Dutch Ksa Tests AI Model to Detect Risky Gambling

A machine-learning model developed for the Dutch gambling regulator is designed to identify potential signs of risky online gambling by analysing players’ actual gambling behaviour rather than relying on what they disclose about themselves.

Researchers at the University of Amsterdam created the open-source algorithm for the Kansspelautoriteit (Ksa). The system analyses factors including betting patterns, gambling frequency and the timing of gambling sessions. It also considers winning and losing streaks and how players respond to those outcomes before generating an individual risk score.

The project applies to online gambling more broadly and is intended to provide operators with an additional tool for their existing player-protection systems.

Focus on Gambling Behaviour

The model relies on activity data because potentially risky gambling behaviour can develop through changes that players may not recognise or report.

Researchers also drew on previous work conducted by Spain’s gambling regulator, the DGOJ, while developing the model. Its open-source format allows operators to use the framework within their own monitoring systems.

However, the Ksa stresses that the model has clear limitations.

A risk score does not mean that a player has a gambling problem. The algorithm is also not intended to replace operators’ existing duty-of-care responsibilities. Instead, it is designed to highlight behavioural patterns that could require further attention.

Operators remain responsible for determining how to respond when the system identifies potential risk and for applying the safeguards required under their player-protection obligations.

Growing Use of Behavioural Data

The Dutch project forms part of a wider European effort to explore whether gambling behaviour can be used to identify potentially harmful activity at an earlier stage.

France has adopted a comparable approach. Its gambling regulator, the ANJ, has used an algorithm to identify around 600,000 account-based players considered likely to be gambling excessively. These players represented approximately 60% of the segment’s gross gaming revenue.

Both approaches reflect a shift toward more targeted responsible gambling measures. Rather than relying solely on general warnings, spending limits or information provided by players, regulators are examining whether gambling data can reveal patterns that may indicate risk.

The Dutch model is not being presented as a definitive solution. Its effectiveness will depend on how operators interpret the scores and respond when players are flagged. The Ksa maintains that the algorithm should serve as an additional tool, rather than replacing human judgement or existing duty-of-care requirements.