$400 million. That’s how much DraftKings reportedly distributed in AI-automated promotions in 2025. Executives reportedly credited that targeting with improving promotion-driven sportsbook margins by 13%. The New York Times investigation behind those figures drew on more than 40 former employees and internal documents. It described a machine-learning model DraftKings built in 2023 that scored users on “elasticity,” meaning how strongly they respond to free bets and other offers. On Thursday, Massachusetts became the first regulator to act on it. The MGC will review how DraftKings uses AI and then look at every licensed operator.
The detail I can’t get past is how simply one former employee described the model’s purpose. Jayden Butts told the Times his job was to help answer whether a user would give DraftKings more than the company gave that customer. If the answer was yes, the company could “open the floodgates” with promotions. The Times also reported that slots players scored as more elastic, and that early 2024 data showed highly elastic slots players spent more than less elastic ones. DraftKings says it “rejects any implication that its marketing practices are unfair or improperly targets customers.” The company says promotions go to customers who show “sustained, engaged use” of the platform, not to customers based on their losses.
The reason this goes beyond a PR problem is that Massachusetts already has a rule covering it. Under 205 CMR 257.02, operators can’t use “any computerized algorithm, automated decision-making, machine learning, artificial intelligence, or similar system” that they know or reasonably expect will make the platform more addictive. Commissioner Paul Brodeur said Massachusetts may be the only state that references AI in its regulations. So the MGC doesn’t need a new rule. It needs to decide whether a model that optimizes promotions for elasticity is one DraftKings “reasonably expected” to increase addictive play. Chair Jordan Maynard kept his first step modest: “I’ve requested that Executive Director Dean Serpa and the team engage with the operator to understand the specifics that were reported.” He said any further action “can then be determined by the commission when and if appropriate.”
The other part of the Times report may matter more. Former employees said DraftKings also worked on a model to predict when a customer was heading toward a crisis and might need intervention. One former employee said it was showing promise, but a planned early-2025 presentation on it was canceled, and two similar efforts were reportedly shelved. DraftKings’ chief responsible gaming officer, Lori Kalani, told the Times the evidence showed that kind of risk modeling wasn’t helpful. That’s awkward for the industry’s usual argument against AI restrictions, which is that the same tools help identify problem gamblers. If the targeting model went into production and the protection model got canceled, regulators will find that argument harder to accept.
Across the industry, AI personalization is how operators grow margins in an increasingly competitive market, especially with prediction markets taking a big share of new downloads. The key date here is only a week old: Massachusetts announced its review five days after the Times story ran. If the MGC finds the elasticity model runs up against 257.02, every operator with a promotions engine will have to ask the same question about its own tools, in every state that decides to copy Massachusetts.
