Blended ROAS Calculation: Where It Quietly Goes Wrong

A blended ROAS number is meant to be a summary, one figure that tells a UA lead whether spend is working across a portfolio of campaigns, markets and monetisation types. It is also one of the easiest numbers in a UA dashboard to get quietly wrong, because a blend is only as honest as the components feeding into it, and several of the most common ways those components go wrong produce a number that looks perfectly healthy right up until it does not.

Currency normalisation hides more than it reveals

Any portfolio spanning more than one market is mixing currencies before it mixes anything else, and the exchange rate used to normalise that mix matters more than most dashboards let on. A blended ROAS calculated with month-end spot rates will drift from one calculated with a rolling average, sometimes by enough to change whether a market looks profitable. The practical failure mode is a dashboard that recalculates historical figures every time it refreshes exchange rates, silently rewriting last month's reported ROAS without flagging that anything changed. A team that does not fix its normalisation method and lock historical figures against it cannot trust a month-over-month comparison, because part of the apparent movement is currency noise rather than campaign performance.

Cohort windows rarely line up across sources

IAP revenue and ad revenue are usually measured on different attribution windows by default, IAP often on a 7-day or 30-day cohort window and ad revenue frequently on whatever window a mediation platform defaults to, which is not always the same setting. Blending the two without reconciling the windows produces a number that mixes a mature IAP cohort with an immature ad-revenue cohort, understating the blended figure for a title where ad revenue accrues more slowly, or overstating it for a title where ad revenue is front-loaded through interstitials shown early in a session. The fix is not to force both revenue types onto an identical window, since that is rarely appropriate given how differently they accrue. It is to record which window each component uses, alongside the blended figure, so anyone reading the dashboard knows what is and is not comparable.

Ad-revenue attribution is the least standardised piece

Of the two revenue types feeding a blended ROAS for a hybrid-monetisation title, ad revenue attribution is by far the less standardised. Impression-level ad revenue reporting varies by mediation platform, some report predicted eCPM at the time of the impression, others reconcile to actual network payout later, and the gap between those two figures can be substantial for a title running several ad networks through one mediation layer. A blended ROAS that uses predicted eCPM will look better earlier and get quietly revised downward as actual payouts reconcile, which is a specific and common way a headline number looks strong in a weekly report and softer by the time a monthly report closes the same cohort.

A practical audit checklist

A blended ROAS figure is trustworthy only after it survives a few basic checks, ideally run before it reaches a budget conversation rather than after a decision has already been made on it:

• Confirm whether the dashboard uses a fixed or rolling exchange rate for currency normalisation, and whether historical figures are locked once reported or silently recalculated.
• Check that IAP and ad-revenue cohort windows are documented next to the blended figure, even if they differ, rather than presented as though they were measured identically.
• Ask whether ad revenue is reported as predicted eCPM or reconciled actual payout, and if predicted, how large the typical gap to actual has been over recent cohorts.
• Compare the blended figure against each component reported separately at least once per reporting cycle, since a blend that looks fine can still be hiding one weak component propped up by a strong one.
• Re-run the same blended figure a full cohort maturation period after it was first reported, and record the drift, building a track record of how much a first-look number typically moves once ad revenue reconciles.

A worked example of the drift

Consider a hypothetical mid-market puzzle title reporting a blended day-30 ROAS of 105% in its weekly dashboard, using predicted eCPM for ad revenue and a 30-day IAP window against a 30-day ad-revenue window that the mediation platform actually calculates on a rolling 21-day basis by default. Once the ad-revenue component reconciles to actual network payout roughly six weeks later, the same cohort's blended figure restates to 96%, a swing large enough to change whether that cohort would have cleared a target payback threshold. Nothing fraudulent happened here. The team was reading a number that was internally consistent with its own dashboard defaults but not with the assumptions a budget decision actually needed, and the gap between 105% and 96% was sitting in the reconciliation lag the whole time, invisible until someone went looking for it.

What a clean blended ROAS calculation actually buys a team

None of this argues against blending IAP and ad revenue into one ROAS figure. A single number is genuinely useful for a fast read across a portfolio, and building a separate dashboard for every revenue type and market combination would be its own kind of failure, trading a fast decision tool for an unusable pile of detail. The point of the audit above is narrower: know exactly what assumptions are baked into the blend before treating it as a decision input, and revisit those assumptions on a schedule rather than only after a budget decision made on a flawed number turns out to have been wrong. As UA Ledger argued in Attribution Windows for Hybrid-Monetisation Games, the underlying discipline is the same in both cases: a dashboard number is only as reliable as the reconciliation work a team is willing to do before trusting it.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• D7 ROAS targets punish long-retention games
• The creative testing statistics most teams get wrong