The Dutch gambling regulator has made public an open-source machine-learning model designed to flag risky online gambling by studying what players do, not what they say about themselves. The system scores each player by analysing betting behaviour, gambling frequency, session timing and responses to winning and losing streaks.
The University of Amsterdam said the model was built for Kansspelautoriteit, the Ksa, and was put on the regulator’s website on Tuesday, 18 August. As covered in July, the Ksa had already pushed back against tying betting limits to income, arguing instead for more targeted protections for vulnerable players.
The new tool is meant to sit alongside, not replace, that approach. The regulator does not treat a risk score as proof that a player has a gambling problem, and it does not see the algorithm as a substitute for human oversight or the broader duty of care required of operators.
According to the university, the project was led by Charles de Leau and developed with professors Reinout Wiers and Johan Bollen. ZonMw funded the work from the Ksa’s Addiction Prevention Fund after de Leau pitched the research idea. The model was trained on all bets made by all players at 13 Dutch online casinos over two years, from 30 July 2023 to 30 July 2025.
The training data came from a legal provision requiring casinos to make user data available for independent research. De Leau said that analysing all bets from 13 casinos over two years had never been done before by independent researchers, and that the size of the dataset allowed patterns to be seen on a scale normally used by casinos for marketing.
The open-source design is central to the project’s pitch. The press release says the algorithm is open source, with the code and methodology public so other researchers can verify and build upon it. The Ksa also says the tool gives it another way to calculate and compare risk scores, rather than relying on closed systems owned by gambling companies.
The regulator’s report on “Markers of Risk” helps explain the backdrop. Online gambling was regulated in the Netherlands in 2021, and licensed operators must keep player data in a digital vault known as the remote gambling data safe. That vault contains deposits, withdrawals, stakes, winnings, game type, changes in player limits and interventions, along with pseudonymised player codes and age and status information.
That report said the goal was to use those data to prevent addiction and improve supervision, with the main task being to compare operators on risky behaviour, evaluate interventions and judge whether they work. It focused on operator-level patterns, not individual player diagnosis, and argued that looking only at extreme losses misses many at-risk players.
Its illustrative analysis covered 25 operators and 2.6 million unique player accounts over the year from October 2023 to September 2024. The median monthly loss per account was 35 euros and the average was 137 euros. About 68 percent of accounts lost 100 euros or less a month, while 6.4 percent lost more than 700 euros and 1 percent lost at least 2,500 euros.
Those top 1 percent accounted for 43 percent of gross gaming revenue. By contrast, players losing between 0 and 100 euros a month made up 48 percent of players but only 11 percent of revenue. Among young adults, 1 percent of players generated 33 percent of gross gaming revenue.
The Ksa says net deposits above 700 euros, or 300 euros for young adults, are in principle a signal of risky behaviour. In those cases, it regards a temporary deposit block as appropriate unless the player can show that their financial situation allows more.
The report also said higher-risk gamblers are often marked by more intense play, more frequent and larger deposits, more reversed withdrawals and more late-night gambling. It said at-risk players often combine multiple gambling products and engage in faster, more continuous games.
For non-incidental accounts, the median number of deposits per day was 2 and the average was 3.6. About 32 percent made no more than one deposit a day in a single month, while 7 percent made ten or more deposits on a single day.
The report defined incidental players as those who gamble fewer than five days in a month and lose less than 300 euros. It said they accounted for 49 percent of all players, but that self-exclusion is not a good stand-alone measure of risk because many at-risk players do not self-exclude, and some who do are not at risk.
The article says the Dutch project fits a wider European shift toward using behavioural data to identify harmful gambling earlier. It pointed to France, where the gambling regulator ANJ used an algorithm to identify roughly 600,000 account-based players considered likely to be gambling excessively, a group said to account for about 60 percent of the segment’s gross gaming revenue.
