Every scaled playable campaign eventually decays. Delivery softens, cost per install drifts up, and the creative that carried the account for six weeks stops working. The mistake most teams make is treating that as a single event with a single cause, when in practice at least three different things produce the same shape in a chart — and only one of them is fixed by making a new creative.
What fatigue actually is
Fatigue is the audience, not the ad. The creative has not changed; the population that has already seen it has grown, and the people left in the addressable pool are the ones who did not respond the first time. Everything downstream follows from that: response rate falls, the network's model bids less confidently, delivery drops, and effective cost rises.
This matters because it tells you which fixes work. If the problem is that your reachable audience has seen the ad, then editing the CTA colour will not help — you need a creative different enough that the model treats it as a new proposition to a partly-overlapping audience.
The three things it gets confused with
An auction shift is the first: a competitor entering your placements at a higher bid raises your costs without anything about your creative changing. The tell is that the decline is abrupt rather than gradual, and it affects all your creatives at once rather than the oldest one first.
A seasonality effect is the second, and it looks like fatigue on a two-week view and like weather on a twelve-week one. The third is a targeting or budget change someone made and did not write down — which is far more common than anyone admits, and is why a change log is worth more to creative diagnosis than most dashboards.
The diagnostic that separates them
Launch an untouched copy of the same creative as a new asset. If it recovers, you were looking at frequency and asset-level decay, not the creative concept dying. If it performs the same as the original, the audience has genuinely seen this idea and a re-upload will not save it.
The signals that move before CPI does
By the time cost per install has moved, you have already spent money learning something the earlier signals would have told you. Three things move first.
Engagement rate is the earliest: the share of impressions that produce any interaction at all. It falls before installs do because it measures the creative's ability to earn attention, which is the first thing to erode. Completion rate follows — the share of people who reach the end card. And frequency, if your network reports it, is the direct measure of the underlying cause.
Watch those three as a set. Engagement falling while completion holds means the hook is tired but the experience still works, and a new opening may be enough. Both falling together means the concept is spent.
What to change, in order of leverage
The instinct is to change the CTA or the colours, because those are cheap. They are also the least effective, for the same reason: a change the audience does not register is not a new creative to them or to the network's model.
- The mechanic. A different interaction is genuinely a different ad and resets the clock properly.
- The first three seconds. Same game, different opening — often enough when engagement fell but completion held.
- The end card structure. Cheapest meaningful change, and it only affects people who got that far.
- Copy and palette. Real but small; use these to extend a winner, not to revive a dead one.
The ordering is deliberate. Teams work up this list from the bottom because the bottom is cheap, and then conclude that creative refresh does not work — when what did not work was changing something nobody noticed.
A refresh cadence that is affordable
The honest answer to "how often should we refresh" is: as often as your data says, which is a different number for every account. What is generalisable is the shape. Run one concept as the control for as long as it holds. Introduce one genuinely different concept on a small share of budget continuously, so there is always a candidate in flight rather than a scramble when the incumbent fails. When a challenger beats the control, promote it and start a new challenger.
That pattern costs one new concept per cycle rather than a batch of variants, and it means you never discover fatigue and start production on the same day. The constraint that makes it possible is production speed — a team that needs two weeks and an engineer for a new playable cannot run a continuous challenger, which is the practical argument for keeping creative production in-house.
Variants are not the same as concepts
A common failure is running five variants of one idea and calling it a creative pipeline. To the auction those five are largely the same proposition, and they fatigue together. Give the model genuinely different ideas to choose between — different mechanic, different promise, different first frame — and you get both better selection and a slower decay curve, because the audience for one is not fully the audience for another.
When to retire rather than refresh
Keep a creative running at low spend after it stops leading. Two things make that worth the budget: it is a control against which the next challenger is measured, and creatives that have decayed in one market frequently have life left in another where the audience has not seen them. Retiring an asset entirely because it lost in your largest geography discards inventory you already paid to produce.
Geography buys you time
Fatigue is a property of the audience that has seen the ad, which makes it geographic. A creative exhausted in your largest market frequently has months of life left in one it has never run in, and the production cost has already been paid. Teams that treat retirement as global throw away inventory they own.
The practical version is to stagger launches rather than going wide at once. It costs nothing, it extends the useful life of every winner, and it has the side benefit of giving you a clean read on whether a decline is creative fatigue or something happening in one market's auction.
Create playable end cards in minutes—no code required.
Open Playable Ads Maker