← Back to feed

DraftKings built an internal machine learning model that scores bettors by expected “elasticity” — their predicted net loss in response to promotions — and used it to target high-scoring users with personalised offers. The Electronic Frontier Foundation published an analysis of the practice on September 24, drawing on a New York Times investigation published September 19 into DraftKings’ use of behavioral data.

The pattern: a user places bets, loses, and stops playing. DraftKings’ elasticity model flags users who are statistically likely to generate the most net loss when re-engaged with promotions. The platform delivers personalised bonuses and “free bet” offers to those users — offers that the model predicts will produce further losing wagers.

What Makes This Different From Standard Ad Targeting

Standard behavioral advertising re-targets users who showed interest in a product but did not convert. DraftKings’ elasticity model specifically selects users who converted, lost, and churned — the profile most associated with problem gambling. The model’s training signal is loss behavior and churn, not general engagement.

While a separate DraftKings project aimed at explicitly identifying problem gamblers was reportedly shelved, the elasticity scoring model achieves a similar practical result: routing the highest-value promotional spend toward users showing the behavioural signatures of gambling disorder. The EFF frames this as online behavioral advertising “supercharged” by machine learning.

Regulatory Gap

No current US federal law specifically prohibits training a model on customer gambling loss data and using it for ad targeting. Most state-level sports betting regulations focus on responsible gambling disclosure requirements — deposit limits, self-exclusion registries, cooling-off periods. Massachusetts is an exception: regulation 205 CMR 257, adopted in 2023, specifically prohibits operators from using AI or automated systems known or reasonably expected to make gambling more addictive, and the Massachusetts Gaming Commission has opened an inquiry into DraftKings under that provision.

The FTC has enforcement authority over deceptive and unfair trade practices, but has not moved against AI-driven behavioral targeting in gambling contexts. The CFPB’s jurisdiction does not extend to non-financial gaming platforms.

Pattern Across Industries

DraftKings is not alone in applying ML to precision-target high-loss customers. The practice mirrors what credit card companies discovered decades ago with paper mail: people in financial distress respond to credit offers at disproportionate rates. AI collapses the cost of identifying those individuals and delivering personalized messages at scale.

The difference in the gambling context is that the product being sold is itself the mechanism of harm. A credit card company is exploiting financial distress. DraftKings is exploiting gambling disorder to sell more gambling.

The EFF’s critique centers on this distinction — that the harm here is not incidental to the business model but structurally embedded in it, and that AI makes it dramatically more efficient.