1. Home
  2. Guides
  3. Creative testing budget

How to budget creative testing and find your winning ad

How much to spend per creative, how many to test, when to kill and when to keep going, on Meta, Google, AppLovin and TikTok.

By Eugene Yakovlev · Updated November 21, 2026 · 10 min read

Spend is extremely lopsided

In gaming video, the top 2% of creatives accounted for 53.5% of spend in Q1 2025, and 60.7% in Q2 2026 (AppsFlyer). Across all verticals, top 10% of creatives account for >3/4 of spend.

However carefully the budget is allocated, the networks will starve most of your ads. The purpose of creative testing is to have enough tickets that one will fall within the top 2%.

Gaming apps spending $7M+ a quarter had ~4,000 variations in Q1 2026, an increase of +45% YoY. Apps under $1M a quarter were static or decreased. If in this category, consider a more refined testing strategy, as you cannot win on volume.

How many tickets you need

578,750 Meta creatives from 6,015 accounts were benchmarked by Motion in 2026. A winner was defined as having 10x or more the median spend in the account, and $500+. 3.8% of creatives in accounts with <$10K monthly spend became winners, and 8.2% in accounts with >$1M monthly spend.

If you create 20 creatives a week, you can expect 1 winner. This is across all verticals. No specific games figure exists, but this is a useful figure to know. If you create 1 or 2 new ads per week, it’s unlikely you’ll see much change.

Concepts beat variations

Andromeda is an ad retrieval algorithm rolled out by Meta in 2025. By Meta's own figure, it gave a 2.4% lift in conversion rates on Facebook. Practitioners have observed that very similar creatives will be grouped together. Copy, colours and music changes will not create new creatives. If the same video is used in a ‘hook test’, it will be considered as 1 creative, even if there are 5 variations.

To get the best out of this, you should aim for 8-12 unique concepts, with a couple of variations within each. To ensure uniqueness, change the hook, value proposition and character. Do not change fonts.

Money per creative

There is no true flat rate. You'll see $100-$500 per creative, 40x CPI, testing at 5-10% or 20-30% of budget. These figures vary between accounts depending on the goals.

The cleanest method is to calculate based on number of installs required for reads. Multiply this number by CPI. If using $2 CPI and needing 50 installs, that's about $100 per variant. Divide this amount between creatives equally, so the delivery system doesn't pick winners for you. Start with IPM for initial reads, and benchmark separately across channels.

Target CPIInstalls for a first readSpend per creative20 creatives a week
$250$100$2,000
$450$200$4,000
$850$400$8,000

If you cannot afford this number of variants per week, reduce the number of concepts in testing and try to make them more unique.

Platform by platform

Meta

  • Meta’s in-built creative testing allows for 2-5 creatives to be tested within an existing ad set. It is recommended to spend no more than 20% of budget. By default, tests will run for 7 days.
  • Only supports Highest Volume bidding. No cost goals or maximum bids can be set. Aim to get ~50 optimised events per week to get meaningful results.
  • Advantage+ app campaigns allow for manual testing of 50 combinations.
  • Many people follow a similar evergreen strategy (from Cedric Yarish via RevenueCat). They will have ~50 ads running, and Meta will put ~80% spend in top 3. Every week, remove the bottom 5 and add in the top 5 from testing. Against a $5 average CPA: $4 is good, $8 is maybe, $10+ is out (0.8x average is good, 2x is bad).

Google App Campaigns

  • Minimum daily budget = 50 x Target CPI (or 10x Target CPA for in app action campaigns). Do not make any disruptive changes before 100 conversions.
  • Max 20 videos, 20 images and 20 HTML5 assets per ad group. Fill in gaps using assets similar to your Best ones. Only replace Low assets when you have reached capacity. Replace gradually.
  • Low/Good/Best depend on other creatives in the campaign and do not affect serving on their own.
  • App asset A/B testing splits 50/50 and requires 80% confidence. Only available on Android.

AppLovin

  • Do not pause poor performing creatives. It is recommended by AppLovin not to manually pause creatives as it will interfere with what the optimiser is doing.
  • End cards drive >50% of clicks in creative sets and should be treated as creative.
  • 1 creative set per concept. Add creatives to evergreen campaigns (ones that are already spending), rather than a separate test campaign. Get ~10 conversions per day per test. Do not choose to stop testing based on CTR, always use cost per conversion or ROAS.

