What AI personalisation really means in mobile games

If you have spent time evaluating AI monetisation tools recently, you have seen the same numbers. Sixty percent higher engagement. Eighty percent higher conversion rates. Double-digit ARPU lifts. Every tool publishes them. Almost none of them explain how those numbers were calculated.

That gap — between a confident headline and the measurement behind it — is where studios get burned.

What "personalised" actually means in most tools

"Personalised" has become the default selling word for anything more sophisticated than a static offer. Some platforms mean it in a genuine sense: the AI considers this specific player's history, session patterns, and spending behaviour before deciding what to show them. Others mean it in a looser sense: different groups of players see different offers based on which spending tier they belong to.

The second approach is standard behavioural segmentation. It has been effective practice in mobile games for over a decade. It does not require a machine learning model. Calling it AI personalisation is technically defensible and commercially misleading.

What neither approach can tell you, without independent measurement, is whether the AI is doing anything that a well-configured rule-based system would not do just as well. True per-player personalisation requires enough signal per player to act on, a model that genuinely fits individual behaviour rather than cohort averages, and a comparison that tells you whether the model's decisions actually moved revenue. Most platform dashboards provide the first two and quietly skip the third.

Why self-reported uplifts cannot be trusted at face value

Here is the typical scenario. A studio integrates an AI monetisation tool in January. By March, revenue is up twelve percent. The tool's dashboard reports twelve percent uplift from personalisation.

What the dashboard cannot account for: whether the studio released new content in February. Whether a content creator posted a video about the game. Whether the mobile games market had a strong quarter globally. Whether seasonal patterns pushed revenue up for every comparable game in the same genre. Whether the studio's organic UA spend increased.

A measurement that compares revenue before and after adding a tool — on the same player base, across a period when other things also changed — is not measuring the tool. It is measuring everything that happened in that window. The AI gets credit for the whole line going up.

This is not fraud. The platforms publishing these numbers believe them. The problem is that "revenue went up while we were running the AI" is a completely different statement from "the AI caused revenue to go up." A studio that cannot tell the difference between those two statements has handed a service a permanent revenue share on their game's organic growth.

The only measurement that answers the real question

The only way to know what a personalisation system actually contributes is to run a group of players through your game without any AI offers, in parallel with the players who do receive them. Same time period. Same seasonal conditions. Same content updates. Same market. The only variable is which group the AI served.

When that comparison shows the AI-served group out-earning the unserved group by a margin that clears statistical significance, you have real evidence. When it does not, the honest answer is that the AI did not help this month.

This is the measurement gamemantra is built on. We hold back ten percent of your players as a permanent baseline group. Every month, we compare actual in-game purchase revenue between the two groups, and you see the same numbers we do.

That structure exists because without it, we have no way of knowing whether we are helping you. And neither do you.

Three questions to ask before you sign up

Most platforms will not volunteer this information. You have to ask for it directly.

Do you run a permanent holdout group, and is it mandatory? The baseline group has to exist for the full duration of your use of the platform — not as an optional test you configure once. If it is optional, most studios turn it off because they believe it costs them revenue from those players. An optional baseline produces optional accountability.

Is your revenue share calculated on incremental revenue or total revenue? Some platforms charge a percentage of all in-game purchase revenue generated while the platform is running. That is a subscription dressed up as a performance fee. It means the platform earns money even in months where the AI made no measurable difference. Genuine performance pricing applies only to the revenue delta between the AI-served group and the baseline group.

Can you see the raw data and verify the calculation yourself? A platform that cannot give you direct access to the event-level comparison — purchase events, player group assignments, the arithmetic — is asking you to trust a number you cannot check. The measurement is only as credible as your ability to audit it independently.

---

Most platforms will give you a confident answer to the first question and a vague one to the second. Very few will give you a satisfying answer to the third.

That is not a reason to stop evaluating AI monetisation tools. It is a reason to make the measurement structure non-negotiable before any contract is signed.

See how we measure uplift → · Talk through it with your game's numbers