By Thomas Aigner, SVP of Business Development, Golden Whale
In brief [TL; DR]
AI is becoming easier to access, but that does not mean optimisation is becoming easier to achieve. With growth uplifts of 100% and more, this is a field that can make or break an operation.
In iGaming, the real value comes from connecting the right goal, intervention and model optimisation process in a way that improves live commercial decisions. Internal teams can build strong capability, but without external comparison it is difficult to know whether an in-house model is genuinely competitive, or how much performance and manpower may be going to waste.
Access is not always an advantage
AI has become much easier to access, which is clearly positive for the iGaming industry. No-code tools, pre-trained models and plug-and-play dashboards have lowered the barrier for operators that want to experiment with machine learning, giving more businesses the ability to test ideas that would once have required large specialist teams and significant investment.
The danger, however, is assuming that access to AI automatically creates operational advantage. As more tools become available, many operators are questioning whether AI-driven optimisation can now be built and managed entirely in-house. On the surface, that argument makes sense. Operators have their own data, internal teams understand the business, and modern tools make model building appear faster and more cost-effective than ever before. But optimisation in iGaming is not simply a model-building challenge.
A churn model, value score or behavioural prediction can be useful, but it does not automatically improve retention, reduce bonus waste or increase player value. The commercial challenge is deciding what to do with the prediction: when to act, how generous an intervention should be and whether that action is commercially justified.
That is why model quality is hard to judge in isolation. A model may produce recommendations, but that does not mean it is improving decisions as much as it could. Operators need reference points, ideally external models or benchmarks, to understand what level of optimisation is possible and how much value may still be left behind.
Where internal AI starts to struggle
The first challenge for internal AI is often field of vision. An operator has an innate knowledge of its own business, players and constraints, but it only sees its own environment. That can make it harder to judge the right combination of commercial goal, player intervention and model optimisation process, because the team is working from a narrower set of patterns than a specialist supplier operating across multiple clients, markets and use cases.
Data quality and context also remain central. Most operators have large volumes of data, but that data is often fragmented, structured inconsistently or missing the commercial context needed to support effective decisions. A model can predict churn, value or behaviour, but if the dataset is incomplete or the intervention logic is weak, the outcome will still be weak.
The second challenge is deployment. Many internal projects stop at dashboards, scores or insights. These may help teams understand behaviour, but they do not automatically change player treatment. A prediction that a player may churn is only valuable if it can be translated into the right action across incentives, loyalty, messaging or risk-sensitive growth workflows. Without that operational connection, AI remains something that describes the business rather than something that improves it.
This is the gap between AI as insight and AI as optimisation. Prediction is only the starting point. To create value, AI has to move beyond identifying opportunities and become part of the decision infrastructure that acts on them.
What working AI should prove
Working AI should therefore be judged by commercial outcomes rather than model accuracy alone. Success should show up in higher retention, better reactivation efficiency, improved bonus discipline, stronger margin control, faster experimentation cycles and measurable uplift against existing human-led or rule-based processes.
True optimisation also requires learning from real interventions. Many systems are trained on historical data, but commercial value emerges when the system learns from what happened after a decision was made. Did the player respond? Was the incentive necessary? Did the action improve margin-adjusted value, or simply increase cost? Without that reinforcement loop, AI may identify opportunities, but it will not consistently improve the decisions that shape commercial outcomes.
Build, buy, or blend?
This is also why the build-versus-buy debate needs more nuance. It should not be framed as a simple choice between internal capability and external partnership, because the strongest model is usually hybrid. Operators should absolutely build their internal foundation. They should own their data, strategy, brand, player relationship and commercial boundaries, because those elements are too central to outsource completely. But highly specialised optimisation intelligence is different.
Building a churn model internally is one thing. Building a self-learning decision infrastructure that reliably improves incentives, journeys, retention and margin outcomes across live operations is another. That requires modelling expertise but also experience in pairing interventions with the right methods, workflows and measurement structures.
This is where expert models can create a real advantage. Specialist optimisation models are not just another tool in the stack, they provide a point of comparison for what better decisioning can deliver. In high-frequency environments such as iGaming, even single-digit improvements in optimisation can compound significantly over time, in some cases doubling growth rates over the course of a year. That is why the question is not simply whether an operator can build a model internally, but whether that model can outperform expert alternatives in practice.
For operators that still choose to build in-house, external comparison remains essential. Testing internal models against specialist outside models can reveal how much performance potential is being left behind, where operational workload could be reduced and whether internal teams are spending time on optimisation tasks that could be handled more effectively by machine learning systems already proven in live environments.
Where human teams create the most value
The point is not that operators need to rebuild their technology stack around AI. In many cases, the stronger approach is to add intelligence into existing workflows across campaigns, loyalty, messaging and gamification. Those systems still have an important role to play, but the most valuable decisions increasingly need to be guided by models that can learn, compare and optimise continuously.
This also changes the role of internal teams. As optimisation becomes more AI-driven, teams spend less time manually configuring journeys and more time defining strategy, creative direction and the boundaries within which machine learning models operate. Human judgement remains essential, but it is applied where it creates the most value.
That is where the build, buy or blend debate should land: not in who owns the most tools, but in which approach produces the best decisions when tested against real alternatives. Operators will still need strong internal capability, but the winners will be those that can combine that capability with expert models, external comparison and precision decisioning that turns prediction into action, and action into measurable commercial improvement.
Read More: Behind the Scenes of Event Marketing: Preparing for iGB L!VE 2026
