ‘information gatekeeper’– also stands to benefit from this pricing arrangement. Indeed, our main result is that, when CPC and CPM are both available and can be used jointly as pricing instruments, the platform maximises its profits by using CPC exclusively.
One might speculate that the CPC and CPM instruments are analogous to two-part tariffs for a standard monopoly: price at marginal cost and use a fixed fee to extract surplus from advertisers. Under this logic, one might erroneously conclude that the gatekeeper should optimally set the clickthrough fee at zero (its marginal cost) and use an impression fee to extract surplus from potential advertisers.2 Where this logic goes wrong is that it ignores competition among advertisers. In the standard setting, one buyer's decision to accept the monopolist's offer does not impact the valuations of other buyers. In the present setting, one firm's decision to have its advertisement displayed does impact the amount other firms are willing to pay, as it reduces the likelihood that their advertisements will be viewed as ‘best’ and hence clicked by consumers. An optimising platform takes this ‘competition for clicks’ into account in its pricing, and this makes the analysis of the optimal fee structure more complex than in the standard textbook case.
The trade-offs between CPC and CPM can be clearly seen in the simple model of optimal platform pricing presented in Section 1, where advertisements exogenously differ in their attractiveness and relevance to consumers. As consumers click on the most attractive and relevant advertisement, a less attractive advertiser might still hope to obtain clicks via relevance. Advertisers enter until the least attractive (marginal) advertiser is just indifferent between utilising this channel or earning zero from its outside option.
Suppose that impression and clickthrough fees are adjusted to maintain the same number of advertisers. Then a $1 increase in the clickthrough fee requires a reduction in the impression fee in proportion to the chance that no other firm's advertisement is displayed, i.e. an order statistic. The platform benefits from such an adjustment in proportion to the chance that at least one firm's advertisement is displayed (i.e. deemed relevant by its algorithm) since otherwise the platform gets no clickthrough revenues. It is harmed by the reduction in impression revenues from all firms. The heart of the argument is establishing that this decrease is proportional to the chance that at least one firm's advertisement is displayed. As the former probability always exceeds the latter, a greater reliance on clickthrough fees raises platform profits and hence the exclusive use of clickthrough fees is optimal (Proposition 1). Put differently, while clickthrough fees operate on an order statistic – only the most attractive relevant firm pays – impression fees affect all firms and hence are a blunter instrument for extracting surplus. Of course, once clickthrough fees are used exclusively, they have the same effect as a reserve price in an auction. Proposition 2 shows how the platform optimally adjusts the clickthrough fee to induce the profit maximising number of advertisers.
While the simple model is a clean illustration of this intuition, one may worry that the result may be altered or reversed when the attractiveness of advertisements or the value of the outside option is endogenised. To investigate this possibility, Section 2 considers a richer model in the spirit of Baye and Morgan (2001), the first study to examine the fee-setting behaviour of a ‘gatekeeper’ serving consumers and firms in a two-sided online market. This model closely corresponds to a setting where the platform is a price comparison site.3 The price charged by a firm determines the attractiveness of its advertisement, while the firm's outside option (i.e. the payoff when not using the site) depends on the intensity of advertising at the site. Thus, both the attractiveness of advertisements and the value of eschewing the platform's channel are endogenous.
Despite these modelling differences, CPC remains the superior instrument for capturing surplus in the market (Proposition 3). Once again, the key driver is the fact that shoppers click on the best of the listed advertisements. Consequently, a firm realises that it is obliged to pay a clickthrough fee only if, ex post, its advertisement best matches the preferences of shoppers visiting the gatekeeper's site. By contrast, an advertiser pays the impression fee regardless of whether it generates any clicks and before any information is revealed about whether its advertisement is ‘best’. Finally, the gatekeeper's expected profits from the impression fee depend on the average number of advertisements induced by the fee, whereas its profits from clickthrough fees depend only on the probability that at least one firm advertises (an order statistic). The contingent nature of clickthrough fees, coupled with this order statistic effect, makes them a superior instrument for extracting surplus from potential advertisers. These results obtain in a model where firms are ex ante symmetric and there is no moral hazard. Obviously, if firms were vulnerable to opportunistic behaviour on the part of the gatekeeper – displaying advertisements to consumers it knows are not interested in purchasing a firm's product, for instance – the superiority of clickthrough fees over impression fees would be even greater.
The institutional structure of the richer model also allows us to study other features of online markets. Conversion rates, the chance that clicks are converted into sales, play a key role. We show that, while the exclusive use of clickthrough fees is optimal for any positive conversion rate, higher conversion rates lead to higher clickthrough fees and higher profits for the gatekeeper. Firms also benefit from higher conversion rates. An important insight to emerge from this analysis is that the platform only partly captures the gains from its investment in improving conversions. Thus, platforms will tend to underinvest in improvements. This perhaps helps to explain why conversions remain stubbornly low (about 5% at most) on these sites.
Although both models share the same implication about the optimality of CPC pricing, Section 3 shows that there are important differences. Both models assume that there are (potentially small) transactions costs to firms wishing to advertise on the platform. In the simple model, platform profits increase continuously as transactions costs fall to zero. In the richer model, the outside option is endogenous and this leads to a discontinuity in payoffs – platform profits exhibit an upward jump when transactions costs are eliminated. In a similar vein, exclusive contracts (i.e. contracts that offer a single advertiser the exclusive right to advertise on the platform) are of no benefit to the platform in the simple model but are potentially helpful in the richer model.
Our article is related to the literature on optimal fee structures in two-sided markets when there is a monopoly platform.4 An important difference is that payoffs in these models are increasing in the number of agents on the opposite side of the platform and independent of the number of agents on the same side of the platform. In other words, competition among users on the same side of the platform is effectively absent. In the settings we study, it is natural to include competition among advertisers, and this is the key driver of the optimality of clickthrough fees over impression fees. Indeed, our analysis suggests that existing results on the equivalence of impression fees and clickthrough fees – cf. Proposition 3 in Armstrong (2006) as well as Proposition 3 in Eliaz and Spiegler (2011)– stem from the absence of user (i.e. advertiser) competition.
Our article is somewhat related to the literature on position auctions (Edelman and Schwarz, 2006; Edelman et al., 2007; Varian, 2007; Athey and Ellison, forthcoming). While these models also study clickthrough fees, their levels are determined by the auction form selected by the platform. Seller decisions in these models consist purely of how much to bid in the position auction.
Like many of these models, our model assumes that there is a single dominant platform in the market. While this potentially limits the applicability of these models, casual empiricism suggests that such market structures are, in fact, common. For instance, eBay is the dominant platform for online auctions in the US; Google enjoys overwhelming market share in search; Facebook is the dominant social networking platform. A burgeoning empirical literature (Liebowitz and Margolis, 1994; Brown and Morgan, 2009; Tellis et al., 2009) documents that single (dominant) platforms are the rule in many markets. Recent experimental evidence (Hossain and Morgan, 2009; Hossain et al., forthcoming) paints a similar picture: unless platforms are strongly horizontally differentiated, the most likely market structure to emerge is a monopoly platform.
