Attribute and fatigue guide

Learn how to separate creative fatigue from audience, budget, and attribution effects, then turn the diagnosis into a controlled next test.

Creative fatigue is a sustained loss of an asset’s ability to produce the target outcome after accounting for delivery, audience, placement, market, and reporting effects. A falling CTR can be one symptom, but it is not a complete diagnosis. The asset may be reaching a broader audience, moving into a weaker country mix, receiving more budget, or waiting for conversions to mature.

The practical question is not “Is CTR down?” It is “Did the creative itself become less effective, and what should the next controlled variant change?”

Which signals indicate real fatigue?

No universal threshold works across every app, network, objective, and spend level. Use a bundle of evidence:

Frequency or cumulative exposure increases.
CTR or install rate declines within a comparable audience and placement.
CPI or CPA rises after controlling for auction and market shifts.
Purchase rate, revenue per install, or ROAS weakens at the same maturity horizon.
The decline persists across multiple observation windows rather than one noisy day.
New variants with a controlled change outperform the original under comparable delivery.

The stronger the diagnosis, the more of these signals agree. A single metric should trigger investigation, not an automatic pause. Age on its own is not a signal at all: the refresh cadences that circulate as weeks are published without a dataset or a method, and a calendar cannot tell a decayed creative from one the engine is still finding new users for. Measure decline; do not infer it from age.

To turn investigation into a repeatable rule, use the creative fatigue measurement protocol: a guarded window comparison with volume floors, a defined decay threshold, and the checks that rule out delivery and reporting look-alikes.

Separate creative effects from delivery effects

Start by fixing the comparison scope. Compare the asset within the same app, platform, country, campaign objective, placement family, and attribution horizon. Then examine whether spend or audience expansion changed at the same time as performance.

Observed pattern	More likely explanation	Next check
CTR falls while downstream conversion is stable	Broader exposure or weaker immediate response	Frequency, reach, placement and audience mix
CTR is stable while CPA rises	Post-click or auction change	Install rate, event conversion, CPM and attribution
CPI rises but D30 ROAS holds	More expensive users with retained value	Revenue per install and payback
All outcome metrics weaken after repeated exposure	Possible fatigue	Comparable new variant and frequency bands
One country falls while others hold	Market-specific effect	Localized creative, auction and store-page behavior

This prevents a common error: replacing a good concept because campaign delivery changed around it. The country row is the one to run first. Split a creative’s change into the part its country weights moved and the part its country rates moved; a flag that disappears once the mix is held still was a delivery change, not a creative one.

Analyze attributes, not only files

An asset is a combination of observable decisions: opening hook, message, format, pace, visual composition, proof, offer, CTA, speaker, product visibility, and duration. File-level analysis tells you which finished ad won. Attribute analysis helps explain what to preserve when producing the next ad.

Use a controlled taxonomy. Each label should be interpretable by the creative team and applied consistently. Avoid vague tags such as “good hook” or “high quality”; they encode the analyst’s judgment instead of the creative fact.

Useful attributes include:

Hook type: problem, outcome, question, demonstration, testimonial.
Format: UGC, gameplay, product demonstration, animation, static.
First product appearance time.
On-screen text density.
Human presence and speaker role.
Proof mechanism: metric, review, demonstration, comparison.
CTA wording and timing.
Duration and scene-change pace.

How to design the next test

Change one decision or one coherent group of decisions. If the winner used a problem-led hook and visible product demonstration, a useful next test might retain the demonstration while changing only the hook. Rebuilding every scene, message, format, and CTA produces a new ad but teaches the team almost nothing.

Write the hypothesis before generation:

Keeping the demonstrated product outcome while replacing the overused opening hook will restore install rate without reducing D30 revenue per install.

Then define the success metric, minimum delivery, comparison scope, and stopping rule. The result should be recorded even when the hypothesis fails; failed tests narrow the next decision.

Use the creative test brief template to carry that diagnosis into a control, one changed decision, held constants, a primary metric, and predeclared win, loss, and inconclusive branches.

If an AI agent is running the analysis, the read-only MCP creative performance workflow preserves those scope and evidence checks before it writes the next brief.

Why outcome-aware attributes matter

Attributes associated with clicks can differ from attributes associated with purchases or long-term value. A fast, provocative hook may win CTR while attracting users who monetize poorly. A slower demonstration may acquire fewer users but improve payback.

Rank attribute patterns against the outcome that matters for the current business decision. Use CTR when the task is attention. Use purchase conversion, revenue, ROAS, retention, or predicted payback when the task is profitable acquisition.

Ranking is where most attribute work goes wrong, and the failure is arithmetic rather than judgment. A tag-level table counts impressions where it should count creatives, which makes almost every gap look certain. Before a row changes a brief it should pass acceptance rules: enough independent creatives on both sides, an interval whose near end still clears the decision threshold, and a correction for the number of comparisons the table ran.

The Lemon workflow

Lemon AI joins attribute labels to creative delivery and downstream outcomes. Teams can inspect which attributes occur in winners, compare performance across breakdowns, generate a controlled variant, and measure it in the same system.

Use Attribute Analysis to explore the pattern layer, Creative Generation to turn a hypothesis into variants, and Creative Analytics to evaluate the result. The methodology explains scope, aggregation, and evidence limitations.

Primary sources

On this page

Creative Fatigue Analysis for Mobile UA Teams