Spain’s DGOJ is building an AI model to flag problem gambling, and the sector has questions

Spain’s Dirección General de Ordenación del Juego (DGOJ) is preparing an artificial intelligence system to detect players with risky behaviour on online betting platforms. The regulator says the model, based on machine learning and XGBoost, can identify 48 types of vulnerabilities — which makes it central to its sector reform, but also a target for criticism from safer gambling specialists.

1. The DGOJ’s system is designed to analyse player behaviour and classify users with risk indicators on online betting platforms. It uses machine learning, specifically XGBoost, and is said to be able to identify 48 types of vulnerabilities in user behaviour.
2. Pedro Romero, head of safer gambling partnerships at BetBlocker, said he was surprised that such a complex project moved ahead without a tender process to choose external partners or advisers. He said independent firms with experience could have helped, especially on two hard parts of the job: identifying at-risk users and deciding how to intervene. He also warned that the regulator may be adopting models used in other jurisdictions without considering all the factors needed to make them work, and called for a dialogue with the sector.
3. Paula Murphy of Mindway AI said there were some meetings with the DGOJ, but they did not help shape the algorithm. The only public consultation on the system opened on 13 October and responses were due by 22 October, giving the sector less than ten days to comment on a complex model.
4. Murphy also questioned the way the model appears to work. Based on her understanding of the proposal, the system would review customers periodically, perhaps once a month, and classify them to determine whether additional controls should apply. In practice, that means a green-light or red-light decision for each player.
5. Her view is that this is too blunt for a problem that exists on a spectrum. Operators, she said, need to know why a player was flagged and how behaviour changes over time — including whether activity is increasing, where the player still has control, and whether a specific product is making behaviour worse. She also said Mindway AI has incorporated the DGOJ model into its GameScanner platform, and that its data scientists and engineers had questions about the model’s scope and the data used to train it. Murphy called for more transparency on what type of players were used in training, saying the algorithm should not be something operators accept on faith.

The practical point for PSPs and gambling operators is straightforward: if the DGOJ’s model becomes part of the compliance baseline in Spain, the questions around training data, intervention logic, and explainability will matter just as much as the detection layer itself.