The two faces of the safer-player algorithm. By Mark McGuinness
Artificial intelligence (A) has become the iGaming industry’s most powerful instrument for safeguarding players. It is also the most powerful instrument for monetizing them. The question facing every corporate board is which purpose comes first.
The industry has spent the past 18 months positioning AI as the great savior of responsible gambling. Every earnings call, every compliance briefing, every trade-press interview now features a reference to predictive markers, early intervention and machine-learning models that identify harm before it becomes a headline. The vocabulary is reassuring. The reality, as ever, is it feels much more complicated.
Behind the glossy earnings presentations sits an uncomfortable truth. In many operators, the same neural networks that identify a player in distress are drawn from the same data lakes, built by the same engineers. They are judged against many of the same key performance indicators (KPIs) as the systems that personalize marketing and maximize lifetime value. This is not an oversight. It is an architectural and operational decision. That decision deserves scrutiny and it is a concern of regulators: black box AI.
The Dual-Use Problem
Every behavioral signal an operator captures is dual-purpose in nature. Session length, deposit velocity, loss-chasing patterns, time-of-day volatility: These are the fingerprints of potential player harm. They are also, almost without exception, the variables that predict a player’s commercial value. A customer who deposits at 3 a.m. after four consecutive losses is, statistically, a high-risk individual. In the same dataset, they are often the most profitable, a moral and commercial dichotomy.and your customer relationship management (CRM) model share inputs, share teams or share accountability, the commercial incentive will, over time, contaminate the protective one. This is not a moral accusation. It is a structural inevitability. Optimization pressure flows downhill and it flows toward revenue. You can’t be judge and jury at the same time.
The UKGC has been unusually direct on this point. Its assurance work through 2024 and 2025 repeatedly emphasized that automated customer-interaction systems must be designed around harm indicators, not retention indicators. The distinction is not cosmetic. It is the difference between a license retained and a license reviewed.
What “Protection” Actually Looks Like
A genuinely protective AI deployment has characteristics that are technically verifiable and commercially uncomfortable. The first is independence of objective function. The model must be trained to minimize harm, not to maximize retention, reactivation or churn prevention. If a data scientist cannot tell you, in a single sentence, what the model is trying to minimize, the model is not protecting anyone.
The second is asymmetry of cost. A protective system must be willing to be wrong in the player’s favor. False positives, where a healthy recreational player is offered a break, has cooled down or is subject to enhanced due diligence, must be treated as acceptable collateral. If the internal debate is dominated by the “over-intervention” problem, the incentives are already skewed.
The third is isolation from commercial telemetry. The marketing stack should not inform the safer-gambling stack. A player flagged by the protection model should not, that same afternoon, receive a reactivation bonus triggered by the CRM or PAM platform. That this still happens routinely across the sector is a documented failure of governance.
The fourth is explainability. The UKGC’s emphasis on algorithmic transparency is not a bureaucratic inconvenience. It is a moral minimum. If you cannot tell a regulator, a court or a bereaved family why your system did or did not intervene, the system is not fit for purpose.
The Lifetime-Value Paradox
Here is where boards and marketing must be honest with themselves. A meaningful deployment of protective AI will, in the short term, reduce revenue. It will close accounts that would otherwise have deposited. It will block bonuses to players who would otherwise have returned. It will generate friction where the commercial function would prefer velocity. If the introduction of AI-driven player protection has had no measurable negative effect on your top-line GGR, the probability is high that it is not doing what you claim it is doing. This is the question I would invite every CEO to put tomorrow morning to their compliance officer: Show me the revenue we have walked away from this quarter. If the answer is a round number or, worse, a blank stare, the protection is at best theatre.
The Commercial Case for Genuine Protection
None of this means the two objectives are permanently irreconcilable. The emerging evidence from the more-mature Northern European markets suggests the opposite.Operators in the Netherlands and Sweden, where interventions have been aggressive and prescriptive, have found that players subjected to early, well-calibrated protective outreach tend to demonstrate higher long-term engagement, lower post-intervention churn, and dramatically reduced complaints volume The recreational player, treated well, stays. The problem player – allowed to deteriorate – eventually leaves in a manner that is catastrophic for both parties and increasingly adverse for the operator’s banking relationships, payment rails, and regulatory standing. The lesson is not that protection is free. The absence of protection is, in the medium term, is considerably more expensive.
There is, therefore, a genuine commercial argument for genuine protection. But it is a medium-term argument and it requires a CEO willing to absorb short-term NGR compression in exchange for a sustainable license – and a durable customer base. In a public-markets environment that rewards the next quarter, that willingness is not evenly distributed across the sector.
The Regulatory Trajectory
The direction of travel is unambiguous. The UK Gambling Commission’s financial vulnerability regime, the Department of Culture Media and Sport, DCMS implementation of the Gambling Act White Paper, and parallel moves from the Malta Gaming Authority and Sweden’s Spelinspektionen point toward the same end state. Operators will be required to demonstrate, not merely assert, that their automated systems act in the player’s interest. “We have an AI model” will no longer suffice. The questions will be: What does it optimize? Who audits it? When it is wrong? In whose favor does it err?
Those who cannot answer these questions with a straight face in 2026 will be answering them under oath in regulatory reviews in 2027.
The Governance Architecture
What is required is a governance framework as rigorous as the one applied to financial controls. Separate teams. Separate objective functions. Separate reporting lines to the board. A safer-gambling function that reports not to the commercial director but directly to the CEO and, via the audit committee, to the board itself. Independent third-party review of model performance. And a cultural understanding, articulated from the top, that revenue foregone in the service of player protection is not a failure but a measure of integrity.
This is not a call for more committees. It is a call for the structural separation of competing interests. That is the principle that governs every other regulated industry, from banking to pharmaceuticals.
The Industry-Honesty Test
The honesty test for AI in our sector is simple and very uncomfortable. That’s a fact. Show me the model. Show me the objective function. Show me the players it protected this quarter and the revenue you walked away from as a consequence. Show me the internal debate and show me who won it. If you cannot answer, any of the above, you do not have a player-protection system. You have a marketing story. In the regulatory climate now taking shape, a marketing story is the most expensive thing you can own. It’s a fact.
ABOUT MARK MCGUINNESS
Mark McGuinness is the Editor in Chief of Sports Betting Operator and a global marketing leader with 20-plus years of experience across Web3, iGaming, and crypto. He specializes in building community-first brands, scaling growth strategies, and integrating NFTs, tokens and decentralized ecosystems into engaging digital experiences. As CMO of Devilfish.com, he is redefining social poker for a new generation of free-to-play players via monetization of digital avatars and micro-transactions within the co-creator ecosystem.
