DraftKings built a model to score bettors by expected losses

DraftKings built a model in 2023 that scores bettors by how much they're expected to lose, according to Tech Times' reporting on a New York Times investigation. The company called the metric "elasticity."

Executives credited AI-driven promotions with a 13% improvement in sportsbook margin in 2025. The model reads play frequency, account balances, and loss-to-wager ratios, then uses that to direct hundreds of millions of dollars in promotional spend. Former employees told the Times its logic surfaced problem gamblers as the most lucrative group to target with those promotions.

DraftKings reportedly built a second model using the same underlying data. That one was meant to score gambling-harm risk. The project got shelved.

DraftKings says it "rejects any implication" that its marketing unfairly targets customers. The reporting doesn't include the company's own account of why the harm-risk model didn't ship.

I've built scoring models. The technical distance between flagging high-value users and flagging users showing addiction markers is close to zero. You're pulling the same features either way. What separates a defensible targeting model from a predatory one is the decision layer sitting on top of the math. Does a flagged user get a bonus offer, or a cooldown prompt. DraftKings built the second model and didn't ship it. That's the fact a regulator asks about first, and it's what turns a PR problem into a legal exposure problem.