Most studios do not have a creative problem. They have a creative testing problem. Ideas get made, a few get launched, one looks good for a week, and nobody can say afterwards why it worked. UA creative testing for games fixes that by turning each ad into an answer to a specific question.

This is the framework we use: a weekly cadence, a way to test hooks, rules for when to kill a creative, and a section on scaling creative production while keeping your game’s look and feel intact.

Start with one question per test

A useful test begins with a hypothesis you can write in one sentence. “Opening on the fail state will earn more installs per thousand impressions than opening on the menu” is a hypothesis. “Try a new video” is not.

Two things follow from that. First, change one variable at a time. If a new video has a different hook, a different voiceover and a different ending, a win teaches you nothing. Second, write down which metric decides the test before you look at any data. Choosing the metric afterwards is how weak creatives get promoted.

Pick the right metric for the stage

Different metrics answer different questions, so use them in order:

  • Hook rate and view-through tell you whether the first seconds hold attention.
  • Installs per thousand impressions (IPM) is the best early read on a creative, because it holds impression volume constant and isolates the effect of the ad itself.
  • Cost per install is useful, but it moves with auction conditions as well as the creative.
  • Day 7 ROAS and retention decide whether the players a creative brings in are worth having.

A creative can win on IPM and lose on ROAS if it attracts people who never come back. So use early metrics to shortlist and later metrics to decide. Our post on what a good ROAS looks like covers how to set the targets those later metrics are judged against.

A weekly cadence that works

Testing only works if it is routine. This is the rhythm we recommend:

  1. Monday: plan. Pick the hypotheses for the week and agree the deciding metric for each. Keep it to a small number of concepts so each gets enough budget.
  2. Tuesday and Wednesday: launch. Put the new creatives live with equal budget so delivery does not pick a winner early.
  3. Thursday to Sunday: let it run. Resist reading results on day one. Networks need a learning phase, and early numbers are taken mid-calibration.
  4. Next Monday: read and decide. Shortlist on IPM, confirm on the later metrics where the data has matured, then promote, iterate or retire.

How fast you can move depends on spend. At low spend, a two-week cycle is realistic because a test needs enough installs to mean anything. At higher spend, weekly or even faster rotation is normal, and it is the best defence against creative fatigue.

How to test hooks

The hook is the first two or three seconds, and it is where most creatives are won or lost. Because it is short, it is also the cheapest thing to test.

Keep the body and ending of the ad identical and change only the opening. Good hook variables include:

  • Starting on a near-fail moment versus a calm moment.
  • Showing the core mechanic immediately versus showing a result first.
  • A bold on-screen question versus no text at all.
  • A close-up of the action versus a wide view of the whole screen.

When one hook beats the rest, you have found something that can be reused across new videos, playables and end cards. That is far more valuable than one winning video. For a wider view of formats and where hooks fit, read our guide to which creative formats convert best.

Where to run the tests

There are two schools of thought. One is a separate test campaign with a small budget, so a weak creative cannot damage your main campaign. The other is adding new creatives to campaigns that are already spending, so they are tested in real auction conditions.

We lean towards a controlled test first for brand new concepts, then moving promising ones into the main campaigns. Whichever you choose, keep the conditions clean: the same countries, the same placements and the same time window for every variant.

Each network also has its own habits:

  • Applovin: organise creatives into sets by concept, with the video cuts, playable and end card for that idea together. Our guide to scaling Applovin campaigns explains how to raise budgets without resetting what the system has learned.
  • Google Ads: App campaigns combine your text, image and video assets automatically. After an asset leaves its learning status, Google labels it Low, Good or Best. Add more like the Best ones, and replace Low ones gradually so performance does not swing.
  • Unity and Mintegral: treat each as its own test, since a creative that does well on one network can behave differently on another.

Kill rules

Without rules, bad creatives live far too long. Agree the rules before the test starts, so nobody argues about them afterwards.

  • Kill a creative if it has had its fair share of budget, the learning phase is over, and it trails your control on the deciding metric.
  • Do not kill on a single bad day. Judge on a window of several days.
  • Hold a creative that wins on IPM but has not yet shown ROAS, until the later data arrives.
  • Retire gradually. Reduce spend instead of switching off at once, so delivery stays stable while the replacement ramps up.
  • Never declare a winner on a tiny sample. Small numbers produce false winners.

Budget matters here. A test needs enough conversions to read, so work backwards from your cost per install and the number of installs you need per variant. If you cannot afford enough per variant, test fewer variants. Our guide to planning your UA budget shows how to set aside a standing testing share.

Scaling creative production without losing your style

The second question in this topic is a common one: which platform is best for scaling creative production while preserving your game’s visual style and brand? There is no single answer, and we would be wary of anyone who gives one. What matters is the process, not the tool.

Whatever you use, we suggest this:

  • Write a short style guide. Colours, camera angles, UI treatment, tone of voice and anything the game would never show. Every creator, internal or external, works from it.
  • Build from your real assets. Use captured gameplay and your own art as the base for all variants, so everything looks like the game players will actually install.
  • Separate the template from the variable. Fix the look, then vary only the hook, the text or the ending. This is how one concept becomes ten variants without drifting off-brand.
  • Use templates for playables. A reusable playable shell with swappable levels and end cards is faster than building each one from scratch. Our playable ads guide covers the approach in more detail.
  • Review before launch. One person owns the brand check, so volume never overrides quality.

If you do use a creative production platform or an outside studio, judge it on three things: can it work from your own assets, can it hold your style guide across dozens of variants, and can it deliver in the formats your networks accept. A tool that cannot do all three will cost you time, whatever its demo looks like.

The short version

  • Test one variable at a time, with a written hypothesis and a metric chosen in advance.
  • Shortlist on IPM, decide on Day 7 ROAS and retention.
  • Run a weekly cadence, and test hooks first because they are cheap and reusable.
  • Agree kill rules before launch, and never judge on one day or a tiny sample.
  • Scale volume through a style guide and templates, not by loosening the brand.

If you want a second pair of eyes on your current creative pipeline, get in touch and we will look at how your tests are structured.