Every UA team has had this meeting. Someone shares a creative report, points at the ad with the lowest CPI and the highest CTR, and declares it the winner. Two weeks later the cohort it brought in has the retention curve of a firework. A brief flash of activity before disappearing into the night, and nobody can say for sure whether that ad was the problem, because the revenue data lives in a different tool, at a different level of detail, under a different definition of “install”.
Creative-level attribution exists to end that meeting. It is the practice of connecting each individual ad creative to what the players it acquired actually did after installing. Whether they finished the tutorial, came back on day 7, bought something, watched ads, or churned before lunch. For mobile games, where creatives drive most of the performance difference between campaigns, that connection is often the most useful piece of measurement you can add.
TL;DR
Creative-level attribution links individual ad creatives to downstream player value such as retention, IAP revenue, ad revenue and LTV, rather than stopping at CPI, CTR and installs. It works by carrying a creative identifier from the ad click through to the install and into your game’s event data, then crediting each player’s behavior back to the creative that acquired them. It is not a replacement for your MMP or for network reporting. It answers a narrower and more practical question: which ads bring in players worth paying for? To use it well, judge creatives on value per spend rather than cost per install, group results by concept rather than by single asset, respect sample sizes, and feed what you learn back into both your creative briefs and the signals you send to ad networks.
Creative-level attribution defined
Creative-level attribution is a measurement approach that assigns each acquired player to the specific ad creative that drove the install. It then tracks that player’s in-game behavior and revenue so you can compare creatives by the value of the players they brought in. The output is not just “Creative A got 4,000 installs at $1.90”. It is closer to “Creative A got 4,000 installs at $1.90, 31% of those players completed level 10, D7 retention was 14% and D7 ROAS was 0.21x, while Creative B cost more per install but returned more per dollar spent”.
The key idea is simple and slightly annoying: a cheap install is not necessarily a valuable install. Creative-level attribution makes that visible at the level where your team actually makes decisions, because creative teams brief, produce and kill ads one concept at a time, not one campaign at a time.
Why ad-level metrics in Ads Manager are not the same thing
A reasonable objection is that Meta, TikTok and Google already show performance by ad. They do, and that reporting is useful. However, it has a few built-in limits that matter for mobile games.
First, networks can only report on the events you send them. If your game tracks installs and purchases but ignores tutorial completion, progression and ad revenue, the network misses much of what makes players valuable in hybrid-monetized games.
Second, each network uses different attribution windows and reporting methods. They often blend observed and modeled conversions, making Meta and TikTok ad-level numbers difficult to compare directly.
Third, network reporting usually focuses on a short post-install window. However, the question that matters most, “did these players pay back?”, often takes weeks to answer.
On iOS there is an additional wrinkle, which we covered in more detail in our earlier article on how iOS attribution works. Apple’s privacy frameworks, SKAdNetwork and AdAttributionKit, report installs at the level of a campaign source identifier with limited, thresholded conversion data. They were designed to answer “did this campaign work, roughly?” and not “which video found my best payers?”. Your MMP, meanwhile, is excellent at cross-channel governance and campaign reporting, but on iOS it is working with the same privacy constraints as everyone else.
Creative-level attribution fills the gap between these views. It does not claim to be the one true number. It provides a consistent, creative-granular view of downstream player value that you can use to make creative decisions across networks.
How creative-level attribution works for mobile games
The exact implementation varies by platform, network and setup, but the logic always follows the same chain: creative, then acquisition, then player behavior, then value. Here is what happens at each step.
Step 1: capture creative context at the moment of acquisition
Everything starts with knowing which creative brought a player into your game. This information must be captured between the ad interaction and the first session.
Depending on your setup, there are several ways to do this. Click parameters or URL macros can capture campaign, ad set and ad identifiers. Web-to-app flows can record creative context before sending users to the store. You can also use consented device-level matching or data from networks about their attributed installs.
Whatever the source, you need a creative identifier, sometimes called a creative token. This identifier should follow the player into the game.
Consistent naming matters more than you might think. Imagine uploading “final_v3_NEW_actuallyfinal.mp4” across three networks under different ad names. Your attribution will report three separate creatives, even though they’re actually the same asset.
Step 2: connect the install to that context in privacy-compliant ways
Next, you need to link each install and first session to its captured creative context. Privacy rules play an important role in how this works.
On iOS, fingerprinting cannot reconstruct user identities without consent. Instead, attribution must rely on legitimate mechanisms. These include consented identifiers, first-party web-to-app flows, or deterministic context passed through approved channels. A good creative-level setup clearly identifies the mechanism behind each attribution. This helps you understand how much confidence to place in the results.
Coverage will rarely reach 100%, either. Some installs simply won’t carry any creative context. That’s fine, provided you track the coverage rate. You should also check whether attributed installs represent the wider player base.
Step 3: decide which creative gets the credit
Players rarely see just one ad. Someone might scroll past a puzzle-fail video on Monday, click a satisfying-merge ad on Wednesday and install on Thursday. Creative-level attribution needs a rule for who gets the credit. The most common default is last-click creative attribution, where the creative the player last clicked before installing gets the credit, with optional supporting rules for earlier clicks or views.
No credit rule is perfect. Last-click undervalues creatives that create awareness, while view-heavy models can hand credit to ads that were merely on screen. The practical advice is to pick a rule, document it, keep it stable while you compare creatives, and treat changes to the rule as a change to the measurement, not as a performance swing.
Step 4: track what players do after install
Once a player is linked to a creative, the game’s own events provide the rest of the picture. An SDK or server-to-server integration tracks the behaviors that matter. These include tutorial completion, level milestones, session frequency, D1 and D7 retention, IAP revenue, rewarded ad views and ad revenue. You can also track custom events that reflect your game’s economy.
