How to Refresh Creative When Frequency and CPM Climb

Pull ad-level frequency, hook rate, outbound CTR, hold, CPA, and CPM on cold traffic only. Fatigue is a winner that decayed: hook and CTR were healthy, then they slid as frequency entered 2.5–3.5. A bad offer is weak from the first 48–72 hours at low frequency. Do not kill at 1.8 because a blog scared you, and do not wait for frequency 7 because last week's leftover conversions still make CPA look polite.

The stack you want: frequency up, hook rolling off 30%+ toward the weak <20% band, CTR off a 1%+ cold floor, CPM up, CPA lagging those moves. Hook and CTR usually slip first because CPA is slower. Hold falling with hook means even people who used to stay are bouncing. CPA up alone with a stable hook can be auction, season, or tracking — a different meeting. Split cold versus retargeting. Warm pools run hotter by design; copying 2.5–3.5 onto a tiny custom audience will make you refresh ads that are doing their reminder job.

Confirm with a new-versus-seen cut if the platform gives you one, or with a simple time series. If new people still hook at 30%+ and only high-frequency buckets are dead, you have fatigue in a pocket, not a dead product. If three recent hooks and two avatars all died at low frequency, stop calling it fatigue and inspect offer, page, and audience size. Small geos and tiny interest stacks climb the band fast — that is an audience ceiling, and a new hook will only buy you days. Write the diagnosis in one sentence before you produce anything.

Change one layer at a time, in order: (1) new hook line and first shot, (2) new avatar, (3) new proof or B-roll, (4) new body script, (5) new offer. Most teams jump to four or five because it feels like work. If hold is still in the 12–20% band and the body recently converted, you need a new open, not a new product story or a new landing page. Write the rung on the brief before anyone opens an editor.

Match the surgery to the injury. Hook under ~20% with decent hold on the people who stay: new first two seconds. Hold well under 12% with a strong hook: the open over-promised, fix the middle. Proof looks repetitive in comments or you are in a demo-led category: new B-roll after hook and face, not before. Offer last, because a new discount will mask whether the open was the problem. Swapping music, colour grade, or caption font is not a rung on this ladder. Those are background. Viewers tire of the face, the first line, and the first visual pattern.

The production shape on rungs 1–2 is still 6 creatives: 3 hooks × 2 avatars on the locked body. If casting is settled, 3 hooks on the winning face is enough. Do not ship twelve synonym opens. Do not ship one precious new film. Three different formulas (question, confession, pattern interrupt) give the auction something to do. Write the rung on the brief so a well-meaning designer does not 'also refresh the CTA' and contaminate the 48–72 hour read. One variable per climb is how the next 48–72 hours stay readable.

Queue the next 3-hook batch while CPA is still acceptable, not after it doubles. Render the grid, QC the first two seconds muted, and name files as refresh plus hook plus avatar. The tired winner stays live at a lower share of spend until the new cells have had 48–72 hours or ~1,000 impressions. Hard-killing the only converter the morning you launch replacements is how you create a CPA hole you then blame on the new creative.

Overlap is cheaper than a gap. The old ad has comments, social proof, and learning; the refresh should take over, not win a day-one knife fight. If the old ad is actually toxic — policy, collapsing hook under ~20%, comment pile of 'I've seen this' — kill it. Otherwise throttle, do not delete. Keep it out of the same desperate budget spike as the new grid if that would starve the test. A clean pattern: test campaign gets the new 3 × 2; scaling campaign keeps the graduate at a modest budget until the readout.

Klip Kanvas is built for this rung: same product brief, new hook lines, optional second avatar, same body, same B-roll. Do not use the refresh as an excuse to reopen the brand kit. The only new work is the first look and whatever the diagnosis required. If you need new footage, shoot or pull it for the proof beat after the open is drafted, so you are not waiting on a studio day while frequency cooks the live ad. Speed is part of the tactic — a late perfect film is how you refresh after CPA has already doubled.

Same rules as any test: one ad set, equal shots at delivery, no edits. At 24 hours, check delivery only. At 48–72 hours, promote the hook that brings hook rate back toward 30%+ without wrecking hold. Retire the cute line that only the Slack channel liked. Then step the winner into the scaling campaign with 20–30% budget increases every 48 hours, not a double overnight.

Read the matrix. If one new hook wins on both faces, you have a message; make that family the next default. If the new face saves an old hook, you had a casting furniture problem. If nothing pops and frequency on the old ad is still the only thing converting, check audience size and offer — you may be hammering a pool that needs widening more than it needs another confession. Do not keep stacking opens into a dead geo. Do not call the format dead because one ladder climb failed after 48–72 hours of honest spend.

When you promote, leave a challenger in the test campaign so the lab never goes empty. Fatigue is a loop. The refresh you just shipped will enter 2.5–3.5 on its own clock, faster if you scale spend. Horizontal scale (new hooks, new angles) is what keeps CPM from becoming a creative tax. Vertical scale without a queued open is how CPM and frequency rise together and everyone blames the platform. Book the next 3 × 2 on the same readout call, even if this refresh looks pretty in preview. An empty test campaign is a fatigue clock with no inventory.

Tie refresh to frequency and hook, not to a superstition like 'new ads every Monday' with no diagnostic. Many accounts need new prospecting hooks every one to three weeks once spend is meaningful. Frequency caps can slow how often one person sees one ad; they cannot invent a new opening. Over-capping a test is a good way to never reach 1,000 impressions.

A keepable cadence: weekly 3-hook batch on the hero product, readout on a fixed weekday, refresh queued when hook rolls off 30%+ and frequency is climbing, even if CPA is still fine. Smaller budgets can live longer on a winner — do not refresh so often that nothing exits learning. Calendar-only cadences beat chaos if they flex when a creative dies early or when a winner is still cheap at 2.2. Write that flex rule down. Monday feelings should not outrank the sparkline. If the only cadence you have is 'when CPA hurts,' you will always refresh late.

Audience is a fatigue lever. Broad, funded prospecting is kinder to a winner; tight stacks cook it. If you must stay narrow, budget a faster hook cadence or accept higher CPA. Use a persistent avatar in retargeting and a rotating cast on cold so the warm pool is not watching the prospecting experiment. Caps belong, if at all, on sensitive warm audiences. For cold AI UGC, spend the energy on the ladder and the next 3 × 2, not on a perfect number in the settings panel that cannot invent a new first two seconds.

When you pause a fatigued winner, tag the hook family, first frame, avatar, and the frequency where hook broke. Put it in the swipe or calendar as retired-in-market, not as a still-available concept. Relaunching the same confession six weeks later can work if the pool turned over; relaunching it next week because someone liked the cut is how you buy CPM on wallpaper.

Retirement notes stop the loop of 'we should go back to the one that worked' without checking whether it worked because it was new. If you reuse a line, change the first frame or the face, and treat it as a test cell, not as a guaranteed restore. Keep the body script — that is the expensive part — and keep rotating opens. Comments like 'I've seen this' are a qualitative confirm, not a reason to ignore the table. Combine them with hook and frequency and you will refresh on time instead of arguing about whether the algorithm 'likes UGC this week.'

Share the log with anyone who can request 'the old ad back' at a higher budget. That request is common and sometimes correct for a short retargeting burst; it is rarely correct as a cold prospecting restore at frequency already past 3. The log makes the conversation about data. Without it, every CPM spike becomes a brand workshop, which is the most expensive refresh of all. A retired winner is still a template for the body; it is not a hall pass to skip the next 3 × 2. Keep the body; rotate the open; argue from the log.