Incentive‐compatible advertising on nonretail platforms

• DOI: http://doi.org/10.1111/1756-2171.12316
• Author: Ran Spiegler,Kfir Eliaz
• Published date: 01 June 2020
• Date: 01 June 2020

RAND Journal of Economics

Vol.51, No. 2, Summer 2020

pp. 323–345

Incentive-compatible advertising on

nonretail platforms

Kfir Eliaz∗,∗∗

and

Ran Spiegler∗,∗∗∗

Nonretail platforms enable users to engage in noncommercial activities, while generating user

information that helps ad targeting. We present a model in which the platform chooses a person-

alized ad-display rule and an advertising fee (which depends on the targeted user group). The

policy that maximizes the platform’s advertising revenues creates an incentive for advertisers to

strategize targeting.We providea condition for incentive-compatibility of the first-best policy, and

highlight the forces that make it harder to satisfy. We apply our result to examples of platforms.

Our analysis of social networks turns out to be related to the “community-detection” problem.

1. Introduction

Recent years have seen a proliferation of online institutions that can be described as “non-

retail platforms.” Users of these platforms access them on a regular basis, in order to engage

in activities such as reading texts, listening to music, exchanging messages, cultivating social

links, etc. In particular, when they access the platform, it is not for the purpose of buying from

advertisers. If a user buys from an advertiser as a result of being exposed to an ad posted on the

platform, the transaction takes place off it; it will have no effect on his activity on the platform,

and it is quite likely that the platform does not even monitor whether the transaction has taken

place. However, the transaction may temporarily depress the user’s demand for similar products,

thus diminishing the effectiveness of advertising them.

Of course, nonretail platforms are at least as old as the village message board. What is

special about the modern online version is that users’ activity on the platform leaves a massive

∗Tel-Aviv University; kfire@tauex.tau.ac.il, rani@post.tau.ac.il.

∗∗University of Utah.

∗∗∗University College London.

This is a substantial revision of a paper formerly circulated (and promoted on YouTube by a movietrailer) under the title

“Incentive-Compatible Advertising on a Social Network.” Financial support from ISF grant no. 1153/13 is gratefully

acknowledged. Wethank Yair Antler, Francis Bloch, Olivier Compte, Andrea Galeotti, Nikita Roketskyi, Omer Tamuz,

Neil Thakral, numerous seminar audiences, and the referees of this Journal, for helpful comments. We are especially

grateful to the editor, David Myatt, for fruitful suggestions.

© 2020 The Authors. The RAND Journal of Economics published by Wiley Periodicals LLC on behalf of The RAND

Corporation This is an open access article under the terms of the Creative Commons Attribution License, which permits

use, distribution and reproduction in any medium, provided the original workis properly cited. 323

324 / THE RAND JOURNAL OF ECONOMICS

trail of information that may be correlated with their consumption tastes in various areas. As a

result, the platform can help advertisers achieve better targeting, which in turn helps the platform

increase its advertising revenues. Here are a few examples of what we have in mind.

Online radio stations like Pandora collect information about users’ musical tastes (in this

respect, they differ from traditional radio), and can use that to target ads for unrelated products.

For instance, whether a user likes Country Music may be correlated with his politics and lifestyle

preferences. But of course, he does not access Pandora for the purpose of being informed about

political candidates or buying vegan food.

Email services may use the content of personal emails to target users. If a user’s emails start

featuring numerous references to babies, he may experience increased exposure to diaper ads on

his email account, although buying diapers is obviously not the user’s primary objective when

checking his email.

Messaging platforms such as Whatsapp or Snapchat may be unable or unwilling to use the

content that users generate, for technical or legal reasons. However, the structure of the social

network among users may provide information about their types. For instance, if users exhibit

homophily—that is, they associate with like-minded individuals—then a large cluster in the net-

work indicates that its members are likely to have similar tastes.

Content sharing platforms such as Reddit are message boards that publish user-generated

content, and may monitor the content that users produce or consume.

Many platforms exhibit combinations of these features. For instance, social media platforms

like Instagram or Twitter can use the network structure of their users as well as the content that

they generate. Although not all of these real-life examples of nonretail platforms currently use

this form of targeted advertising, the potential to do so is inherent in them.1

In this article, we study novel incentive issues that arise in advertising on nonretail plat-

forms. The source of the potential incentive problem is that advertisers have private information

regarding the consumer-preference types they would like to target. The platform relies on their

targeting requests to allocate display ads to individual users, utilizing its own private informa-

tion about users. When advertising fees vary with the targeting request, it becomes a strategic

decision that involves trading off the likelihood of a transaction against the fee. Indeed, real-life

ad-tech intermediaries help advertisers cope with such trade-offs by searching for the target au-

dience that gives the “best bang for the buck.” This may involve diverting the client’s ad to a

less-than-ideal audience to save costs.2

Users’ ad-generated (offline) purchases can affect their willingness to make subsequent

purchases—for example, because they are temporarily satiated. However, this change in their

consumption-driven behavior has no visible effect on their platform activity, which is not

commerce-oriented to begin with. We will see that as a result of this feature, the platform may

want to diversify the type of ads it shows to an individual user. Even if a Country Music fan is

relatively unlikely to be interested in vegan food, exposing him to such ads every once in a while

may increase the long-run expected number of transactions generated by such a user.The article’s

basic insight is that this diversification motive creates an incentive for advertisers to misrepre-

sent their ideal targeting. Our aim is to understand the conditions in which this incentive problem

prevents the platform from attaining its first-best.

In our model, there is a group of consumers with constant access to some nonretail platform.

Each consumer comes in one of two (private) preference types. A type can describe whether

the consumer is interested in “healthy food,” whether he likes “highbrow” movies, whether he

enjoys outdoor recreational activities, etc. The platform obtains a noisy aggregate signal about

1Search engines are an example of hybrid platforms with both retail and nonretail features. Consumers use search

engines to specifically look for a product or service to buy, but they also use them to find information unrelated to any

transaction. Wedo not address such platfor ms in this article.

2For example, AdEspresso.com is a company that offers to help small businesses launch advertising campaigns

on social media. On their website they wrote, “The audience you choose will directlyaffect how much you’re paying…if

your perfect audience is just more expensive, that’s just the way it goes.”

C

The RAND Corporation 2020.