DraftKings Built AI to Predict Which Bettors Would Lose More After Promos - Scandal Explained
Key Points
- DraftKings began testing a machine learning model in 2023 that scored casino customers based on how much they were expected to lose after receiving promotional offers.
- Similar technology was later developed for sportsbook promotions, with the Times reporting that DraftKings continued refining its targeting methods into 2026.
- DraftKings said at its March 2026 Investor Day that AI automated and personalized $400 million of promotional spending during 2025.
- Multiple internal efforts to predict which customers were heading toward gambling problems were shelved
DraftKings AI Model Scored Customers by Expected Losses After Promotions
An investigation published by the New York Times, found that DraftKings built a machine learning model using customer betting records to estimate how people would respond to bonuses and other incentives. The Times based its reporting on interviews with more than 40 former employees, along with internal research, presentations, Slack messages and betting records from customer experiments.
The model went beyond predicting whether somebody would use an offer. It attempted to estimate how much money an individual customer would ultimately lose after receiving one, giving DraftKings a way to determine where promotional spending was expected to generate the greatest return.
The model ranked casino customers by promotional value
DraftKings began testing the system on online casino customers in 2023. The model analyzed dozens of variables, including betting frequency, daily account balances, previous losses relative to wagers and the likelihood that a customer would stop playing.
Customers received an internal "elasticity" score. Higher scores indicated that DraftKings expected to generate more revenue from that customer after deploying a promotion, while lower-scoring customers could receive fewer incentives.
Former DraftKings analyst Jayden Butts tested the model on roughly 5,000 casino customers in September 2023 before the experiment expanded. Butts told the Times in an on-the-record interview that, under the financial logic of the system, "the best investment would be a problem gambler."
Around the same period, DraftKings began developing similar models for sports betting promotions, according to former employees interviewed for the report. DraftKings has continued working on promotional targeting systems as recently as 2026.
The scale is no longer experimental. At its March 2026 Investor Day, DraftKings said AI had automated and personalized $400 million of promotional spending during 2025, while sharper segmentation and decision-making helped improve its net revenue margin on promotional sportsbook wagers by more than 1,300 basis points year over year.
Problem gambling prediction projects were shelved
DraftKings employees were also exploring how machine learning could identify customers at risk of developing gambling problems before they reached a more obvious crisis.
In 2024, data scientist Nestor Hernandez began developing a model using factors including deposits, withdrawals and apparent loss-chasing behavior to generate customer risk scores. The goal was to give DraftKings an earlier warning that a bettor might require intervention.
Hernandez left DraftKings before the model was completed, and other employees continued the project. Former employees told the Times that the project was eventually shut down, while two additional attempts to build similar predictive systems were also shelved.
DraftKings chief responsible gaming officer Lori Kalani told the Times that company leaders collectively decided against using predictive problem gambling models because they did not consider the approach sufficiently evidence-based. She said DraftKings believes its existing monitoring system provides a better way to identify potentially risky behavior.
DraftKings says it already monitors risky gambling behavior
DraftKings strongly disputed the suggestion that its marketing practices improperly target customers because they lose money. The company told the Times that promotions are directed toward customers demonstrating sustained engagement with the platform and called Butts' 2023 testing preliminary and inconclusive.
The company says it monitors more than two dozen behavioral indicators nationwide, including changes in deposits and playing patterns. Depending on what its systems detect, DraftKings can send responsible gambling information, provide educational material, ask customers to complete questionnaires or close an account.
DraftKings has also continued adding consumer-facing safeguards. On September 14, the company announced expanded responsible engagement tools, including customizable cool-off periods and another Gamalyze product developed with Mindway AI. Kalani said in that announcement that responsible engagement is embedded in how DraftKings operates.
Why This Matters For Bettors
Bettors already know sportsbooks do not hand out bonuses for charity. Promos are customer-acquisition and retention tools, and the operator expects to make that money back somewhere.
What makes the Times reporting more troubling is the level of precision DraftKings was reportedly trying to bring to that calculation. The company was not simply measuring whether promotions increased betting activity across a broad group of customers. Its model attempted to estimate how much individual users would lose after receiving an offer and score them according to their expected value to DraftKings.
That creates an enormous information imbalance. DraftKings can analyze betting frequency, deposits, balances, previous losses and other behavior at a scale no customer could realistically replicate. If those signals tell the company that giving one bettor another $25 in bonus bets is likely to generate significantly more revenue, that offer starts to look very different from a simple reward for being a loyal customer.
Using that data to make promotional spending more efficient is a normal commercial goal for a sportsbook. The concern is what happens when the behaviors that make a customer attractive to the marketing system may also be signs that the customer needs intervention. DraftKings continued developing increasingly sophisticated promotional targeting while several internal efforts to predict gambling harm were shelved.
That leaves bettors with a fairly basic question: which system wins when the same customer looks valuable to one side of the business and vulnerable to the other?
DraftKings says its existing responsible gambling controls monitor risky behavior without relying on those predictive models. But if the company can use customer data to identify when another promotion is likely to produce more losses, bettors have reason to expect its safeguards to be capable of recognizing when that promotion should never be sent.
What Happens Next
The Times investigation does not by itself establish that DraftKings violated a gambling regulation. Any regulatory consequences would depend on the rules in individual states and what regulators determine about how the models actually operated.
The most obvious questions concern the data behind promotional targeting: which behavioral signals influence offers, how customers showing potential harm are excluded and whether responsible gambling teams have sufficient access to the predictive capabilities already being used elsewhere in the business.
DraftKings has a clear counterargument in its existing monitoring system and growing suite of player-control tools. The pressure point now is whether regulators consider those safeguards sufficient as promotional technology becomes increasingly personalized and predictive.
Cole cut his teeth as a sportswriter in Texas, covering everything from Longhorns games to small-town Friday night lights. A lifelong bettor stuck with offshore books for over a decade thanks to Texas' slow path to legalization, he eventually found his way into the world of social sportsbooks - where he uncovered a fast-growing, community of bettors.
Today, he writes for the millions of Americans in states without legal books, helping them explore safe ways to bet without running afoul of the law.
As editor-in-chief, he aims to keep BettingScanner honest, human, and grounded in what bettors actually care about: fairness, fun, and finding your lane - even when the state won’t give you one.
