A New York Times investigation published Saturday alleges that DraftKings, one of the largest sports betting operators in the United States, built a machine-learning model designed to identify which customers were most likely to lose money — and used those scores to guide who received bonus bets and free-play promotions.
The report describes a system where the same predictive technology that made promotions more profitable could have been used to spot customers at risk of gambling harm. According to the Times, that protective application was started but never finished.
How the Model Worked
The model, built in 2023, scored DraftKings customers by how much money they were expected to lose after receiving a free bet or bonus, according to the Times. To make that prediction, it examined signals including how frequently a user played, their account balances, their loss-to-wager ratios, and a separate estimate of whether a user was likely to stop gambling altogether.
A higher score meant a customer was predicted to generate more revenue per promotional dollar spent. In other words, the people most likely to receive free bets were, by design, the people the algorithm expected to lose the most.
Jayden Butts, a former DraftKings data analyst, described the underlying logic to the Times: "We are looking for traits and features that we can target that indicate a good investment."
The Business Results
The approach appears to have worked commercially. DraftKings executives said data science and analytics improved promotion-driven sportsbook margins by 13% in 2025, and that AI helped personalize hundreds of millions of dollars in promotional spending, according to the Times.
That figure illustrates why predictive targeting has become central to the sports betting industry's economics. Bonus bets are expensive customer acquisitions; a model that predicts expected losses can direct each promotional dollar toward the users most likely to convert them into sustained wagering.
The Tool That Wasn't Finished
The investigation's sharpest finding concerns what DraftKings chose not to build out. Former employees told the Times that a separate initiative to create risk scores identifying customers with potential gambling problems was shelved — even though, per the reporting, the underlying data infrastructure could have supported it. The infrastructure to look in both directions reportedly existed; only the profit-oriented model was completed.
DraftKings pushed back on the framing. The company's chief responsible gaming officer, Lori Kalani, told the Times that DraftKings monitors customers for risky behavior, and said the company declined to deploy risk-prediction technology because it was not sufficiently evidence-based. DraftKings also told the Times it "rejects any implication" that its marketing unfairly targets customers, maintaining that promotions go to users who show sustained, engaged platform use rather than users ranked by how much they lose.
The company publicly highlights responsible-gaming initiatives including a collaboration with Mindway AI's Gamalyze tool and a partnership with IC360 for integrity and compliance monitoring. Whether those tools close the specific gap described by former employees — proactively scoring and protecting users at elevated risk — remains unclear based on available reporting.
An AI Ethics Test Case Beyond Gambling
The story lands amid intensifying scrutiny of how consumer-facing companies use predictive AI. The core tension is not that the technology exists — scoring users by predicted value is standard practice across subscriptions, advertising and fintech — but what happens when the same model that identifies the most profitable customer also identifies the most vulnerable one.
Regulators have been moving in this direction for some time. In the gambling sector specifically, officials have questioned the industry's use of data to shape player behavior, and the Philadelphia area has seen litigation over DraftKings' promotional practices. If the Times' reporting holds up, it gives lawmakers and gambling regulators something more specific than general industry concern: an alleged case where harm prediction and profit prediction were technically feasible, and only one was finished.
Для користувачів практичний висновок незручний. Безкоштовна ставка, яка надходить у вашу поштову скриньку, може відображати кероване даними судження щодо вашої ймовірності програшу, а не винагороду за лояльність. І модель, яка робить це судження, знає речі — як часто ви граєте, скільки ви гонитесь за програшами, чи демонструєте ви ознаки зупинки — про що ви ніколи явно не розповідали з цією метою.
DraftKings не звинувачують у порушенні закону, а цільові акції є законними. Але оскільки системи штучного інтелекту все більше вирішують, хто бачить яку пропозицію, розрив між тим, що компанії можуть передбачити, і тим, що вони вирішують діяти, стає центральним питанням споживчої етики штучного інтелекту — і звіт DraftKings є одним із найяскравіших прикладів цього розриву на практиці.
Будьте попереду ШІ
Від додатків для ставок до передових лабораторій, ШІ змінює те, як компанії бачать своїх клієнтів. Щоб дізнатися більше про етику ШІ та галузеву підзвітність, читайте AI Buzz Wire.
Читати останні новини ШІ →
