Product1 publisher3 min readPublished
Former DraftKings staff describe a model that ranked bettors by how well promotions worked on them
The New York Times, citing more than 40 former employees, says a 2023 elasticity model scored customers on how strongly they responded to promotions, and that three internal crisis-prediction efforts stalled.
The Product Desk · Product desk
What happened
- The New York Times reports that DraftKings built a model in 2023 to score customers on elasticity, a measure of how much their behaviour responds to free bets and promotions.
- A crisis-prediction model that former employee Jake Shanin said was showing promise lost its early 2025 presentation when the meeting was canceled the day it was due.
- DraftKings told the Times that promotions go to customers who demonstrate sustained, engaged use of its platform and not to customers based on their losses.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- contradiction DraftKings' defence and the Times' account can describe the same score: sustained, engaged use is roughly what an elasticity model measures, and the reported slots figures put the heaviest responders and the heaviest losers in one group.
- constraint Kalani's position that risk modeling technology was not helpful is hard to test when the models that might test it did not reach a presentation, so the internal evidence base stays as it is.
- decision Any team scoring customers by expected contribution now has to decide whether the model that says send this person less gets the same headcount and executive time as the one that says send more.
- exposure With 40-plus former employees and internal memos in a published account, the contents of a targeting score become readable by the customers it ranks and by anyone licensing the operator.
Redditors have described DraftKings' emails and push notifications as difficult to turn off [1]. Elasticity, the property the New York Times says the 2023 model scored, is an economic measure of responsiveness to incentives, and on its face it does not relate directly to traits like problem gambling [3][4]. A customer who bets more when the free bets land and less when they stop is elastic. By the Times' account, the more elastic users generated more revenue for the company [5].
Jayden Butts, a former employee who spoke to the Times on the record, said his basic task was applying a model that evaluated the worth of users on one question: "Is this person going to give us more than we're giving them?" [6] When the answer was yes, Butts said, that customer became a target for promotions and the company would "open the floodgates" [7].
DraftKings told the Times it "rejects any implication that its marketing practices are unfair or improperly targets customers" [8]. Promotions, the company said, are "directed toward customers who demonstrate sustained, engaged use of our platform, not toward customers based on their losses" [9]. For that distinction to hold, engaged use and losing have to come apart in the data. A 2023 memo the Times says it reviewed found slots players more elastic, and early 2024 data showed highly elastic slots players losing more money than less elastic ones [10][11].
A 2016 U.K. study of gamblers found problem gambling among online slots players was the second most prevalent of the player types it studied, behind in-person poker players [12]. Slots is the game type the 2023 memo scored as most elastic.
Other former employees worked on a model pointed the other way, one built to predict when a user was heading toward a crisis and would need an intervention [13]. Jake Shanin told the Times the internal crisis prediction model was showing promise [13]. In early 2025 his team prepared to share it with company officials including Lori Kalani, DraftKings' chief responsible gaming officer, and the meeting was canceled the day of the presentation [14]. Two other employee attempts at similar algorithms were also shelved, according to two former employees [15]. Counting Shanin's, three internal attempts at crisis prediction appear in this account, against one scoring model that was being applied to customers [1].
Kalani told the Times that DraftKings needs "customers who are betting within their means, are betting for entertainment and betting for fun" in order to stay in business, and that it monitors for "potentially risky behaviors" [16][17]. She also said evidence had shown that risk modeling technology wasn't helpful [18]. The elasticity model dates to 2023 and the crisis model lost its meeting in early 2025, about two years apart [2].
No regulator, court or enforcement action appears in this account, which rests on former employees and documents the Times says it reviewed [3][2]. For a team running value-based targeting, two checks separate this from ordinary personalization. The first is whether the feature list of your ranking model reads the same to the customer being ranked as it does in the internal document. The second is whether the suppression side, the model that says send this person less, has a named owner, a budget and a slot on the same executive calendar as the targeting side.
What to watch
- Whether DraftKings publishes the feature set behind its promotion targeting, or a state gaming regulator asks for it.
- Whether any of the three shelved crisis-prediction models is revived and actually presented to company officials.
- Whether the evidence Kalani cites for risk modeling technology being unhelpful is ever published or described.