Operators and suppliers increasingly use algorithms to identify players at risk of gambling harm. But the industry has no common, objective way to test whether these tools work. Our benchmarking framework, published in the Journal of Gambling Studies (2026), sets out how to evaluate player risk detection models on a like-for-like basis.

Why it matters

Benchmarking gives defensible, third-party assurance that harm detection works. It shows where the gaps are, gives regulators something stronger than self-assessment, and offers a rigorous way to compare capabilities across brands, markets and product verticals.

How it works

The approach uses a train-on-synthetic, test-on-real design. Detection models can be compared without pooling sensitive player data, and evaluation can extend across different markets and product verticals.

Current status

We are running a proof of concept using a dataset from our co-authors at Harvard, ahead of a case study applying the benchmark with operators.