Creative ROAS: Can app revenue be tied to one creative?

Creative ROAS explained: the formula, where MMPs tie app revenue to one creative, where Google, Meta, TikTok, and SKAN model it, and how to approximate it.

Two hands hold a phone playing a city-building game ad of a miniature town with one coral building, on a rooftop terrace at blue hour.
On this page
  1. Where creative ROAS can be measured directly
  2. Where the creative split is modeled or missing
  3. Measured vs modeled creative ROAS
  4. How to approximate creative ROAS where it is missing
  5. Is modeled creative ROAS good enough to cut a creative?
  6. Example: Creative ROAS review for a puzzle game
  7. Creative performance with Starti AI Studio
  8. Frequently asked questions

App revenue can be tied to one creative when the install reaches the measurement partner with an ad or creative ID attached. In multi-asset ad formats and for iOS traffic without tracking consent, the split by creative is modeled or missing. The metric itself, creative ROAS, is the revenue earned from the users one ad creative acquired, divided by the spend on that creative.

The calculation for one creative and one cohort day:

Creative ROAS (day N) = cohort revenue through day N from installs credited to the creative ÷ spend on the creative × 100

AppsFlyer's Creative Optimization dashboard reports it this way, with cohorted revenue and ROAS from day 0 to day 90 and the formula "Revenue / Cost * 100."

Where creative ROAS can be measured directly

Creative ROAS is measured when the creative's identity travels with the install:

  • Ad-level link parameters: on networks that use attribution links, AppsFlyer link parameters carry an ad name and ad ID that the advertiser or publisher fills in, and Adjust links have an optional creative level that Adjust's link guide offers "to gather insights about which creative works best for your campaigns."
  • Deterministic installs: Android installs and iOS installs from users who allow tracking keep those parameters on link-based traffic. Adjust's 2026 report puts the ATT opt-in rate, measured among iOS users who permit tracking requests, at 38% in Q1 2026 and 39% for gaming, so most iOS users on that base do not opt in.
  • One creative per ad: when an ad runs a single video, its ad ID identifies the creative, and AppsFlyer matches the same video across ads and networks into one creative record.

Where the creative split is modeled or missing

Several common formats report above the level of a single creative:

  • Multi-asset ads: for Meta Advantage+ and Google App campaigns, AppsFlyer reports that Meta provides data up to the ad level and Google up to the ad group level, so it splits installs, revenue, and ROAS across assets by the network's install ratio. For Meta Flexible Ads, its partner notes add that only the best-performing asset of an ad is shown.
  • Google App campaigns: Singular's guide states that "Only network data is available at asset level (there is no tracker data)," with no revenue at that level. Since January 7, 2026, Singular adds an estimated creative breakdown of installs and revenue for Google Ads on top of that network data.
  • Meta: Adjust's Meta reporting page notes that without Meta's Advanced Mobile Measurement terms signed, the campaign, ad group, and creative fields come back empty and Meta-attributed engagements show as unattributed.
  • TikTok: for the upgraded Smart+ campaign, ad-level names and data do not reach MMPs, according to TikTok's reporting guide, and the creative maps to the MMP's ad name.
  • SKAN and AdAttributionKit: Apple's source identifier holds up to four digits in SKAdNetwork 4 and AdAttributionKit, and "You can encode information about your advertisement in each set of digits." Each install gets a postback data tier that depends on crowd size; in Tiers 0 and 1 the first postback keeps two digits, so detail encoded in the third and fourth digits drops out. Conversion data arrives in three windows that end on day 2, day 7, and day 35, with the fine value only in the first.

AppsFlyer divides impressions, clicks, installs, and cost across assets in proportion to the network's asset-level metrics, and network asset numbers do not add up the way campaign numbers do. Google's asset reporting note warns that "Asset-level reports will often show higher metrics because they count each individual component," that they may show served rather than billed cost, and that teams should "rely on campaign-level totals."

Measured vs modeled creative ROAS

Four numbers appear in creative reports, and each answers a different question:

  • Measured creative ROAS: cohort revenue from installs carrying the creative's ID ÷ its spend. Reliable where the ID survives, on link-based Android and opted-in iOS traffic.
  • Modeled creative ROAS: campaign revenue split across assets by the network's install ratio, so each asset in an ad gets the same revenue per install.
  • Network-reported value by asset: the conversion value an ad platform credits to an asset, such as Meta's action values for Dynamic Creative. It counts what the platform attributes, under its own windows.
  • Campaign ROAS: total attributed revenue ÷ total spend, the reference point for every creative-level figure.

Upstream metrics such as IPM and hook rate show why a creative wins the install; creative ROAS shows whether the install paid. Cohort rules for reading revenue by day are covered in D7 ROAS.

