News · 2026-09-20

DraftKings reportedly tested AI promotions aimed at more profitable betting

DraftKings reportedly tested a machine-learning ‘elasticity’ model that ranked casino customers by whether promotions would make them gamble and lose more money, while the company’s own filings show AI-personalized promotion at substantial scale. The reported system is not evidence that DraftKings diagnosed addicts or caused a particular person’s losses, but it raises a concrete governance question: why are some behavioral predictions permitted to steer marketing while harm predictions are rejected as too uncertain?

Key facts

• The New York Times reports an initial September 2023 test involving approximately 5,000 casino players.
• The reported model estimated promotional response and expected profitability, with scores refreshed weekly.
• DraftKings says it automated and personalized $400 million of promotional spending in 2025 through AI.
• Primary news source: the New York Times investigation.

The phrase ‘AI targeting addicts’ is stronger than the available evidence. The NYT says it reviewed internal documents and spoke with former employees about an elasticity model. Its purpose was business-like and coldly specific: estimate whether promotional incentives would produce enough extra gambling and loss to justify the offer. Lower-scoring, ‘inelastic’ customers received fewer incentives; higher-scoring customers remained eligible for more promotional investment. The report describes free bets, profit boosts, and deposit bonuses delivered through email and phone alerts.

That is a prediction problem, not mind reading. Picture an online shop deciding who receives a discount coupon. The model looks for people whose behavior is likely to change enough to make the coupon profitable. In gambling, the object being optimized is more troubling because the desired behavior is wagering, and player losses are often revenue. The NYT report says the system used recent betting behavior and was refreshed weekly. It does not provide source code, feature weights, model accuracy, thresholds, a final rollout population, or evidence that a promotion caused any individual loss.

DraftKings’ public materials independently establish the surrounding commercial system. Its 2025 Form 10-K says it uses personalized promotions, including free bets and matching deposits, to attract, retain, and re-engage users. Its 2026 Investor Day presentation says the company ‘automated and personalized’ $400 million in 2025 promotional spending through AI and marketed the ‘right message’ to the ‘right user’ at the ‘right time.’ That does not prove the NYT-described model controlled every dollar, but it does corroborate the scale and direction of AI-assisted personalization.

The most consequential reported asymmetry involves responsible gambling. The NYT says a separate team developed a predictive risk model using behavior such as deposits, withdrawals, and attempts to chase losses, but DraftKings decided not to use predictive technology for problem gambling because it was not sufficiently evidence-based. DraftKings says its trigger-based approach is better. The company’s response to the article calls the promotion pilot preliminary and inconclusive and says promotions go to customers demonstrating sustained engagement, not merely because they lost money. Both statements should be held in view.

The strongest counterargument is that this is ordinary segmentation with an AI label pasted on it. A straightforward classifier can rank expected promotional return, and companies have long sent different offers to different customers. That technical observation is true, but it does not dispose of the policy issue. In a setting where harm is foreseeable and measured behavior is rich, a low-tech optimization can still be ethically consequential. The decision to operationalize a revenue model while shelving a risk model is a governance choice whether the implementation is a neural network or a spreadsheet.

Regulation is catching up unevenly. New York’s Gaming Commission notice considers restricting AI-powered personalized promotions, suggested wagers, and suggested wager amounts. It is a consultation, not a final rule and not a case against DraftKings. New Jersey’s responsible-gaming best practices show a different model: automated signals such as escalating wagers and withdrawal cancellations should trigger education and human intervention.

The caveat is crucial for a fair story. The NYT reporting is strong journalism, but the underlying internal model materials are not public. It does not establish the share of targeted customers who had gambling disorders, the geographic deployment, current model use, or causal harm. The clean conclusion is still powerful: behavioral prediction is no longer a future risk in gambling marketing. It is a deployed commercial capability, and policymakers must decide which objectives it may lawfully and ethically optimize.

The source distinction matters: the NYT account reports on unpublished materials, while DraftKings filings corroborate only the broader personalization machinery. That evidentiary boundary should remain visible.

Primary source, verified: read the paper →

Ground Truth. (2026, September 20). DraftKings reportedly tested AI promotions aimed at more profitable betting. Ground Truth. https://groundtruth.day/news/draftkings-tested-ai-promotion-elasticity-model.html

@misc{groundtruth:draftkings-tested-ai-promotion-elasticity-model,
  title  = {DraftKings reportedly tested AI promotions aimed at more profitable betting},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/draftkings-tested-ai-promotion-elasticity-model.html}
}

Topics: ai-governance · machine-learning · consumer-protection · personalization · policy