Anatomy of a creative that scaled, then died

When a winning creative stops working, the word that appears in the review meeting is fatigue. The remedy that follows is a refresh: new colours, a re-cut hook, a different voice line, the same underlying promise. Three weeks later the refresh has also died, and the team concludes, with some relief, that the concept has nothing left in it.

Fatigue is the wrong diagnosis in most of these cases. A scaled creative dies because the delivery algorithm found a specific pocket of users who respond to a specific claim, then extracted that pocket until it ran out. Nothing about the visual surface caused the decline, so nothing about the visual surface will reverse it. The useful question isn't how to make the ad fresh again but which claim to make to which pocket after this one, and that requires an autopsy the fatigue label lets teams skip.

How the pocket forms

Modern delivery systems do not show a creative to an audience, whether the system belongs to Meta or Google or AppLovin or one of the other major networks. They search for the users most likely to complete the optimisation event and concentrate spend where the response is strongest, which means a creative that scales is simply a creative for which the algorithm found a dense pocket quickly.

The claim defines the pocket, not the format. A merge game ad that leads on decorating a ruined mansion finds people who respond to restoration and progress, while the same game advertised on a puzzle-fail hook finds people who respond to competence and correction. These are different people. The algorithm doesn't know or care why; it only knows where conversions are cheap.

Once the algorithm has found the pocket and spend has concentrated there, two counters start running: frequency within the pocket climbs, and the share of impressions reaching users the campaign has never touched falls. Both erode conversion rate, but they erode it for different reasons, and only one of them is fatigue in the strict sense.

Three autopsies, not one

Before anyone commissions a refresh, run three checks on the dying creative. Most analytics stacks and platform reports can support at least two.

The reach curve first: what share of the last fortnight's impressions went to first-time viewers, and how has that share moved since the peak. If new reach is collapsing while total impressions hold, the pocket has run dry.

Then the frequency curve, meaning average and top-decile frequency in the pocket. If frequency is high and rising, some of the decline is genuine repetition wear and a refresh may buy time.

Then the cohort quality curve: D1, D7 and early payer rates for installs from this creative, by week. If quality has held while CPI rose, the algorithm is still finding the right people and just paying more for them; if quality has fallen, it has started reaching outside the pocket to hit volume.

The pattern that appears most often in a scaled winner is the first and third together: new reach down, quality down, frequency only modestly up. That is a pocket exhausted and an algorithm improvising. A visual refresh keeps the same claim and therefore hunts the same drained pocket. It performs for a few days because the platform treats it as a new asset and gives it exploratory delivery, then it converges on the same users and dies faster than the original.

The second-order damage

The part of this story that gets little coverage is what the winner did to the rest of the portfolio while it was alive. A creative that scales attracts budget. In campaign structures that let the platform allocate across assets, the winner takes the majority of spend and the other concepts starve. They never accumulate enough delivery to find their own pockets, so they look like losers, and they get retired on that evidence.

When the winner dies, the bench is empty. The team is now iterating on a dead claim under time pressure, which is the worst condition for creative work. This is a structural consequence of how allocation works, not a failure of the creative team, and the mitigation has to be structural too: a protected minimum delivery for concepts that make a different claim, or separate campaigns for genuinely different angles so the platform cannot collapse them into one.

There is a second, quieter effect. Months of concentrated spend on one claim teaches the platform's model that your app is for the pocket that responded to it. When you launch a different claim, the model's initial guesses about who to show it to lean towards the old pocket. Early results for the new angle look weak for reasons unrelated to the angle. Give a genuinely different claim more exploratory budget and more days than you would give a variant of the old one.

Refresh, reframe or retire

A decision rule, using the three autopsies.

Frequency high, new reach stable, quality stable: refresh, meaning the same claim on a new surface, and expect it to buy weeks rather than months.

New reach falling, quality falling, frequency modest: reframe. That means a new claim aimed at a different motivation and launched with protected budget, and it means not measuring it against the peak CPI of the dead winner.

New reach falling, quality falling, and the two previous reframes also died inside a fortnight: retire the concept family and accept that the pocket for this game's current positioning is close to fully addressed on this channel. Money does more on a new channel or a new store positioning than on another cut.

An illustrative example, with numbers for arithmetic only. A hidden-object creative peaked at a 1.8 unit CPI with first-time viewers making up around 70 percent of impressions. Six weeks later CPI is 3.1, first-time share is under 30 percent, D7 retention of its installs has fallen by roughly a fifth, and average frequency has moved from 2 to 3. Reframe. The refresh that was already in production should ship because it's nearly free, but it shouldn't be the only thing shipping, and nobody should use its performance to judge whether the concept family has life left.

Earlier in the year "Creative Variant Retirement Rules for a Live Portfolio" laid out when to switch a variant off. The point here sits one level up: the retirement decision is only as good as the diagnosis of why the creative is dying, and the fatigue label is a way of not making one.

The last thing to check after any scaled winner dies is whether the store listing and the first session still deliver the claim it made. A creative that finds a pocket also raises expectations in that pocket, and if the reframe promises something different, the onboarding needs to know.

Related archive reading

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

• Frequency is a bidding problem, not a creative one
• Most UA best practice is survivorship