Abstract
Firms increasingly use algorithms to personalize products and services. A common concern is algorithmic monoculture: providers converging on similar predictions and delivering homogeneous offerings, thereby hindering social welfare. We build a competition model incorporating noisy user preferences that endogenizes providers’ algorithmic targeting decisions and introduce an interpretable metric for the level of monoculture. The model yields two actionable insights. First, duopoly creates a monoculture trap: both firms target the majority user segment. Second, with three or more competitors, a threshold effect in minority share endogenously induces some firms to serve minority users, thereby reducing monoculture. With heterogeneous providers, as the stronger provider’s advantage grows, the unique equilibrium is mixed—both randomize across labels, and the weaker provider more often serves the minority, breaking monoculture.
Using a welfare metric grounded in users’ realized experiences, we prove that competition increases social welfare for both majority and minority users, with larger gains for the latter. These results provide regulators and managers with concrete policy levers: maintain at least three effective competitors or promote algorithmic heterogeneity to safeguard welfare; and monitor a simple monoculture index to diagnose uniform outcomes without micromanaging individual models.