Written in by More Playables production team
IPM, CPI, Retention, and ROAS: Reading a Creative Test
Formulas, a synthetic funnel, attribution boundaries, and questions that prevent one metric from standing in for the whole decision.
Metric definitions and synthetic example; no campaign result
Updated
The formulas
Metric | Formula | Question it helps answer |
|---|---|---|
CTR | Clicks ÷ impressions | Did the asset produce the recorded click event? |
CVR | Selected conversions ÷ selected starting population | What share completed the defined conversion? State the denominator. |
IPM | Attributed installs ÷ impressions × 1,000 | How many attributed installs were recorded per thousand impressions? |
CPI | Spend ÷ attributed installs | What spend was recorded per attributed install? |
Retention | Users active at the defined later period ÷ eligible installed cohort | Did the acquired cohort return under the chosen definition? |
ROAS | Attributed revenue ÷ spend | How much attributed revenue was recorded per unit of spend? |
Every formula depends on the reporting system, event definition, attribution model, cohort eligibility, currency, and time window. Preserve those alongside the number.
Synthetic worked funnel
The following values are invented to demonstrate arithmetic. A creative receives 200,000 impressions, 8,000 recorded clicks, 1,000 attributed installs, and $2,500 spend. CTR is 4%. IPM is 5. CPI is $2.50. If 300 of the 1,000 eligible installers meet the stated day-one return definition, that retention measure is 30%. If $1,250 in revenue is attributed under the selected window, ROAS is 0.5, or 50%.
Why one metric can point the wrong way
Pattern | Possible reading | What to inspect |
|---|---|---|
CTR up, IPM flat | More clicks did not become more attributed installs | Click quality, store step, attribution, audience |
CPI down, retention down | Cheaper installs may be lower quality | Cohort definitions, targeting, creative/game expectation gap |
IPM up, ROAS unknown | Early conversion improved but value is not mature | Revenue window and cohort age |
All early metrics flat | Treatment may be equivalent or test may be underpowered | Delivery, variance, effect size, data quality |
Playable interaction metrics are diagnostic
Starts, first actions, completions, fail states, and outbound clicks can locate friction inside the creative. They are not substitutes for installs or downstream quality. A design that maximizes completion could make the ad easier without improving the audience's interest in the actual game.
Attribution and privacy boundaries
Networks and measurement partners can use different windows, models, thresholds, and privacy-preserving reports. Compare treatments inside the same defined system where possible. Do not combine incompatible denominators or describe modeled attribution as direct user-level observation.
A decision-ready read
- Confirm creative IDs and test allocation.
- Freeze formulas, denominators, windows, and source reports.
- Read the primary metric with at least one downstream guardrail.
- Check delivery and product confounders.
- Record uncertainty and permit an inconclusive decision.
- Translate the result into a specific production action.
Put the formula and source for each metric into the creative test matrix before comparing treatments.
Sources
Put this into practice
- Creative testing framework
Design the comparison before reading the result.
- Creative test matrix
Record definitions, source, and next action.
Make your next test playable
Turn a creative question into a scoped playable, variant set, and delivery plan.
Start a playable brief