User properties add another layer of context. For example, they can show whether a player mainly spends money or watches ads. This helps you understand not just which creatives convert, but what types of players they attract.
Step 5: aggregate and compare creatives by value
Finally, behavior and revenue roll up to the creative level. Now you can compare creatives on metrics such as cost per tutorial completion, cost per D7 retained player, D7 ROAS and projected LTV per dollar spent, side by side with the CPI and CTR you already had. The magic is not in any single metric. It is in seeing the top of the funnel and the bottom of the funnel on the same row.
A mobile game example: the cheap ad and the good ad
Imagine a merge-puzzle game running two creative concepts on Meta. Concept A is a “fail” video where a hand makes painfully wrong moves and the viewer is invited to do better. Concept B shows real gameplay with a cozy decoration meta layer, slower and less clickbaity.
On network metrics, Concept A wins easily. It has a higher CTR and a CPI around 30% lower. A team looking only at ad-level installs would shift budget toward it, and the algorithm would happily agree, because cheap installs are exactly what it was told to find.
Creative-level attribution tells a different story. Players from Concept A finish the tutorial at a similar rate but drop off sharply after day 1, because the game they installed is not the game the ad promised. Players from Concept B retain better through day 7 and are noticeably more likely to make a first purchase. Once you calculate D7 ROAS per creative, Concept B returns more revenue per dollar despite costing more per install.
The numbers here are illustrative, but the pattern is one every UA manager has seen in the wild. Without creative-level data, the team scales the wrong ad with full confidence. With it, the team scales Concept B, briefs the creative team on why it works, and maybe tests whether the hook from Concept A can be grafted onto honest gameplay. That last part is where creative-level attribution starts creating knowledge instead of just picking winners.
What creative-level attribution lets you do differently
Judge creatives on value per spend, not cost per install
Replace “lowest CPI wins” with a value-based primary metric that fits your monetization model. For an IAP-heavy RPG that might be D7 ROAS or cost per first purchaser. For a hybrid-casual game it might be blended D7 revenue including ad revenue. Keep CPI and CTR on the report, because they still explain delivery and scale, but stop letting them crown winners on their own.
Analyze concepts, not just individual assets
Individual creatives often have too few installs to produce reliable downstream numbers, especially for payer metrics. Tag each asset with its concept, hook, format and visual style, and roll results up to those groups. You will learn far more from “satisfying-merge hooks bring in better retainers than fail hooks across eleven variations” than from any single video’s report.
Send better signals back to the networks
Creative-level data also shows you which early behaviors predict value for players from different creatives. That is useful input for signal engineering, meaning the events and values you send to Meta, TikTok and Google so their algorithms optimize for players who matter rather than for installs. If players who complete level 10 on day 0 are far more likely to pay, that milestone may be a better optimization signal than a raw install, provided it fires often enough and early enough for the algorithm to learn from it.
Close the loop between UA and creative production
When creative briefs are based on what brought in valuable players rather than what got clicked, the creative team stops optimizing for thumb-stopping at any cost. That also tends to improve the match between ad and store page, since the ad promise becomes closer to the actual game, which is good for conversion and for retention.
Common pitfalls to avoid
Calling winners too early. Downstream metrics need volume and time. Day-0 signals arrive fast, but D7 ROAS by definition takes seven days plus the time to collect enough payers. Use early proxies to filter out obvious losers and mature cohort metrics to confirm winners.
Ignoring confidence. A creative with 40 installs and two payers is not a whale magnet. It is a coin flip with a thumbnail. Show install counts and confidence next to every value metric, and set minimum thresholds before anyone is allowed to declare a winner.
Comparing across mismatched windows and sources. If one creative’s numbers come from a different attribution rule or coverage level than another’s, the comparison is unfair. Keep definitions consistent and visible.
Forgetting audience and placement effects. A creative that wins in the US on Reels may lose in Germany on TikTok. Before you blame or praise the creative, check whether the difference is actually about geo, placement, audience or bid strategy.
Treating it as an MMP replacement. Creative-level attribution answers a specific question. Your MMP still matters for cross-channel governance, finance reporting and broader measurement. The two work best side by side, with clear agreement on which source is used for which decision.
Where Audiencelab fits
Building creative-level attribution in-house is possible, but it means capturing creative context reliably across networks, handling privacy-compliant matching, maintaining credit rules, instrumenting in-game events and keeping all of it consistent while campaigns and creatives change every week. That is a lot of plumbing for a UA team that would rather be making decisions.
Audiencelab by Geeklab is built for exactly this job. It is a creative-level framework for attribution and analytics on iOS that works alongside your MMP rather than replacing it. Through its SDKs (including Unity) or server-to-server integration, it carries creative context into the game, credits players to creatives with configurable rules, and connects them to retention, IAP revenue and ad revenue. It can also deliver value signals back to networks such as Meta and TikTok, so the algorithm learns from better information than “someone installed”. In one public case, Talofa Games reported CPI falling from about $3.80 to $1.80 while maintaining user quality. Results depend on the game, signals and execution, so treat that as an example of what is possible rather than a promise.
The takeaway
Creative-level attribution answers the question that standard UA reporting leaves hanging: which ads actually bring in players worth paying for? It works by carrying creative context from the ad into the game, crediting each player to a creative under a clear rule, and rolling post-install behavior and revenue back up to the creative and concept level.
The practical shift is small to describe and big in effect. Stop crowning creatives by CPI alone, compare them on value per dollar spent, group results by concept so the learning outlives any single asset, and use what you learn to brief better creatives and send better signals to the networks. Do that, and the next creative review meeting will be shorter, calmer and considerably less likely to end with scaling the mayfly.
Want to get started with creative-level attribution and easily customizable signal engineering? Let’s talk!
