Scores as Infrastructures
Automated exposure, reliability and strategic discipline in advertising platforms.
Overview
Digital platforms increasingly use scores not only to transmit information, but also to determine access to automated allocation mechanisms. In advertising markets, for example, a favorable eligibility score can give an ad access to additional auctions, demand and exposure.
This project studies the strategic consequences of this dual role. A platform privately observes an imperfect signal about the quality of an ad and then assigns a public favorable or unfavorable score. After a negative signal, it may strategically upgrade the ad in order to obtain additional demand and exposure.
Main mechanism
Upgrading a low-quality ad creates an immediate benefit, but it can be detected through probabilistic auditing. Detection damages the credibility of the scoring system and moves the platform into a persistent low-trust state in which the score becomes uninformative and automated exposure falls to its minimum level.
The model therefore creates a feedback loop: the scoring rule determines the reliability of the score, reliability determines how much exposure downstream actors delegate to it, and this exposure changes the platform's incentive to preserve the score's credibility.
Main results
- Derives the incentive constraint governing truthful scoring by comparing the immediate gain from upgrading with the expected loss of future platform value.
- Identifies a central quantity, G, which determines when exposure and strategic discipline are complementary.
- Shows that when G > 0, greater exposure can make truthful scoring easier to sustain because the future value destroyed by a loss of credibility grows sufficiently strongly.
- Endogenizes exposure as an increasing function of score reliability and obtains a high-reliability truthful equilibrium and a low-reliability pooling equilibrium.
- Characterizes a unique interior equilibrium separating the two regimes. Under the adjustment dynamics used in the paper, the two extreme equilibria are locally stable while the interior equilibrium is unstable.
- Shows that stronger auditing can support the high-trust equilibrium and shift the tipping point in its favor, but does not eliminate the degraded low-exposure equilibrium.
In the reference calibration, the unstable tipping point occurs at an upgrading probability of approximately 0.370. Below this point, the platform prefers not to upgrade after a negative signal; above it, upgrading becomes the preferred action.
Interpretation
The model highlights a form of bistability. A reliable score can sustain high automated exposure, which increases the value of preserving its credibility. But once the system moves sufficiently far toward strategic upgrading, lower reliability reduces exposure and weakens the future value that disciplines the platform.
As a result, stronger monitoring can improve the resilience of a trusted system without necessarily being sufficient to restore a system that is already trapped in the degraded equilibrium.
Methods
The project combines dynamic Bayesian games, signaling, reputation mechanisms and nonlinear equilibrium analysis. The analysis characterizes incentive constraints, stationary equilibria, local stability and comparative statics with respect to exposure and auditing intensity.
Research context
This project was developed during a six-month research internship at CNRS – HEC Paris under the supervision of Marie Laclau. The motivating application is advertising-platform eligibility scoring, although the mechanism can apply more broadly to digital platforms where scores simultaneously transmit information and control access to automated allocation.