Can operators trust AI to predict player behaviour if the data foundations are still messy?

Fragmented player data, unclear definitions of churn and LTV, predictions that never make it into daily decisions, and technology that looks impressive without actually changing what teams do.

So where should operators really start, and when does a prediction model become useful enough to act on?

We spoke with Aditya Satyadev, Co-founder & Director at BizAcuity, about 20 years of change in iGaming data, the gap between prediction and execution, GAMWIT, and why the industry may still be talking about its data foundations in 2028.

Q1.  You led PartyGaming’s data warehouse and BI work through the IPO, stayed through the bwin. party merger, then co-founded BizAcuity in 2011 with Sudhir Pidugu and Nishant Jain. What problem were you still seeing that made starting a company the right move?

A. By the time we'd been through the IPO and a few M&A processes, I had watched first-hand what a proper data platform did for the business. BI wasn't a back-office function, it was shaping how commercial, marketing and VIP teams made decisions every day, and the value was undeniable. 

But the systems delivering it were top-of-the-shelf enterprise applications, with the licensing and infrastructure costs to match, so that capability was effectively reserved for companies of PartyGaming's size. 

Around the same time, open source and the wider tech advances were changing the math. You could now stand up something genuinely powerful for a fraction of what those enterprise stacks cost. That was the opening: if the value was real and the cost of delivering it had collapsed, the same benefit could reach operators who'd never been able to justify a tier-1 build. 

We started BizAcuity in 2011 to do exactly that: help other iGaming companies, and businesses well beyond gaming, create real value out of the data they already had. The idea was simple “solve business problems with data by building solutions using the right technology at the right price”.

Q2.  You have seen iGaming data for about 20 years, from spreadsheets and overnight batches to dashboards and then live prediction. What changed in a way that moved the numbers, and what looked useful but did not?

A. The honest truth is that the basic transactional data in an iGaming system hasn't changed much in 20 years - a bet, a deposit, a withdrawal look about the same as they always did. 

What changed the game is the data around those transactions. It now arrives in real time, and there are countless moments where a point-in-time action decides whether a player places that next bet or walks away. We've also moved from simple transaction recording to rich behavioural telemetry, which has exploded the volume which sometimes becomes unmanageable unless the architecture is built for it, or the data lake quietly becomes a swamp. 

Add external data like regulatory feeds, region-specific behaviour, payments and fraud, abuse and compliance, moving from a side concern to mission-critical, and player behaviour became easier to understand yet far harder to model. 

The piece that made all of it usable was stitching it into a single view of the player across channels and devices. What looked useful but didn't move numbers were the dashboards built for their own sake and the "collect everything" projects producing beautiful charts nobody acted on. If a number didn't change a decision someone made that day, it never moved the needle.

Q3.  What surprised you most about that shift: the technology, how ready operators were to use it, or how long it took for prediction to become part of the daily workflow instead of a quarterly presentation?

A. Easily the last one. How long it took for prediction to move from a quarterly slide into the daily workflow. 

The technology matured far faster than the habits around it. Operators were often readier to buy a model than to change how a CRM team or a VIP desk starts its morning, and that is where prediction lived or died. 

A forecast only matters if someone trusts it enough to act before the day begins, and that trust is an organisational change, not a technical one. We had churn and LTV models that were genuinely useful years before most teams wired them into a daily rhythm. 

The bottleneck was never the algorithm. That is the lesson that shaped how we build now at BizAcuity and with GAMWIT. We make the prediction ranked, explainable and dropped straight into the tools people already use, so acting on it isn't a special project. It becomes part of the day.

Q4.  A lot of operators still run the business off reports that look backwards. When does a prediction model deserve a place next to the daily dashboard, and where do you still see teams buying AI before the underlying data is in good shape?

A. A model earns its place next to the daily dashboard when it changes a decision that gets made every single day, and when you can measure the lift it creates. A churn score that reroutes retention spend, an early-VIP flag a host acts on that afternoon, a fraud signal that stops a payout before it leaves. If it just sits on the screen looking clever and nobody does anything differently, it isn't ready, however good the model is. 