How to approximate creative ROAS where it is missing

Teams that cannot measure every creative directly combine five practices:

  1. Clean test cells: run creative tests with one creative per ad, so the ad ID carries the creative into the measurement partner, and read measured ROAS from Android and opted-in iOS cohorts.
  2. iOS cells: on iOS, give each test creative its own campaign or source identifier value, and use the network's ad-level SKAN reporting where it exists, which TikTok's iOS reporting guide suggests "to understand creative-level performance."
  3. Labeled modeling: use the measurement partner's asset split for multi-asset campaigns, and label it as modeled in every report.
  4. Attribute tags: tag each creative by hook, format, and offer, and compare the ROAS of attribute groups, which carry more volume than single creatives.
  5. Volume floor: read a creative's day-7 ROAS once its cohort holds enough payers to be stable, and use day 0 to day 3 purchase events as the early signal before then.

Android payers and iOS payers can behave differently, so a creative that wins on Android gets its own iOS read before it scales there.

Is modeled creative ROAS good enough to cut a creative?

It is good enough to pick candidates for a cut. Since a modeled split gives every asset in an ad the same revenue per install, it says nothing about whether one creative brings better payers than another. AppsFlyer's worked example shows the mechanics: with the network reporting 60 and 40 installs for two assets and 50 installs attributed to the ad, the assets get 30 and 20, and "Down funnel metrics, such as Revenue and ROAS, are calculated based on this ratio as well." The final call needs a measured read: before cutting a creative that looks weak in a modeled split, run it in a clean cell against the current best creative and compare measured ROAS on the same cohort day.

Example: Creative ROAS review for a puzzle game

The review:

  • Setup: three videos in one Meta Advantage+ campaign, plus the same three in single-creative ad sets on Android at a smaller budget, each cell run until its day-7 cohort holds at least 200 payers.
  • Modeled split: the measurement partner's install ratio gives video A 52% of the campaign's day-7 revenue, B 31%, and C 17%.
  • Measured cells: on Android, D7 ROAS reads 11% for A, 16% for B, and 7% for C.
  • Reading: A wins delivery inside the campaign, while B's installs pay more where revenue is measured per creative; C trails on both.
  • Decision: C is retired, B gets a larger share of the test budget and two new openings, and A stays in the campaign while its iOS read comes in.

Creative performance with Starti AI Studio

Smart Insight follows one creative across platforms and campaigns, with its spend, engagement, and video completion in the same record, for Meta, TikTok, Google Ads, and configured custom networks. It compares nine creative attributes against CTR, completion, and other outcomes, and teams can turn the findings into a brief for their creative team or continue into production in AI Studio.

Book a demo to plan a creative test.

Frequently asked questions

What is creative ROAS?

Creative ROAS is the revenue from users acquired by one ad creative, over a set cohort window, divided by the spend on that creative, times 100.

Can an MMP report ROAS by creative?

Yes, where the install reaches the MMP with an ad or creative ID, as with ad-level link parameters on Android and opted-in iOS traffic. For multi-asset formats such as Meta Advantage+ and Google App campaigns, MMPs split revenue across assets by the network's install ratio.

Why don't Google App campaigns show revenue by asset?

Google provides App campaign data up to the ad group level, and its asset reports show network metrics such as installs and in-app actions. Measurement partners report at the ad group level or estimate the creative split.

Can SKAdNetwork attribute an install to a creative?

It can when the ad network encodes the creative in the hierarchical source identifier and the postback data tier returns enough digits. In Tiers 0 and 1 the first postback keeps two digits, which leaves 100 possible values.

How do you approximate creative ROAS?

Run tests with one creative per ad, read measured ROAS from Android and opted-in iOS cohorts, give iOS test creatives their own campaign or source identifier, use the measurement partner's modeled split for multi-asset campaigns, and compare ROAS by creative attribute.

Starti AI Studio

Discuss your next campaign with Starti.

Book a Demo
Hands pack ski mittens and goggles into an open backpack while a television behind a low partition plays a ski resort ad featuring a distant skier in a coral jacket.

TV attribution: How it works for linear and streaming TV

TV attribution explained: how linear and streaming TV ads get credit for visits and sales, what OTT attribution covers, and how TV attribution windows work.

A folded gray stroller stands in a sunlit entryway while a television behind a low plaster partition shows a stroller with a coral canopy on a park path.

Programmatic TV advertising: How it works and how to buy it

Programmatic TV advertising explained: how one streaming TV impression is bought, how linear TV differs, programmatic vs direct buying, and how to start.

A hand selects a sailing shot from a row of hanging footage stills, in navy halftone with a coral rectangle behind the chosen print.

Video asset management: How ad teams find and reuse footage

Video asset management for ad teams: why video is harder than images, indexing by shot, reusing footage, the workflow to publishing, and what software needs.