Research

We study how preferential access to consumer data shapes competition and data management strategies in digital display advertising markets. We develop a model in which two demand-side platforms (DSPs) compete to purchase a single advertising slot through a first-price auction. The two DSPs differ in their access to consumer data: one is vertically integrated within the ad-tech stack and consequently observes consumer preferences with positive probability, while the other receives only a noisy signal. We characterize the unique Bayesian Nash equilibrium of the auction and show that greater data access for the integrated DSP softens competition, increasing expected payoffs for both bidders. At the same time, enhanced data availability improves ad relevance for consumers. We then study the effects of a mandated data-sharing remedy, deriving conditions under which forced data sharing imposes a net cost on the integrated platform even under the most favourable monetisation terms, offering a formal explanation for the persistence of data silos in digital advertising markets.

I study optimal screening on a platform where seller quality generates aggregate demand externalities. Heterogeneous sellers invest in quality at private cost, while consumers decide whether to participate based on the aggregate quality of the platform catalogue. In the headline model, the seller-side menu is observable to consumers, so the platform's tariff affects participation through the quality schedule it induces. I characterise the direct benchmark and compare regimes in which the platform can and cannot charge a consumer fee. With both instruments available, the fee absorbs the bargaining parameter governing surplus division, leaving the optimal quality schedule independent of incidence. When the fee is unavailable, the quality schedule is distorted, and an isoelastic-uniform example shows that the one-sided regime delivers weakly lower quality than the two-sided benchmark. I then study implementation. A standard nonlinear tariff implements the benchmark only as a selected equilibrium and may support multiple fixed points, including zero trade. A state-contingent insulating tariff restores unique implementation when realised market size is verifiable.

Within AI literature, an enthymeme – a hidden premise rarely explicitly stated but materially significant – dwells, used to support the case for rational agents as an efficient form of organizing artificial cognition. We recognize this enthymeme, arguing that the agentic form is efficient for acting on given preferences, while artificial agents are deployed to respond to complex and dispersed demand that is still in formation when encountered. We argue that gathering and resolving such demand is not a preprocessing step but a form of cognitive work in its own right, and that an integrating centre is worth building only where interdependence among demands is dense, frequent, and costly to reverse.