Where I still see it go wrong is teams buying AI before the foundation is in shape. Fragmented player identities, no agreed definition of what churn, LTV or even a VIP means across departments, transactional data that doesn't reconcile. AI is not a magic wand. It works on the data you feed it, and it is still junk in, junk out. Put AI on a shaky foundation and it just automates the wrong numbers faster and with more confidence. 

With GAMWIT the data migration itself is substantial, but the integration, the exploratory data analysis, the QA and coming to agreement on certain definitions are what take priority, because that is what makes the models trustworthy. The discipline comes before the model. The model is the easy part.

Q5.  Two things are really in play here. One is prediction, the churn, LTV and VIP models that tell you who a player is, what they are worth and where they are heading. The other is the CRM layer, the bonus design, personalisation and game recommendation that decide what you actually do about it. Should operators treat these as one system, and if so, which comes first, or how do the two merge?

A. They are two halves of the same system, and the mistake is running them as two disconnected projects. Prediction must come first, because it is the read of the player that everything else depends on. But prediction on its own changes nothing. 

It only creates value when a CRM acts on it, through the bonus, the personalised journey, the next game to surface. Here I want to be clear about boundaries. 

GAMWIT is not a CRM and does not try to be one. It is the prediction layer, the brain that sits behind the CRM. 

Most tools in the market are built CRM first, with prediction bolted on later, almost as a feature. GAMWIT is the opposite. It is prediction first, and those predictions, churn, LTV, VIP, along with responsible gaming and AML, become the input into whatever CRM the operator already runs. 

Because that brain holds the view of each player's value and risk, it also lets you track the ROI of your CRM efforts, whether a bonus or a campaign moved the predicted value and retention, rather than just guessing. That order matters. When prediction leads, the CRM acts on a signal it can trust, and you can measure the result, the two stop working against each other, and grow and protect and finally pull in the same direction.

Q6.  SBC Summit is days away. What are operators asking this year that they were not asking two years ago? And is there anything BizAcuity or GAMWIT is putting in front of them in Lisbon because of those questions?

A. Two years ago, the conversation at a show like this was still fairly basic. Can you give us dashboards, can you predict churn. This year the questions are sharper and more demanding. 

Operators want prediction that stays independent of any single CRM or execution stack, so they are not locked into one vendor just to act on their own data. They are asking about responsible gaming and AML as real models, with explainability and audit trails, not as compliance checkboxes. 

And almost everyone now follows up with "prove it, show me the ROI," because budgets are tighter and the market is full of AI claims. There is also a lot of noise about AI agents and autonomous action, but the real question underneath is whether the prediction the agent acts on is any good in the first place. That is what we are talking to operators about in Lisbon. 

GAMWIT as an independent, prediction-first layer that feeds whatever CRM they already run, with the risk and compliance models built in. And we now have a real proof point. For one of our clients we benchmarked GAMWIT's predictions head-to-head against a well-known engagement and bonus-optimisation specialist, and our numbers came out materially better, at a noticeably lower price. That is the kind of evidence that answers "prove it."

Q7.  Given that history, what do you think the industry will still be debating about in 2028 that it should already have settled: the data foundation, safer gambling models, acting in the moment, or something else?

A. The data foundation, and it genuinely frustrates me.

It is the least glamorous and the most decisive thing on that list. In 2028 we will still be arguing about consistent player identity, agreed definitions of churn, LTV and VIP, and clean event-level data, when all of that should have been settled a decade ago. 

Everything the industry gets excited about, safer gambling, acting in the moment, personalisation, sits downstream of that plumbing. Yet we keep buying the exciting layer and skipping the boring one and then wonder why the models underperform. 

It comes back to what I said earlier. AI is not a magic wand, and junk in still means junk out. Safer gambling models will stay in the debate too, and that one is legitimately hard, because acting in the moment forces a real trade-off between commercial pressure and duty of care. But the foundation is not hard. 

It is just unglamorous, and that is exactly why it keeps getting deferred. If the industry fixed the foundation first, most of the other debates would get a lot quieter.