Apr 25, 2025 | 3:01 am
The gambling sector is looking at new ways to assess artificial intelligence and machine learning systems designed to identify customers who may be at risk of gambling harm. Research supported by Playtech Protect examines how shared benchmarking methods could help casino operators, technology providers, and regulators better evaluate these tools.
AI-based player risk systems are becoming increasingly common across gambling operations. These systems analyse customer behaviour, including betting activity, playing patterns, session duration, and product choices, to identify possible warning signs. Operators use these insights to decide when customers may benefit from additional support or intervention.
The challenge is that different gambling companies and technology providers often build and test their systems in different ways. They may use different datasets, definitions of harm, modelling approaches, and performance measurements. As a result, claims about accuracy can be difficult to compare.
The research highlights the need for clearer ways to understand whether a system identifies the right customers at the right time while limiting incorrect alerts and missed cases.
Playtech Highlights AI Testing Needs
Playtech has been exploring the issue of benchmarking player risk algorithms for several years. The company helped bring attention to the topic through a presentation at the International Association for Gambling Regulators in 2024 and supported the academic work that developed the benchmarking concept.
The research does not introduce a new risk detection model. Instead, it reviews current approaches to AI and machine learning in player protection and outlines a possible framework for comparing different systems.
The proposed benchmarking model would use anonymised player information, with data separated into training and testing groups. Developers would receive access to the training data, including selected indicators linked to gambling harm, and would use that information to prepare their algorithms.
The testing data would keep the relevant harm indicators hidden. Providers would submit their predictions for players in that dataset, allowing an independent benchmarking process to compare results against the withheld information.
The researchers suggest using several benchmarks rather than a single ranking system. Different evaluations could examine gambling-related risks appearing during individual sessions, across several days, or over longer periods.
The framework could also include different types of gambling customers, covering different products, regions, and levels of engagement. This approach would allow systems to be assessed across a wider range of situations instead of relying on one narrow dataset.
Data Privacy and AI Confidence
The research supported by Playtech Protect also addresses concerns around privacy, security, and commercial information. Any benchmarking system would need to provide meaningful assessments while preventing unnecessary disclosure of sensitive customer or business data.
The paper discusses several possible methods for achieving this balance, including independent data trusts, regulator-managed repositories, synthetic datasets, hidden testing environments, and systems that allow models to be evaluated without requiring companies to reveal their internal details.
The researchers suggest that regulators, independent research groups, data trusts, or other neutral organisations could oversee such benchmarking efforts.
For gambling operators, clearer testing standards could help them make better decisions when choosing risk detection technology. Providers could also use independent evaluations to show how their systems perform under consistent conditions.
The absence of shared benchmarks currently makes it harder for the industry to separate strong-performing solutions from systems that rely mainly on marketing claims. Playtech Protect’s involvement in the discussion reflects a wider industry focus on creating more transparent ways to assess AI-based player protection tools.
Top United Kingdom casinos
£
40
T&C's Apply
+ 100 Free Spins
New Customer Offer. T&C’s Apply. 18+. #ad. Min Deposit £20, excl. PayPal & Paysafe. Max bonus £40 on selected games & 100 Free Spins on Big Bass Splash. Wagering and max win apply. T&Cs Apply.
23 Free Spins
New customer offer, T&Cs Apply, 18+. #Ad. 18+, 10x wagering on the 23 Free Spins. 77 Cash Spins after 1st time deposit and £10 wager on slots. No max cash out on Cash Spins. Skrill Deposits Excluded. |T&C's Apply! Gambling can be addictive. Play Responsibly. 18+.
Regulators and operators may face greater expectations
The research indicates that companies using or supplying AI risk systems will likely face greater expectations around evidence and accountability. Operators and providers may need to show that their models perform effectively beyond the data used during development and that interventions connected to these systems help reduce gambling-related harm.
The paper suggests several practical steps that could improve model oversight. These include keeping clear records about how systems work, defining risk categories consistently, carrying out testing with new data, checking for fairness, watching for changes in player behaviour patterns, and reviewing whether algorithm-based interventions achieve their intended purpose.
The researchers also note that if the industry does not create credible and transparent evaluation methods, regulators may introduce their own requirements. Spain is highlighted as one jurisdiction moving toward more detailed oversight, alongside wider European efforts to create more specific rules around technology and models.
The academic paper behind the research involved eight academics specialising in player risk AI algorithms from institutions including the University of Nevada (Las Vegas), the Division on Addiction at Harvard Medical School, the University of Calgary in Canada, the University of Sydney in Australia, Washington State University, and Focal Research Consultants.
Playtech Protect has contributed to ongoing industry conversations about how gambling AI systems should be assessed, while the research provides a framework for considering how future benchmarking could operate in practice.
Source:
Benchmarking player risk algorithms, Industry Research Brief Vol. 6.(1) – Benchmarking player risk algorithms, playtech.com, July 2026.