Performance Creative System

When I arrived at TGC, Sky's user acquisition creative was leading with beauty — sweeping landscapes, atmospheric music, the feeling of flight. It was true to the game. It wasn't working hard enough for the business.

The problem wasn't the aesthetic. It was the assumption underneath it: that a beautiful image would be enough to make someone stop scrolling, understand what they were looking at, and decide to download an unfamiliar game. Beauty without specificity doesn't convert. It gets admired and skipped.

Sky had been running paid UA for years — but without a unified audience architecture, the creative was optimizing for the wrong signal. Different teams were briefing different kinds of creative with no shared framework for who the game was actually for, or what language would make a specific kind of person feel seen enough to act. The result was beautiful creative that reached a broad audience and converted a narrow one.

I rebuilt the audience persona architecture from scratch — not demographic segments but identity-based profiles: who these people actually were, what they were looking for, and what language would make them feel seen specifically enough to act. Introverted players. Neurodivergent players. People who wanted social connection without social risk. Cozy gamers. Couples looking for something to do together. Each group needed different creative, different copy, different proof points.

From that architecture I developed a creative testing system that could run dozens of variations simultaneously — different hooks, different emotional registers, different formats native to each platform — and read the signals quickly enough to shift budget toward what was working before a campaign window closed. The metric I optimized toward wasn't click-through rate. It was installs per mille: the quality of the conversion, not the volume of the curiosity. Those are different things and optimizing for the wrong one produces the wrong audience.

One important caveat on the data: UA attribution was fragmented before 2023 due to a platform migration, so the clearest apples-to-apples comparison is 2023 to 2024. Within that window, IPM grew +153% year on year while CTR remained essentially flat — meaning the same proportion of people clicked, but significantly more of those clicks converted to installs. That's an audience quality signal, not a volume signal. The goal was never more clicks. It was better ones.

Over the full tenure, the system produced 40+ monthly campaign and evergreen ad variations localized for seven markets, with consistent improvement in acquisition efficiency year over year.

The measurement layer went deeper than standard UA metrics. Using Affogato — a real-time player insights platform — I tracked not just CTR and IPM but downstream player quality by segment: which audience, which creative angle, which format drove high-LTV installs versus cheap ones. Those findings briefed the next creative wave. The system produced 180+ assets annually, localized to 5+ languages, continuously refreshed based on signal rather than assumption.