Growth
11 min read
By
Puli Liyanagama
CTO & Co-founder
Nov 4, 2024
Candy Crush taught us that a segment-based-system doesn’t work
It’s 2015, and we’re in King’s Shared Tech and Player Kingdom office in sunny Barcelona. Candy Crush is at its peak—over a billion downloads and 250 million daily active players. We’re working through a problem, when our colleague leans in with a grin. “What if we added Game Progression State as another segment dimension? We could group players based on their level bands, making our campaigns even more targeted.”
We thought, “Oh great, just what we needed… more segments. We already have Spend state, Engagement state, Churn probability, Geographic region, Previous campaign participation and many more dimensions. Adding this will multiply our segments by tenfold! Tens of thousands of segment permutations. No big deal. Just the small matter of keeping all that up-to-date.”
Little did we know that this moment would be the catalyst for our eventual shift away from a segment based system entirely. It set us on a path to find a better way—real-time AI personalization that’s not only scalable but also cost-effective.
The Three Layers of “Segmentation”
Let’s make sure we’re on the same page about what “Segmentation” really means. We view the evolution of segmentation as having three key stages.
The first stage was static segmentation. This approach creates fixed lists of players based on common traits and updates them every few days. Both the segment and the player are treated as static during that period.
The second stage moved to a rule-based system. While the lists are no longer static, the rules are. For example, a marketer might set a rule like, “If the player lives in South London and reaches level 10, show this specific offer.” The rule stays the same, but the player’s behavior doesn’t.
The third stage? It’s no longer segmentation in the traditional sense. The rules are dynamic, adapting in real time to each player’s unique behavior. Neither the player nor the rules are treated as fixed. This is the holy grail of personalization—something many still believe is out of reach. But here’s the exciting part: it’s not. We’ve built it.
The first two stages of segmentation have clear limitations, which we dive into in this article. Yet, they remain the standard for 99.9% of the industry today. The third stage is the future, and it’s already here. We’ve built it.
Why segments can’t keep up
On paper, segmentation sounds like the golden ticket to personalization. Group players by behaviors, interests, and habits, and give each group exactly what they need. Simple, right? Except, it doesn’t work.
Here’s why: take the classic “payer segment” as an example. This is a list of all players who’ve ever made a purchase. The problem? It lumps together players who bought something 40 days ago and haven’t spent since with those who’ve been increasing their spend consistently week after week. The difference in these players' behavior is obvious, yet they’re grouped together—despite having completely different motivations and needing different strategies.
In gaming, data gets stale faster than office coffee. Players are constantly evolving. One minute they’re crushing candies and leveling up, the next they’re diving into new live-ops events. By the time you’ve calculated the perfect segment, the player’s behavior has already changed.
Back then, we built custom tech to keep up. Fast, optimized databases? Check. Compressed bitmaps to calculate segment overlap in milliseconds? Absolutely. Systems that updated data every few minutes? We had that too. We threw everything we had at this—time, engineers, marketers, and specialists. But as the complexity grew, so did the cost.
And after all that effort, here’s the kicker: it still wasn’t enough.
The Breaking Point: Segmentation Fails to Scale
At Data Tiger, our last start up, we were determined to solve this segmentation headache once and for all. We did things differently - stage two of the evolution. Instead of painstakingly pre-calculating segments, we built a real-time trigger system. It was a “simple” rule based system: evaluate each player’s behaviour in the moment and make decisions in real-time. Data didn’t have time to get stale anymore because we weren’t relying on old data in the first place. And this approach scaled beautifully. Apple acquired us for it.
But, as is often the case, solving one problem reveals another. The more complex our user journeys became, the more tangled they got. Every new rule added a new branch to an ever-growing web.
The truth hit us like a sugar crash: the real weak link in personalisation wasn’t the data or the tech—it was the human.
Enter Machine-Driven Personalisation: Ending the use of segments for good
We realised that relying on humans to manage these journeys and rules just wasn’t scalable. So, when we started Metica, we decided to take humans out of the operational loop. Instead, we now focus on empowering strategists to set high-level goals and then let machines handle the rest. The machine, given a pool of actions—whether it’s a personalised offer or an in-game event—can pick the right action for each player based on their current behaviour, all in real-time.
We accomplish this by leveraging advanced machine learning techniques such as supervised learning, which predicts player behaviours like churn or spending based on historical data. Reinforcement learning allows the AI to continuously optimize player interactions, adjusting actions to maximise engagement or revenue. Real-time predictive models like our pLTV model help estimate a player's future value, using dynamic data to predict engagement and spending more accurately as player behavior evolves.
No more worrying about stale segments, no more mind-boggling complexity, and definitely no more piñatas of data hanging over our heads. The AI learns and adapts, optimising the player experience continuously without human intervention.
Segmentation is Dead: Here’s why
Here’s the problem with segmentation: it assumes players are static, that you can slot them neatly into predefined groups. But the reality is, players are constantly evolving. And so does the game. The big spender today might churn tomorrow, and the non-spender might surprise you with a purchase. Using segmentation is like trying to navigate using last week’s map—it’s outdated the moment you start.
Real-time personalisation, on the other hand, moves with the player. It’s fluid and dynamic, responding to each player as they are right now. No more grouping, no more stale data. The best part? It scales without getting complicated. Segmentation had its moment, but we don’t need it anymore.
Segmentation isn’t just inefficient; it’s a relic. As AI continues to evolve, the need for predefined segments will vanish altogether. Instead, we’ll have systems that treat each player as a unique individual, adjusting their experience in real-time.
The future of personalisation in gaming
It baffles us that "segmentation" is still talked about as a good thing, even in 2024. It’s still believed by many to be something to aim for, as if it’s a sign of progress when, in reality, it’s holding games back.
The world has moved on a lot since those early days at King, and so has technology. Our mission has always been to drive the evolution of how games connect with their players. That’s why, at Metica, we believe the future of personalization isn’t about putting players into static categories. It’s about recognizing each player as their own unique category, adapting to their behaviors in real-time. With AI taking care of the heavy lifting, carefully guided by human oversight, we’re already making this a reality. We’ve written a blog on personalization here. It works. Don’t be late to the game. We’ve seen significant improvements in lifetime value (LTV) with our clients, proving that AI-driven personalization is the key to transforming player engagement and retention.
If you want to get into the weeds with us on all things LTV optimization for games, and ROAS optimization for games, or just to learn more about this, schedule a call.
We are also offering a free analysis to discover how we can unlock opportunities for your game.