Safer gambling needs smarter checks, not simply more checks

How do we know whether safer gambling regulation is actually working?

We can count checks. We can count customer interactions. We can measure how many people set limits or self-exclude.

But those numbers do not tell us if operators are getting better at identifying the people who genuinely need help.

That matters as the UK introduces another layer of financial checks.

In July 2026, the Gambling Commission “confirmed that Financial Risk Assessments will be introduced in stages” 

The aim is sensible: identify high-spending customers experiencing financial difficulties without asking most of them for payslips, bank statements or other documents.

But perhaps we are still asking the wrong question.

Instead of constantly debating when a player should be checked, we should get much better at identifying who actually needs intervention.

Safer gambling needs smarter checks, not simply more of them.

Spending alone does not tell us enough

The same amount of money means completely different things to different people.

A £500 gambling loss could cause serious financial problems for one customer. For another, it may be affordable entertainment spending.

Financial information still matters. But operators have something else: behavioural data.

Imagine two customers.

One has deposited £1,000 every month for several years. Their activity is stable and there is no obvious change in behaviour.

Another normally deposits £100 a month. Suddenly they start depositing several times in an evening. Sessions get longer, stakes increase and they repeatedly chase losses.

£1,000 a month, stable, no change in pattern

Long-term steady depositor

Behaviour is consistent over several years with no obvious change.

£100 a month, then sudden spike

Behaviour-change depositor

Multiple deposits in one evening, longer sessions, rising stakes and loss-chasing.

Author's illustrative example, not measured data

Who deserves closer attention?

A simple spending threshold cannot answer that properly.

The Gambling Commission already recognises this. Its “remote customer-interaction rules” require operators to consider several indicators, including spend, changes in spending patterns, time spent gambling, gambling behaviour and use of gambling-management tools.

The challenge is bringing those signals together.

Gambling could learn from risk-based monitoring

I have worked in clinical research and clinical trials, where risk-based monitoring is already a familiar concept.

The principle is straightforward.

You do not treat every data point, research site or potential issue as equally important. You focus attention where the risk to participants or the reliability of the trial is greatest.

Regulators including the US Food and Drug Administration “support risk-based approaches to clinical-trial monitoring“.

I think safer gambling should move further in the same direction.

It does not mean monitoring less. It means concentrating attention where the evidence suggests the risk is highest.

Otherwise, we risk creating a compliance exercise: more checks, more alerts and more documentation without knowing whether the right players are actually being protected.

This is where AI could be useful

AI is mentioned so often in gambling that the term itself is starting to lose meaning.

But player protection is one area where it has an obvious use.

Human safer-gambling teams cannot continuously study the behaviour of hundreds of thousands of customers. Software can look for changes and combinations of signals at scale.

Mindway AI, which is majority-owned by Better Collective, is one example.

Its GameScanner technology uses AI alongside assessments from human experts to identify patterns associated with risky gambling and assign players different risk profiles.

The interesting part is not Mindway itself. It is the approach.

Instead of asking only whether somebody has crossed a financial threshold, technology can ask whether their overall behaviour has changed in a way that indicates risk.

That could allow operators to intervene earlier when several warning signs appear together.

It could also leave customers showing no meaningful signs of harm largely alone.

That is what a genuinely risk-based system should try to achieve.

Are we counting the wrong thing?

There is a good reason to question the current approach.

The Gambling Commission says major online operators carried out around 20.5 million customer interactions in 2025–26, up from 13.3 million the previous year.

Twenty million interactions sounds impressive.

But the Commission “also says it does not currently collect data that allows it to measure the quality or impact of those actions” (https://www.gamblingcommission.gov.uk/about-us/impact-metric/ro1-protecting-from-harm-or-exploitation/operator-customer-interactions).

That is the problem in one sentence.

Is 20 million better than 10 million?

We do not really know.

If operators are identifying vulnerable customers earlier and preventing harm, then more interactions could show progress.

If millions of low-risk customers are receiving automated messages they immediately dismiss, the number tells us very little.

We have become good at measuring whether the process happened.

We need to become better at measuring whether it worked.

AI cannot become the decision-maker

A risk-based approach also creates its own problems.

Algorithms make mistakes.

A football bettor may suddenly gamble more during the World Cup. A racing customer may spend considerably more during Cheltenham.

Unusual behaviour is not automatically harmful behaviour.

There are also questions about privacy, transparency and incorrect risk scores.

If AI contributes to a decision that restricts someone’s account, an operator should be able to explain why that customer was flagged.

AI should help safer-gambling teams identify risk. It should not become an unexplained machine deciding who is allowed to gamble.

Human judgement still matters.

The goal should be better interventions, not more interventions

This is where the debate should move next.

We should stop asking only:

Did the operator intervene?

We should also ask:

Did the intervention work?

Did the operator identify a problem earlier? Did the player’s behaviour become less risky? Did an intervention prevent further harm? How many customers were incorrectly flagged?

These questions are harder to measure than the number of emails, pop-ups or checks completed.

But they are much closer to what safer gambling is supposed to achieve.

Better data, behavioural monitoring, financial information and AI could allow operators to become more targeted.

Customers showing genuine warning signs could receive earlier and stronger intervention.

Customers showing no meaningful signs of harm could experience less unnecessary disruption.

That is the point of a risk-based approach.

We should not judge safer gambling by how much monitoring takes place.

We should judge it by whether we are getting better at finding the right players, at the right time, and doing something that actually helps.