DraftKings Built Machine Learning Models to Target Its Biggest Gambling Losers, Report Says
The sportsbook reportedly used predictive models to identify and retain customers likely to lose the most money, raising fresh questions about AI-driven gambling harm.
DraftKings developed machine learning models designed to identify customers who were likely to lose large sums and then target them with retention offers, according to a report drawing on the company's own practices. The claim puts one of the largest US sportsbooks at the center of a growing argument over how predictive analytics are used in consumer gambling products.
The mechanics described are familiar from other consumer industries. A model trained on account data — bet size, frequency, deposit patterns, time spent in the app — can produce a score estimating how much revenue a given customer will generate, and how likely they are to quit. Marketing and promotions then follow that score. In gambling, the customers with the highest projected losses are also the ones most likely to be experiencing harm, which is what separates this use case from, say, predicting who will renew a streaming subscription.
DraftKings has expanded aggressively since the US Supreme Court cleared states to legalize sports betting in 2018. The company now operates in dozens of jurisdictions and spends heavily on customer acquisition, including the free-bet promotions that drove search traffic around this story. Free bets are a standard acquisition tool across the industry, but they also function as a data-collection funnel: a new account generates behavioral signals that feed directly back into the models.
The report lands as regulators sharpen their focus on AI in gambling. Several state gaming commissions have begun asking operators to explain how algorithms shape promotional offers and whether problem-gambling safeguards are applied consistently. Federal attention has grown too, with lawmakers questioning whether sportsbook marketing reaches vulnerable users. Operators argue their models serve legitimate business purposes, such as fraud detection and responsible-gaming flags, and that any targeting is bounded by licensing rules.
What is not yet established is exactly how DraftKings' models were configured, who oversaw them, and whether any internal limits existed on promoting to high-loss accounts. Those are the questions likely to drive the next round of reporting and any regulatory inquiry. For now, the episode makes a broader point: as AI moves deeper into consumer finance and entertainment, the line between personalization and exploitation is drawn less by the technology than by the targets a company chooses to optimize.
DraftKings has not issued a detailed public response to the characterization of its modeling. Watch for state gaming regulators to request documentation, and for shareholder and responsible-gambling groups to press the company on what its models optimize for. The answer will shape how every sportsbook in the country defends its use of customer data.