Fraser, you’ve said that personalisation is talked about far too generically. When operators say “personalisation,” what are they usually referring to, and what are they confusing it with?
I think it’s important to start by saying that personalisation is one of the most important concepts in digital product strategy, and iGaming is no exception. Done well, it underpins almost every meaningful dimension of the player experience, from how content is surfaced to how journeys are structured and offers are timed.
The problem is that the term itself has become too broad. It’s now used to describe everything from targeted emails and CRM journeys to AI-driven recommendation engines, VIP management and product configuration. Once everything becomes “personalisation,” the word starts to lose its usefulness.
There are different types at play here: content personalisation, UX personalisation, journey personalisation, offer personalisation and product configuration. These are different problems, requiring different data, different capabilities and different success metrics, but the industry often collapses them into one catch-all concept.
The other important point is that operators are often solving for the wrong interpretation of a genuine user need. Research tends to show that when customers say they want personalisation, what they really mean is relevance and lack of friction. They want to get to what they want quickly and easily. They are not asking for algorithmic sophistication for its own sake.
Where do you see the biggest category errors being made?
One of the biggest is the confusion between explicit and implicit personalisation. Explicit personalisation is user-led. That’s where players actively configure their own experience, whether through preferences, settings or the type of messaging they want to receive. Implicit personalisation is system-led. That’s where the platform infers what the player might want based on their behaviour and adapts accordingly.
In gaming, we focus far more on the implicit side than the explicit side. That is partly because the rise of AI and machine learning has created a kind of hype cycle around what is possible. People get drawn to the sophistication of the engine, rather than stepping back and asking whether they are solving the right problem in the first place.
What’s interesting is that the companies most often cited as proof that implicit personalisation works, such as Netflix and Spotify, invest heavily in explicit signals. Users rate things, like things, dislike things and build taste profiles. Those explicit inputs are what make the implicit systems better. In gaming, we tend to skip that step. We run models on historical behaviour and hope that is enough to drive improved engagement.
I don’t really see explicit and implicit as a spectrum. I see them as two sides of the same coin. The best recommendation systems are enhanced by explicit input, but that’s an area the gaming industry still underuses.
Gaming is habitual and time-based, as you’ve pointed out. How does that reality fundamentally limit the value of hyper-personalised experiences once a player already knows what they want to do?
A lot of the research I’ve been involved in, including ethnographic work and focus groups, points to the same conclusion: gameplay is much more habitual than discovery-led. For the majority of returning sessions, the player is trying to complete a task. They want to get to a market, place a bet or play a game they already know. They are not necessarily in an exploratory mindset.
The behavioural science is clear on this. Once a behaviour becomes habitual, motivation and deliberation largely drop out because it becomes routine. So, if someone is logging in every Tuesday evening to play the same game, surfacing recommendations at that moment may be working against their psychology rather than with it.
That doesn’t mean personalisation has no role, but it does mean its value is concentrated at very specific lifecycle moments. Onboarding is one. Reactivation is another. New product launches can also be relevant. But applying hyper-personalised experiences as a persistent layer across every session, regardless of context, is where the industry often gets it wrong.
At what point does personalisation start getting in the user’s way? What does over-personalisation look like in a live casino or sportsbook environment?
It starts getting in the way when it interrupts the journey instead of supporting it. A user in an in-play sportsbook environment or a live casino context is usually high-intent and low-tolerance for friction. If they have to navigate past recommendations, changing content hierarchies or shifting layouts just to get to a known destination, you’ve made the product worse in the name of relevance.
There’s a core UX concept called wayfinding, which is about helping people orient themselves and move confidently through an interface. Users rely on familiar landmarks and patterns. If a lobby is reordered every time they visit, or navigation shifts session to session, those habitual paths are broken.
The safest model is a static structure with personalised content inside it. So, a “recommended for you” component can exist in a fixed place, while the titles within it change. That works. But when the architecture itself starts moving around, that’s when personalisation becomes disruptive.
There’s a good anecdote from Apple when it introduced shuffle on the iPod. The system was genuinely random, but users complained it didn’t feel random because hearing the same artist twice in a row felt wrong. Apple had to make shuffle less random so it felt more random. It’s a useful reminder that algorithmic logic and human perception are not the same thing. Personalisation can be technically correct and still feel jarring in practice.
Discovery often happens off-platform rather than on it. If that’s the case, are operators over-investing in on-site personalisation at the expense of fixing core UX fundamentals?
In many cases, yes. This is not unique to gaming, of course. Across digital industries, a lot of purchase decisions and discovery moments happen before the user ever arrives on-platform. That comes through peers, affiliates, search, social, influencers and other external channels.
In iGaming, that pattern is particularly strong. A lot of site traffic arrives with intent already formed. The player has already seen the game, the market or the brand somewhere else, and by the time they land on site they already know broadly what they want to do – which creates an attribution problem. On-site recommendation engines can claim credit for engagement that may have happened regardless, simply because the content was surfaced at the point of visit. That can inflate the apparent ROI of on-site personalisation. In reality, a lot of that intent may have been formed entirely off-platform.
So yes, I do think there is a risk that operators are over-investing in personalisation layers while under-investing in the fundamentals. If the UX is not intuitive, if navigation is broken, if the core journeys are weak, then no recommendation engine is going to compensate for that.
Many personalisation strategies assume high-quality behavioural data. How realistic is that assumption given fragmented data stacks, regulatory constraints and cross-brand complexity?
Honestly, for most operators, the assumption is far less realistic than their ambitions suggest. The fragmentation problem is structural. Players often hold multiple accounts across multiple brands, so any one operator only ever has a partial view of that customer’s behaviour.
