DraftKings used AI to target the gamblers likeliest to lose

A New York Times investigation says DraftKings built a machine-learning “elasticity” score to predict which customers would lose more after promotional bonuses — then steered incentives toward high scorers — while efforts to use similar tech to spot problem gambling were stalled.

FINANCE desk — Using AI to aim bonuses at the bettors predicted to lose more after a promo is a consumer-protection story about model incentives: the score was trained to find extra losses. DraftKings disputes that it targeted people because of those losses.

The company, as the Times reports it, named the number an “elasticity” score. In this setting, elasticity means how much extra betting — and extra customer losses — a free bet or bonus was expected to produce. A higher score meant a promotional dollar spent on that person was predicted to come back as more play, and more losses, for DraftKings. File that definition as the Times’. This desk did not score an account.

The Times says DraftKings built the casino model in 2023. It looked at how often someone played, how their account balance moved from day to day, how much they usually wagered, how much they lost compared with that wagering, and how likely they were to stop. Loss-to-wager ratio is a simple fraction: losses divided by total bets. File those inputs as the Times’. This desk did not rebuild the model.

The score was then used to steer free bets, bonuses, and other incentives toward high scorers — the people the model said would respond by gambling more, and losing more, after the offer. File that targeting as the Times’. DraftKings disputes the “because they lose” reading; that dispute is below.

What is not in dispute is that DraftKings has told investors AI now aims a large share of its promo budget. At its 2026 Investor Day the company said it automated and personalized about $400 million in promotional spending with AI in 2025, and that data science helped lift the margin on promotion-driven sportsbook bets by 13% that year. File those Investor Day figures as DraftKings’, via the Times. This desk did not sit in the investor room.

Jayden Butts, a former DraftKings data analyst who the Times says tested the elasticity system on thousands of casino players, told the paper: “We are looking for traits and features that we can target that indicate a good investment.” He added: “The best investment would be a problem gambler.” A second unnamed former analyst told the Times, “It is as predatory as it sounds.” File those lines as theirs, via the Times — not as this desk’s finding, and not as a company admission.

The Times also reports a second track that did not ship. In mid-2024, data scientist Nestor Hernandez began a separate model meant to give the responsible-gaming team a risk score before a customer hit a crisis. Responsible gaming here means the company’s tools and outreach for people whose betting may be becoming harmful. Hernandez left in November 2024 with the model unfinished. By early 2025, the Times says, a planned briefing was canceled and the project was shut down. File that sequence as the Times’.

Lori Kalani, DraftKings’ chief responsible gaming officer, told the Times that leaders made a “collective decision” not to use predictive technology for problem gambling because they did not see the approach as “evidence-based.” File that as Kalani’s, via the Times. It is the company’s stated reason — not a finding that a harm model could not work.

DraftKings told the Times it “rejects any implication that its marketing practices are unfair,” and that promotions go to customers who show “sustained, engaged use of our platform, not toward customers based on their losses.” It called Butts’s analysis “preliminary and inconclusive.” The company also says its business depends on people betting within their means, and that it already watches more than two dozen behavioral indicators, with deposit limits, time limits, cooling-off periods, self-exclusion, messages, and, in some cases, account closure. File the dispute and those tools as DraftKings’, via the Times.

Scale, still the Times’ reporting, not a desk audit: Citizens Bank research cited by the paper put DraftKings’ 2025 gross gambling revenue near $8.7 billion and promotions near $3 billion. The Times also cites the company saying it has about 11 million customers, up from five million in 2022. Gross gambling revenue is the money kept after winning bets are paid — not profit after all costs. File those figures as cited. This desk did not recount the books.

Plain English for the rest of the card: sportsbook = an app or site that takes bets on sports. online casino = the same company’s slots and table games. elasticity score = DraftKings’ internal number for how much extra play, and extra losses, a promo was expected to produce. promo / free bet / bonus = an incentive meant to get someone to keep betting. loss-to-wager ratio = losses divided by total bets. problem gambling = betting that is becoming harmful to the person doing it. responsible gaming = the company’s harm tools and outreach. predictive risk tool = software that tries to flag that harm before a crisis. This filing is the Times investigation plus the company dispute — not a court finding, and not a new law.

REPORTED here: The New York Times’ 19 Sep 2026 investigation — Tier B originating newsroom, not a DraftKings newsroom PRIMARY. TechTimes is a same-day pickup of that Times story — not a substitute primary. The elasticity score, the 2023 casino model, the frequency / balance / loss-ratio inputs, the steering of bonuses toward high scorers, the Butts and unnamed-analyst lines, the Hernandez harm-model shutdown, Kalani’s “evidence-based” reason, the Investor Day $400 million / 13% figures, the Citizens Bank $8.7 billion / $3 billion scale, and the 11 million-customer line are Times-attributed. The “sustained, engaged use” dispute, the “preliminary and inconclusive” line, and the existing responsible-gaming tools are DraftKings’, via the Times. NOT claimed: that this desk saw the internal memos, that DraftKings admitted it targeted problem gamblers, a court finding, a new statute, that the harm model would have worked, a stock tip, or investment advice. Distinct from the already-filed anthropic-claude-financial-advisors, openai-chatgpt-financial-services, synapse-analytics-13m-series-a, and bis-ai-market-vulnerability.