Targeted Advertising Platforms: Data Sharing and Customer Poaching
Klajdi Hoxha · 2025
Read Targeted Advertising Platforms: Data Sharing and Customer Poaching by Klajdi Hoxha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-27112.txt.
Introduction
Klajdi Hoxha, in a 2025 study of online markets, takes up a design puzzle for shopping sites. Big platforms ask shops to pool what they know about buyers, then aim ads with the shared facts. Shops fear a catch: rivals may use the pool to steal their own buyers. Hoxha asks what rules would make shops share anyway.
He builds a math model and finds the best setup for the platform. At its core are three markets. In one, shops sell all their data at a set price. In the next, shops swap data for ads that draw more clicks. In the third, rich shops buy those strong ads at full price. He notes the same design may fit carbon credit trades and shared public funds.
Catalog record
| CREATOR | Klajdi Hoxha |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 108 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-27112
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Targeted Advertising Platforms: Data Sharing and Customer Poaching Thanks: I am grateful to Michael Ostrovsky, Andrzej Skrzypacz and Weijie Zhong for their guidance and support. I am thankful to Bing Liu for numerous discussions during the early stages of the paper, as well as Ellen Muir, Spencer Pantoja, Ilya Segal, Eric Tang, Frank Yang and seminar participants at Stanford University for helpful comments and suggestions. I owe special thanks to Varanya Chaubey for exceptional editing advice. The author declares that he has no relevant or material financial interests that relate to the research described in this paper. Klajdi Hoxha Abstract E-commerce platforms are rolling out ambitious targeted advertising initiatives that rely on merchants sharing customer data with each other via the platform. Yet current platform designs fail to address participating merchants’ concerns about customer poaching . This paper proposes a model of designing targeted advertising platforms that incentivizes merchants to voluntarily share customer data despite poaching concerns. I characterize the optimal mechanism that maximizes a weighted sum of platform’s revenues, customer engagement and merchants’ surplus. In sufficiently large platforms, the optimal mechanism can be implemented through the design of three markets: OPEN i ) i) selling market , where merchants can sell all their data at a posted price p p , OPEN i i ) ii) exchange market , where merchants share all their data in exchange for high click-through rate (CTR) ads, and OPEN i i i ) iii) buying market , where high-value merchants buy high CTR ads at the full price. The model is broad in scope with applications in other market design settings like the greenhouse gas credit markets and reallocating public resources , and points toward new directions in combinatorial market exchange designs. Keywords : Exchange market, click-through rate (CTR), customer data, bundling, mechanism design, platform, poaching, scoring rule, targeted advertising. Introduction E-commerce platforms are rolling out ambitious targeted advertising initiatives that rely on merchants sharing customer data with each other via the platform. These initiatives are in response to recent privacy policies—such as California’s CCPA, European Union’s GDPR and Apple’s ATT—limiting third-party cookie tracking, a cornerstone of targeted online advertising. 1 1 1 96% of iOS users have opted out of app tracking and 80% of advertisers rely on third-party cookies according to https://www.wordstream.com/blog/ws/2021/10/26/online-advertising-statistics and https://www.imore.com/96-iphone-users-have-opted-out-app-tracking-ios-145-launched With Safari, Firefox and Edge having phased out third-party cookies, Chrome is now the only major browser still relying on them as it seeks alternative solutions. 2 2 2 https://privacysandbox.com/news/privacy-sandbox-update/ One promising solution involves e-commerce platforms pooling customer data from merchants’ own datasets. Doing so helps the platform build a larger and more targeted dataset, thereby delivering targeted ads to customers from their preferred merchants. Yet current platform designs fail to address participating merchants’ concerns about customer poaching. These designs give participating merchants the opportunity to run advertising campaigns on the platform that target every customer in the shared dataset. To determine which merchants get to target which customers, such platforms commonly allocate targeted ads by running ad auctions based on bid amounts and ad quality scores, where the latter measures the probability of a click. If competitors with ads at least of similar quality bid more aggressively, merchants’ ad campaigns may reach fewer customers than before joining the platform. Our running example that illustrates this setting is Shopify, an e-commerce platform that powers the (online) stores of millions of merchants. In 2022, Shopify pioneered these initiatives by launching Shopify Audiences, a tool designed to pool customer data from participating merchants to improve their targeted advertising. However, a frequently asked question on Shopify’s official page reveals a growing concern: can merchants “exclude competitors from using their data”? 3 3 3 https://help.shopify.com/en/manual/promoting-marketing/shopify-audiences/faq Shopify’s ambiguous response—“…the data you contribute will be used to benefit other participating merchants”—only validates the customer poaching concerns. This paper proposes a model of designing targeted advertising platforms and solves for the optimal platform design. Specifically, a designer designs a platform where a finite number of merchants participate by contributing exclusive customer datasets and compete for targeted advertisements designed using the aggregated dataset . Each customer is represented by a profile of click-through rates (CTRs), one for each merchant’s ad. Merchants earn a private profit-margin per click (type). 4 4 4 I assume for simplicity that customers always convert (i.e. buy the product) if they click the ad. To elicit this information truthfully, the designer assigns a participation fee (transfer)—possibly negative—and credibly commits to some advertising mechanism such that all merchants are willing to participate. The designer maximizes a weighted sum of platform revenue and social value. The novelty in the model stems from the heterogeneity in customers’ CTRs for each merchant’s product and the exclusivity of customer datasets initially held by individual merchants. This modeling choice is motivated by the existing inefficiencies of merchants’ individual targeting efforts outside the platform using their exclusive customer datasets. Such decentralized efforts yield both smaller audience reach and reduced targeting precision due to data sparsity ( Lambrecht and Tucker 2013 , Bleier and Eisenbeiss 2015 ). On the platform, however, participating merchants compete for the ability to target any customer within the shared dataset. Given a large aggregated dataset and heterogeneity in customers CTRs, the platform can potentially design targeted ads that deliver higher expected clicks for most—if not all—merchants. Using a mechanism design framework, I solve for the optimal mechanism that improves upon individual targeting efforts while ensuring merchants’ willingness to share their customer data. I show that the optimal advertising (direct) mechanism allocates targeted ads in bundles according to a non-decreasing merchant-specific scoring rule that is a function of each merchant’s (type). Specifically, each customer in the dataset is targeted by the merchant with the highest quality score defined as the weighted merchant-specific score by the customer-specific CTR. In the case of ties, the designer constructs a customer-specific tie-breaking rule that ensures participation of every type. Due to the differences in CTRs and merchants’ types, some poaching may still occur in the optimum. To alleviate poaching concerns, adversely affected merchants with negative net expected clicks receive positive monetary transfers to guarantee their participation and often give strictly surplus. By allocating targeted ads through a single merchant-specific scoring rule, the designer pools merchants’ misreporting incentives for different ads—hence reducing information rents. Consequently, a single bid determines each merchant’s entire on-platform advertising campaigns: which of her own customers she continues to target and which new customers from other merchants she gains access to. On the platform, merchants anticipate retaining advertising access to only their high CTR customers, while selling the rest to the designer. Simultaneously, they expect to buy high CTR customers from competing merchants’ datasets. This creates countervailing incentives : merchants have incentives to overestimate data value when selling and underestimate it when buying. The designer leverages these countervailing incentives by constructing a single scoring rule—based on each merchant’s initial dataset—to optimally design menus of data bundles and non-itemized transfers that essentially describe on-platform advertising campaigns. Designing data bundles often involves resolving for each customer tie-breaks among merchants with identical quality scores, making this a multidimensional mechanism design problem, hence computationally untractable. In the second part of the paper, I extend the model by