The New York Times just published an investigation called “How DraftKings Uses A.I. to Target the Gamblers Likeliest to Lose.”
The key takeaway: an algorithm trained on a massive dataset can predict how likely a player is to spend, and how much, if you offer them a promo. The same data can also tell you which players are going to develop a problem, but the company only chose to do the first part. And wouldn’t you know it, the player who’ll spend the most is probably the one who’ll end up with a problem. I hope nobody’s shocked to hear that the company was more than happy to figure out which players would pay off, but for some reason didn’t see the need to figure out which ones were in trouble.
Here’s what I found most interesting.
DraftKings was spending hundreds of millions of dollars a year on promos but knew very little about whether they worked. For a sense of scale: last year the company pulled in about $8.7 billion in gross revenue from sports and casino players and gave out about $3 billion in promos. DraftKings now has 11 million customers, up from 5 million in 2022.
“The best investment would be a problem gambler.” — Jayden Butts, former DraftKings data analyst
The company built a model designed to find the people it could get to lose the most. Some of the employees who worked on it now regret it and call it dangerous.
The model scored every player, and the highest score went to whoever would lose the most per promo. Internally, that score was called “elasticity.” It’s a nice clean econ term, and it’s there to hide an ugly truth. It’s just morally easier to say “this player is elastic” than to call things what they are.
And guess what turned out to be the most elastic?
Online casino games, including slots and blackjack, bring in almost a third of the company’s revenue, even though they’re only offered in a handful of jurisdictions. Slots revenue was “more elastic” than other casino games, meaning promos were especially good at pushing people toward slots. Players with high elasticity scores also played more slots on average than players with low scores. And slots are widely considered one of the most addictive forms of gambling.
Built into the elasticity model was another one that predicted who was about to quit gambling, and a separate system tried to win those players back. In other words, the company went looking for the people who were trying to stop.
DraftKings itself says its promos go to active, engaged players, not to the ones who lose the most. But a model that ranks people by how much they’ll lose per promo speaks for itself.
DraftKings told investors that data science and analytics helped it boost margins on promo-driven sports bets by 13% in 2025.
Meanwhile, DraftKings stalled or killed models for spotting problem gamblers, supposedly because there was no evidence they were useful. And really, what’s the use in turning away the players who lose the most? The conflict of interest is baked in.
Employees built a model that gave players risk scores, but the company shelved it. Its competitors, FanDuel, Fanatics and PrizePicks, say they use risk scoring. The head of responsible gaming at PrizePicks said that without this kind of technology, he couldn’t look you in the eye and say the company was doing its best.
That said, I wouldn’t jump to the conclusion that DraftKings is the villain and everyone else is a saint. It’s still unclear how the competitors actually use risk scoring. And the same article mentions that FanDuel has also bragged about using customer data to personalize promos, without explaining what that means. My guess is nobody’s done a deep investigation into them yet.
Responsible gaming at DraftKings is handled by a team of about 50 people, at a company of 5,000.
DraftKings’s head of responsible gaming compared gambling to shopping. Other industries have the same issues, she said. “Shopping can be problematic for people.”
The only example of “protective” AI the company gave: language models read what players write to customer support and look for signs of distress. So they only catch the people who speak up.
The piece also talks about how players see promos as free money when really they just pull you back in, how inboxes get flooded with offers, and how hard it is not to play again even when you’re in recovery. But I think you already know all that.
Google searches like “Am I a gambling addict?” are on the rise. Calls about sports betting to Ohio’s problem gambling helpline have more than quadrupled since the state legalized it in 2023.
The regulator recommends setting up automatic triggers to flag problem gamblers. DraftKings won’t really say how that works in practice: it won’t share the thresholds or how many accounts it has closed, so there’s no way to check. My guess is not very aggressively.
In my view, regulators should require this, not recommend it, and independent audits should check whether the rules are actually being followed.