Even within multi-brand groups, those signals are not always connected. If you operate several brands in the same market, you may not be able to unify that data in the way a recommendation engine would ideally want. Across regions, it becomes even more complicated because the same player profile may behave differently under different regulatory frameworks and different brand propositions.
So, the data foundation is often much weaker than people assume. Add legacy technology stacks that were never designed for real-time activation, and it becomes clear that the infrastructure challenge is just as important as the data challenge. Building a genuinely complete picture of player intent is extremely difficult in an industry where behaviour is fragmented by design.
From your research work, where does personalisation genuinely move the needle, and where does it mostly create noise?
The most effective form of personalisation the industry has ever had is probably VIP management. It’s deeply personalised, but it’s human rather than algorithmic. Good account managers know their players, understand their preferences and engage with them in a way that feels relevant. Commercially, that has probably been the most powerful form of personalisation in gaming.
The irony is that the players most heavily targeted for algorithmic personalisation are often the ones who need it least. The highest-value players are usually the ones with the strongest habits and the clearest navigational confidence. They already know what they want to play and how they want to play it.
Where personalisation can genuinely add value is at lifecycle moments where intent is not yet fully formed or has lapsed. Onboarding is one. Reactivation is another. Cross-sell can work in some contexts. These are the moments where players are more open to discovery.
There is also a commercial issue that often gets overlooked. A recommendation engine optimising for relevance is not the same as one optimising for business outcome. You might have one title scoring slightly higher for relevance but another delivering significantly higher GGR. If the model is only optimising for its own relevance score, the commercial outcome becomes secondary. It has to be designed into the system from the start.
How should operators think about personalisation relative to UX consistency? Is there a point where tailoring experiences undermines familiarity and user confidence?
Absolutely. Consistency is a prerequisite for good UX. Users build mental models of products. They learn where things are, how flows work and what to expect. That is what allows them to move quickly and confidently. If personalisation starts shifting the architecture between sessions, users pay a relearning cost every time they return. For habitual players, that is particularly damaging because it disrupts the confidence and familiarity they rely on.
So the right model, in my view, is consistency in structure and navigation, with personalisation applied to content rather than architecture. That preserves usability while still allowing relevance. There is also a practical benefit. Consistent UX is easier to test, easier to optimise and easier to improve over time. If you are changing the interface too radically from one session to the next, it becomes much harder to build a reliable view of what is actually improving performance.
AI has made personalisation easier to deploy, but not necessarily better. Are we at risk of automating poor assumptions at scale?
Yes, definitely. A lot of recommendation systems use algorithms that learn from previous behaviour and then keep surfacing whatever appears to have performed well. The risk is that if you are not measuring uplift properly, you create a self-fulfilling loop. Content gets surfaced because it previously performed well, then it performs well because it was surfaced, and the algorithm interprets that as proof that it made the right decision. At that point, the system is not necessarily learning genuine user preference. It is learning its own signal.
One of the ways to address that is through negative sampling, where you deliberately introduce random content into the recommendation set as a control. That allows you to test whether the supposedly “recommended” content is genuinely outperforming what users might have engaged with anyway. Counterintuitively, adding some randomness can improve overall performance because it stops the model becoming too narrow and closed in its own assumptions.
The broader point is that AI will amplify whatever assumptions are already built into the system. If the product logic is flawed, if the data is incomplete, or if the historical patterns are misleading, then AI will simply scale those problems faster.
If you were advising an operator starting from scratch, what would you tell them to get right first before even thinking about advanced personalisation layers?
Fix the foundations. That would be the first message. If deposit errors are not handled clearly, if navigation breaks on mobile, if a player clicks through from an ad and lands in the wrong place, if the KYC journey creates unnecessary friction, those are the problems to solve first. They will nearly always have a higher ROI than an advanced recommendation layer.
The second thing is data hygiene. Understand what data you actually have, what quality it is at, where the gaps are and whether it is genuinely usable for the type of modelling you want to do. This means looking not just at on-platform behaviour, but also at what external variables may matter, whether that is demographics, event calendars or other contextual signals.
And third, be very precise about what you are trying to personalise. Content, UX, offers and journeys are different challenges with different objectives. If you treat them all as one initiative, you end up spreading resources too thinly and not solving any of them properly.
Looking ahead, do you think the industry will move away from “personalisation” as a headline concept altogether? If so, what replaces it as the more honest framing?
I don’t think it will move away from personalisation. If anything, I think it becomes even more central. What changes is the layer at which personalisation happens.
Today, personalisation is mostly built on historical behavioural data, segmentation and outputs pushed into relatively static interfaces. The player experiences it as a slightly better product, rather than as a distinct layer.
What AI and LLMs may change is the interface itself. Over time, I think the distinction between “the product” and “the personalised version of the product” starts to collapse. Instead of navigating a static site and then receiving recommendations, users may increasingly interact with more conversational, intent-driven interfaces that respond in real time to what they are trying to do.
How quickly that happens is hard to say, particularly given regulation, privacy and the commercial reality that some businesses are already stepping back from certain LLM integrations. But directionally, I do think that is where things are heading. Personalisation does not disappear in that world. It simply becomes inseparable from the interface itself.
The post Why iGaming Needs to Rethink Personalisation appeared first on G3 Newswire.
Fraser, you’ve said that personalisation is talked about far too generically. When operators say “personalisation,” what are they usually referring to, and what are they confusing it with? I think it’s important to start by saying that personalisation is one of the most important concepts in digital product strategy, and iGaming is no exception. Done…
The post Why iGaming Needs to Rethink Personalisation appeared first on G3 Newswire.