Reading Your CPI Results Without Fooling Yourself

CPI measures how much you paid for one install during a test, and it is the most common first filter publishers use. But the number is easy to misread. A low CPI from the wrong audience means nothing. A high CPI from one creative does not condemn the concept. And a test with too few impressions is noise dressed as data.

Why CPI gets misread

Three reasons, all common.

1. The number feels precise. A CPI of $0.27 looks like a measurement, but it is an average across audiences, times, and placements that may vary by a factor of three.
2. It is the first number you get. Other metrics take days or weeks to mature. CPI arrives fast, and fast numbers attract premature conclusions.
3. It conflates concept and creative. A high CPI could mean people do not want the game or it could mean the ad did not show the game well. Those are very different problems.

Understanding what the number can and cannot tell you is the difference between acting on signal and acting on noise.

What CPI actually measures

CPI measures the cost of converting an ad impression into an install. It reflects three things combined.

• Click-through rate (CTR). How many people tapped the ad after seeing it. This is mostly about the creative.
• Install rate. How many people who reached the store page installed. This is partly about the store listing and partly about the concept.
• Auction dynamics. What you paid depends on who else is bidding for the same audience, which changes by day and by region.

A single CPI number mixes all three. When the number is bad, you need to pull it apart to know what is bad.

CPI tells you how cheaply you acquired installs. It does not tell you whether those installs were worth acquiring.

Sample size: when to trust the number

The most common mistake is reading a test before it has enough data.

A minimum viable test needs at least 1,000 impressions per creative variant, and ideally closer to 5,000. Below that, random variation dominates the result and differences between creatives are not reliable.

If your budget only allows 2,000 impressions, run one creative rather than splitting across four. One result you can trust is worth more than four you cannot.

The budget math for running a proper test is laid out in Creative Testing on a $500 Budget.

Separating creative signal from concept signal

This is the critical question after a high CPI: is the game the problem or is the ad the problem?

Three indicators that the creative is the issue.

• Low CTR but decent install rate. People who reach the store page install, but not enough people are clicking. The ad is not compelling.
• CPI varies widely across creatives. If one creative hits $0.20 and another hits $0.80, the concept can reach $0.20 and the question is creative quality.
• The ad does not show the core loop. If the creative shows a cinematic or a misleading mechanic, the CPI reflects something other than your game.

Three indicators the concept is the issue.

• Decent CTR but low install rate. People click but do not install after seeing the store page. The game itself is not converting.
• CPI is consistently high across multiple creatives. When three different ads all produce the same high cost, the common factor is the game.
• The store page accurately represents the game. If the ad and the store are honest and people still do not install, the concept is not pulling.

The distinction matters because creative problems cost days to fix and concept problems cost months. Spending a month on a creative problem is a waste. Spending a day on a concept problem is a bigger waste, because it produces false hope.

The creative side is covered in How to Write an Ad Hook in Four Beats.

Geographic and audience traps

CPI varies dramatically by market, and testing in the wrong one produces misleading results.

A common pattern: test in a cheap market to save budget, get an encouraging CPI, then discover it doubles or triples in the target market.

Three rules for geography.

1. Test in your target market. If you plan to launch in the US, test in the US. A $0.15 CPI in Southeast Asia tells you nothing about US performance.
2. Be careful with broad targeting. Letting the platform optimise for cheapest installs will find the cheapest installs, which may not be the audience that retains.
3. Compare within the same market. A CPI comparison between two creatives is only valid if both ran in the same region during the same period.

The same applies to audience segments. A test targeted at "all users" produces a different CPI from one targeted at "people who play puzzle games." The narrower test is more expensive and more informative.

When a retest is valid

Not every high CPI deserves a retest. Some do.

A retest is valid when you changed something specific and want to measure the effect of that change. Running the same test again because you did not like the result is not a retest, it is hoping for a different outcome.

Valid reasons to retest.

• New creative that shows a different part of the game. The concept is the same, but the presentation changed meaningfully.
• Updated store page. If install rate was weak, a better screenshot set or video can shift it.
• Different market or audience. If you tested broadly and want to try a specific segment.
• Significant game change. If you rebuilt the opening based on what the first test taught you, per Why Most Prototypes Fail in the First Ten Seconds.

A retest without a change is not a strategy. It is a delay.

Reading the result alongside retention

CPI alone is half the picture. A cheap install that leaves on day one costs money, not less.

The real question is whether CPI multiplied by retention produces a viable unit economy. A $0.30 CPI with 40% day-one retention is often better than a $0.15 CPI with 15% day-one retention.

Publishers look at both numbers together, and so should you. The thresholds differ by publisher and are covered in How Publisher Submission Bars Differ.

If your CPI is just above the bar but your retention is strong, say so in the submission. A publisher looking at both numbers may greenlight what a CPI filter alone would reject.

What we would do

Run the first test with one or two creatives in your target market, spend enough to reach 5,000 impressions per creative, and then stop and read the data before making any decisions.

Pull apart CTR and install rate. If CTR is the weak link, make new creatives. If install rate is weak, revisit the store page. If both are mediocre, pressure-test whether the concept is the problem.

Never retest without changing something specific, and always note what you changed and why so the second result means something.

The short version

• CPI conflates creative quality, concept appeal, and auction dynamics.
• You need at least 1,000 impressions per creative to trust the direction, and 5,000 for reliable comparison.
• Low CTR with decent install rate points at the creative. High CTR with low install rate points at the concept.
• Test in your target market, not the cheapest one available.
• A retest is only valid if you changed something specific and measurable.
• Read CPI alongside retention, because a cheap install that leaves costs more than an expensive one that stays.

Run one honest test in the right market before deciding anything. If you want help reading your CPI results, send us the numbers.

Related reading: Creative Testing on a $500 Budget, How Publisher Submission Bars Differ, and CPI Went Up After a Test: Now What.