The online gambling giant was spending hundreds of millions of dollars every year on promotional incentives: “free” betting money advertised through emails and phone alerts. But the company knew little about their effectiveness.
So in 2023, DraftKings took customer betting records and built a machine learning model, a form of artificial intelligence that seeks patterns in data, to answer the question: Who was more likely to respond to promotions by gambling — and losing — more?
Butts had reason to be concerned. DraftKings makes money when gamblers lose money. And the model sought to identify those it could get to lose the most. It scored each customer based on their habits: The higher the score, the more money a gambler was likely to lose for each promotion offered.
Since Butts ran those tests, DraftKings has continued to hone its methods, using data science, to target losing gamblers with promotions that encourage more betting, according to six former employees who worked on them. At the same time, four other former employees said, DraftKings has stalled or squashed efforts to use similar technology to predict who might develop a gambling problem based on their betting activity.
Silicon Valley firms spent years analyzing every digital interaction to predict what will keep users clicking on advertisements. Now, as companies like DraftKings have made gambling accessible to millions on smartphones, they too have collected an extraordinary wealth of data.
An investigation by The New York Times shows what DraftKings has chosen to do — and not do — with that power.
The Times interviewed more than 40 former DraftKings employees and obtained internal research memos, presentations and Slack messages as well as betting records from experiments conducted on customers.
The documents show how the model Butts worked on analyzed dozens of data points for gamblers, including how frequently they played, their daily account balances and how much they typically lost compared with how much they bet. It also incorporated another model that calculated how likely a user was to stop gambling.
This betting data may also contain signs that a person is headed for trouble. Yet when employees developed a machine learning model that would have assigned users “risk scores,” the company sidelined it, according to two former employees who worked on that project.
In late 2024, DraftKings fired Butts for performance reasons, he said, amid a short blip in business that led some in the company to believe its promotional experiments weren’t working as intended. The company briefly paused some of its data science work — before revving back up again with a flurry of new machine learning projects.
Butts and five other former DraftKings employees who worked on promotional targeting told the Times they regretted building technology they now viewed as dangerous.
“It is as predatory as it sounds,” said a former DraftKings analyst who, like many interviewed for this article, requested anonymity because he feared retribution. “If you lose more, we give you more, so you keep playing more.” He quit in 2024. [...]
Promotions, which take on forms like a free bet, a “profit boost” or a deposit bonus, play a vital role in DraftKings’ business: The company brought in around $8.7 billion in gross revenue from sports and casino gamblers last year and gave out about $3 billion in promotions, according to research by Citizens Bank.
DraftKings and some competitors, including FanDuel, have boasted publicly about their use of customer data for promotions — without disclosing what those efforts entail. A DraftKings executive recently told investors that data science and analytics helped it improve its margins on promotion-driven sports bets by 13% in 2025 and that it used AI to personalize hundreds of millions of promotional dollars.
One former DraftKings data scientist who worked on promotions, Jacob Shulkin, said that they were effective because they took advantage of gamblers’ psychology. “I feel I’m getting free money,” he said, “but really, it’s dragging me back in.”
Several gamblers told the Times that promotions fueled their addictions. Bryan Biehl lost nearly $70,000 gambling online, more than half of it at DraftKings. Biehl recalled how in late 2024, when he started therapy for his addiction, his email inbox began to feel like a relapse risk.
“I would get flooded with bonuses and deposits,” Biehl said. “If you are in addiction, you are not going to say no.”
In the first two weeks of December 2024, Biehl received 40 promotions from DraftKings, emails show. He succumbed to temptation one last time on Christmas Day before putting himself on self-exclusion lists, which blocked him from gambling apps.[...]
DraftKings already had a system that weighed factors like a gambler’s skill, how much they bet and tax rates on gambling revenue in the state where they lived.
Figuring this out required machine learning. Unlike traditional data analytics, where researchers decide which patterns to look for, machine learning models can sift through hundreds of variables on their own to find combinations that help predict particular behaviors.
Data scientists had trained the new casino model on historical data. It was Butts’ job to test it on real customers. Each week, the model vacuumed up information about a user’s recent activity. The score it calculated was known internally as “elasticity,” a term borrowed from economics.
Users with below-average scores were deemed “inelastic” and marked for fewer incentives. The “elastic” bettors remained.
In September 2023, Butts tested using the elasticity model to influence promotions for about 5,000 casino players. He later expanded the tests to a larger population.
He initially thought DraftKings aimed to save money by avoiding people who were unlikely to be profitable. But he said his supervisors told him that the company did not want to reduce its promotional spending but rather to “redeploy” it. He understood this to mean the goal was to direct more promotions toward the biggest losers.
Data scientists had trained the new casino model on historical data. It was Butts’ job to test it on real customers. Each week, the model vacuumed up information about a user’s recent activity. The score it calculated was known internally as “elasticity,” a term borrowed from economics.
Users with below-average scores were deemed “inelastic” and marked for fewer incentives. The “elastic” bettors remained.
In September 2023, Butts tested using the elasticity model to influence promotions for about 5,000 casino players. He later expanded the tests to a larger population.
He initially thought DraftKings aimed to save money by avoiding people who were unlikely to be profitable. But he said his supervisors told him that the company did not want to reduce its promotional spending but rather to “redeploy” it. He understood this to mean the goal was to direct more promotions toward the biggest losers.
Image: Tony Luong/The New York Times
