Frequency is a bidding problem, not a creative one

The standard response to a frequency chart that bends upward is to open a creative brief. The concept has worn out, the argument goes, so the fix is a new hook, a new opening three seconds, a fresh batch of variations. That reflex is expensive and, more often than not, wrong. On the self-optimising networks where most game spend now sits, the bidder decides frequency before any viewer sees the creative. The creative is downstream of the problem, not the cause of it.

The claim here is narrow but uncomfortable for creative teams: when frequency rises and the concept isn't old, the correct first move is to change how you are bidding, not what you are showing. Refreshing creative in that situation buys a few days of relief, destroys the comparison you needed, and leaves the cause untouched.

Why the bidder concentrates impressions

A performance network's model answers one question per impression: how likely is this device to complete the optimisation event, and what is that worth at the current bid? The users with the highest predicted probability win the impression, and as a campaign runs that pool shrinks, because the obvious converters have either installed or seen the ad and declined. The model doesn't respond by widening the net. It responds by returning to the devices whose predicted probability is still highest, which are frequently the same devices it already reached.

That is what a rising frequency curve shows: the model has exhausted the cheap part of its confidence distribution at your bid and event. The creative didn't get worse. The auction got narrower.

Three bidding inputs drive that narrowing. The optimisation event comes first: a deep event (a purchase, a level-30 completion) gives the model a small set of high-confidence users, so it recycles them harder, while a shallower event widens the pool at the cost of intent. Then the bid or target; a tight ROAS target or low CPI cap tells the model to stay in the safe zone, which means the users it already knows. Last, the audience definition. Geo and OS set the ceiling, together with any inclusion list, and if the ceiling is small, frequency climbs regardless of creative quality.

Network-level frequency caps do very little on SDK inventory because the cap applies per campaign and per network, and often per app, rather than per user. A user can hit your cap on three campaigns across two networks in an afternoon and none of the four systems involved will report it.

The second-order damage from a creative refresh

The trade-off that most coverage of creative fatigue misses is what a refresh does to your evidence. When you swap a concept out during a frequency spike, three things happen at once.

The new creative enters the network's exploration phase, which briefly forces the model to try devices it had stopped trying. Frequency drops. It looks like the creative fixed it. Within a week or two the model has learned the new asset, returns to the same high-confidence pool, and frequency climbs again. You now believe the second concept fatigued faster than the first, which is not true; the pool was already saturated when it arrived.

Meanwhile the retired concept gets logged as fatigued in your creative library. Six months later a strategist decides not to revive it because the data says it burned out. A bidding condition nobody recorded has contaminated that data.

And the refresh cost is real. Sensor Tower's State of Mobile work over the last two years has consistently pointed to creative volume rising across the top advertisers. Spending that production capacity on a bidding symptom is a poor use of it.

A diagnostic before you brief anything

Separating the two causes takes a week of data you already have. Read three series together by campaign: frequency, click-through rate at constant frequency, and reach growth.

If frequency is rising, CTR among first-exposure users is roughly flat, and reach has plateaued, the bidder is recycling. Creative is fine. Adjust the bid side.

If frequency is flat, CTR is falling at every exposure count including the first, and reach is still growing, the creative is genuinely losing its pull with fresh viewers. Brief the refresh.

If both are moving, fix bidding first, wait for the pool to reopen, then judge the creative on the users it has not met yet.

As an illustrative example, take a puzzle title with a single tier-one campaign whose average frequency has moved from about two to nearly five in three weeks. A creative-first team ships four new variations and sees frequency fall to three. Within twelve days it has climbed back. A bidding-first team instead loosens the ROAS target by a modest step and splits out a second campaign on a shallower event, after which frequency settles back near two and reach resumes growing, while the original creative keeps performing on the new users it is now allowed to meet. Those numbers are illustrative; the mechanism is not.

The bidding levers, in order

Work through these from cheapest to most disruptive:

• Move the optimisation event one step shallower on a copy of the campaign and let the two run side by side. Watch whether frequency diverges.
• Widen the target. A small relaxation in a ROAS or CPI target often reopens more inventory than any creative change.
• Add a second campaign with a different event or a different geo cluster rather than a bigger budget on the same one. Two models exploring separately reach more users than one model exploiting.
• Only then add creative, and add it as an exploration lever, not a fatigue cure.

Where the creative team still owns it

None of this argues that creative fatigue is a myth. Concepts do wear out; a strong hook eventually stops working with fresh viewers once a category imitates it, and that is a creative problem. It shows up as falling first-exposure CTR and deserves a real brief.

What the creative team should push back on is a frequency chart handed over as the brief. Ask for the first-exposure CTR series alongside it. If the buyer can't produce one, the request is premature, and the most useful thing the creative lead can do is send it back with a question about the optimisation event rather than a new storyboard.

Related archive reading

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

• Re-engaging lapsed players: a campaign design
• Why your best network degrades as you scale it