TikTok

  • 3-5 creatives in ad group, 3-5 ad groups in campaigns.
  • Value proposition in <3s, hook in <6s.
  • Rotate every 1-2 weeks. TikTok will burn through creatives very quickly.

Metrics, in order

  1. Hook rate (3-second views / impressions). A good hook rate is between 25-35%. Below 25% means your opening is bad.
  2. Hold rate (15-second views / 3-second views). Good hold rates are ~25%.
  3. IPM (installs per 1,000 impressions). First real indicator of conversion.
  4. CPI and D7 ROAS. Only metrics which pay your bills.

Based on these metrics, you can determine what needs improving in your ad. A high hook rate and low hold rate indicates a problem in the middle of your ad. A low hook rate and high hold rate indicates that the concept is good, but the first second of the ad is poor. These figures are pulled from practitioners, rather than from platforms. Consider them as a guide and when you have enough median data, change to your own.

What a winner is, and how long it lives

Ad is good if it outperforms current control by CPI/D7 ROAS, and sustains this for more than a couple of days. A single good day in performance is noise.

Many sources suggest game ads have lifespans of 9 or 15 days. However we could find no primary source for this data, and one publisher removed their figures due to this issue. The best data we could find was from AdMapix for the US. In launch campaigns, new creative sets have a lifespan of 25-67 days. Evergreen, proven assets have very long lifespans. One playable in Whiteout Survival was live for 486 days, one video in Last War for 361 days, and one ad in Royal Match for 305 days. However, new tests at one publisher were assessed within 5-11 days.

Therefore consider the lifespan of tests to be very uneven. New tests should be killed quickly. Once you have proven winners, continue showing them as long as IPM remains steady. If IPM begins to decrease without a change in spend, you are seeing fatigue. At this point you would want to brief the next concept in the same direction, not from scratch.

Formats that punch above their weight

In 2025, Liftoff’s creative index suggested that in high spend games, 35% more spend was in playables, compared to other games. In smaller advertisers, playables converted at up to 16x better (impressions to installs) than other formats. (Not directly comparable, but significant.) UGC had a 152% improvement over other formats in the same metric. If your test strategy contains only videos and gameplay, this is the first gap to close.

Our rules of thumb for smaller studios

We would recommend approx. 20-30% of budget spend on testing, and the remainder on scaling successful creatives. If you have a small budget, reduce the number of concepts being tested, but ensure they are very different to each other.

Reasons to stop a test early:

  • Hook rate <25% after 1,000-2,000 impressions.
  • Half the budget has been spent with CPI >2x target.
  • IPM decreasing without change in spend in an old winner (fatigue).

Our take

Sometimes the most unexpected creatives, maybe made technically the weakest, turn out to be the most effective. Or creatives where some small mistakes were made, or some other nuances that, let's say, humanize this content and make it more natural. Sometimes this works really well.

— Eugene Yakovlev

A one-page plan

  1. Calculate weekly testing budget as CPI × 50 installs × concepts.
  2. Ship 8-12 unique concepts per week (fewer if needed, but never fewer unique ones).
  3. Look at hook and hold rates after day 1, IPM after first read. Look at CPI and D7 ROAS at end of test.
  4. Add winning creatives to evergreen. Do not change them until IPM begins to drop.
  5. Feed data from losing creatives into next brief. Even if a concept has failed, it may still be valuable as it tells you what kind of opening is failing.

A popular shortcut is to test hooks on organic social media to avoid spending on clear underperforming creatives. While this does not replace paid testing, it helps to eliminate clear failures cheaply. We have written a guide on testing on organic first, then scaling winners.

Want to see what we'd make for your game?

Send us the store link and any materials you have. We'll reply to confirm a couple of details, then make two concepts for your game, free. No call needed.

  • A reply from us within one business day
  • Two concepts within 48 hours after you confirm
  • Watermarked previews, no strings attached

We only use this to send your concepts. No newsletters. Prefer email? info@scrollstoplab.com