Subsidizing Sequential Search

Abstract: We study markets where firms compete for consumer attention by subsidizing costly product inspection. These subsidies do not change product quality, but they alter the order in which consumers search by lowering inspection costs. We establish a subsidy-sorting principle: in any equilibrium, higher-quality firms provide weakly larger subsidies, leading consumers to search in descending subsidy order. A unique equilibrium survives forward-induction reasoning in the spirit of the Intuitive Criterion: low-quality firms are never inspected, intermediate-quality firms separate with strictly increasing subsidies, and high-quality firms pool at the full subsidy. This equilibrium maximizes information revelation among all possible outcomes and ensures efficient inspection. We then extend the analysis to AI-mediated platforms that can create and price inspection tokens. The platform's optimal linear pricing leads to excessive inspection relative to the social optimum. While this distortion does not reduce consumer welfare, it reallocates surplus from sellers to the platform and consumers.

• The paper introduces a model where firms strategically subsidize consumer search costs to signal match quality.
• It establishes the Subsidy–Sorting Principle, ensuring higher-quality firms offer larger subsidies and consumers follow a descending-subsidy search order.
• The analysis uncovers a unique step-increasing-step equilibrium and welfare trade-offs from platform-controlled inspection token pricing.

Subsidizing Sequential Search: An Expert Analysis

Introduction and Motivation

The paper "Subsidizing Sequential Search" (2605.28985) offers a comprehensive equilibrium analysis of attention markets wherein firms compete to offset consumer search frictions not by altering inherent match quality, but by subsidizing the inspection cost. Unlike traditional frameworks where search cost is exogenous and fixed, this model endogenizes search frictions as contractible, strategic variables: sellers can reduce consumer cost through explicit monetary subsidies. This paradigm is particularly pertinent in agentic economies mediated by AI, where inspection is decomposed into computational, auditable events (e.g., queries, database access) whose cost can be directly priced and manipulated by platforms or sellers.

Model Architecture

The model comprises $n$ firms, each with a private type $t_j \in [0,1]$ corresponding to the match probability for a representative consumer. Types are drawn i.i.d. from a common prior. The consumer wishes to find at least one product matching her needs, gaining utility $u$ from a match; discovery requires costly inspection ($c > 0$ per evaluation). After privately observing $t_j$, each firm simultaneously chooses a subsidy $s_j \in [0, c]$, paying the consumer upon inspection. Consumers observe the entire subsidy profile $\mathbf{s} = (s_1, ..., s_n)$, update beliefs, and plan a sequential search to maximize expected utility. The cost of providing a unit of subsidy is $p$, which is also microfounded as the price per inspection token in the platform extension.

The interaction is cast as a symmetric Perfect Bayesian Equilibrium (PBE), with sequential search and signaling features. The analysis is performed both in the standard decentralized environment and in the case where a platform can centrally set $p$ to maximize revenue.

The Subsidy–Sorting Principle

One of the principal contributions is the formalization and proof of the "Subsidy–Sorting Principle." This result holds in all equilibria: higher-quality (higher $t$) firms offer weakly larger subsidies, and the consumer's optimal search order is descending in realized subsidy. The reasoning leverages single-crossing conditions in firms' payoffs w.r.t. inspection probabilities and subsidies, ensuring that monotonicity is strict when incentive compatibility binds. The structural upshot is that subsidy size acts both as a signal and as a search-friction-reducing instrument. The consumer follows a descending-subsidy index rule—a variant of Weitzman's Pandora's Rule—stopping when the expected surplus from inspecting the next product is non-positive.

Equilibrium Structure and Refinement

A key technical result is the unique equilibrium refinement, following forward-induction in the sense of Cho-Kreps and Banks-Sobel. The refinement, which only admits economically credible off-path beliefs, yields a unique symmetric equilibrium featuring a “step-increasing-step” (SIS) subsidy policy:

• Low types: Firms with $t_j \in [0,1]$0 (where $t_j \in [0,1]$1) pool at zero subsidy and are never inspected.
• Separating region: For intermediate $t_j \in [0,1]$2, subsidy $t_j \in [0,1]$3 is strictly increasing, producing full revelation of types within this interval and efficient sorting.
• Pooling at the cap: High types $t_j \in [0,1]$4 pool at the full subsidy $t_j \in [0,1]$5 due to the binding upper constraint.

This outcome is robust: under the equilibrium-dominance refinement, all other equilibria either admit strictly dominated off-path deviations or violate monotonicity.

Welfare, Platform Pricing, and Excess Search

A welfare decomposition distinguishes consumer surplus, producer surplus, and aggregate social efficiency. Several comparative statics follow directly: higher $t_j \in [0,1]$6 (intense competition) increases aggregate inspection and consumer surplus while compressing per-firm rents; higher $t_j \in [0,1]$7 or $t_j \in [0,1]$8 induces entry (and inspection) only for higher-quality types, raising entry thresholds and reducing both inspection intensity and producer surplus.

The agentic economy extension rigorously analyzes the implications of platforms that control the price per inspection (token pricing). Platforms maximize revenue by lowering token prices below the social optimum, inducing excessive search and pooling among high types. This mechanism increases both platform revenue and consumer realized matches (and welfare), but reduces total surplus due to over-inspection (efficiency loss relative to the Weitzman/virtual surplus optimum). Importantly, in the platform optimum, types with negative virtual values are still inspected—a sharp deviation from standard mechanism design benchmarks.

Numerical and Empirical Claims

The paper provides precise formulae for equilibrium subsidy schedules, attention probabilities, and match likelihood, and full characterization of all cutoffs and regions. Strong analytical claims include the uniqueness of the SIS equilibrium under refinement, strict sorting in all equilibria, and the direction of welfare comparative statics with respect to $t_j \in [0,1]$9, $u$0, and $u$1. Furthermore, the analysis of platform pricing provides an explicit revenue decomposition, boundary characterization for optimal caps, and conditions under which the platform's chosen price induces over-search.

Implications and Outlook

The analysis yields several direct implications for the design and regulation of attention markets, especially in settings where AI agents mediate search and computational events are auditable and contractible:

• Predictive features: Higher observed subsidies should covary with proxies for match quality, with congestion and inspection probability inversely related to the size of the top pool.
• Platform design: If platforms use linear pricing of attention tokens, profit maximization will generically induce informational pooling at the top and over-inspection, redistributing surplus from firms to both platforms and consumers.
• Empirical validation: The predicted structure of subsidies, search order, and welfare distribution provides sharp hypotheses for the study of platform logs and experimental attention allocation.

Theoretically, the work generalizes seminal signal-extraction and attention allocation frameworks, endogenizing search cost and linking them to platform pricing and algorithmic search mechanisms. Practically, as agentic search becomes standard, the results suggest channels where regulation (e.g., on inspection token pricing, transparency in ranking) could align platform incentives with welfare or at least discipline the degree of information loss due to pooling.

Future Directions

Open questions include the analysis of multiproduct search, dynamics (e.g., reputation effects over repeated interactions), correlated types or externalities between firms, richer pricing contracts, and platforms employing non-linear pricing for attention. Further, empirical calibration using search and inspection logs from agentic platforms could substantiate the mechanisms identified here and guide policy or platform engineering for efficient attention allocation.

Conclusion

"Subsidizing Sequential Search" unifies the economics of directed search, attention allocation, and platform pricing in environments where sellers can directly subsidize consumer inspection. The robust characterization of equilibrium structures, analytic comparative statics, and the welfare implications for both decentralized and platform-centered markets comprise a rigorous foundation for subsequent work in both economic theory and practical mechanism design for AI-mediated marketplaces.