AnalyticsScalingCheatsheetMeta Ads

Frequency & CPM Diagnostic Sheet

A diagnostic sheet for rising CPMs: frequency bands (cold fatigue 2.5–3.5), mix-shift detectors, auction-vs-creative isolation tests, and the scale / refresh / wait / rollback action grid.

Updated 2026-08-2410 min read

LanguageEnglish

When CPM jumps, most buyers either cut budget or blame the auction. Both can be wrong. This diagnostic sheet separates auction inflation from mix-shift from creative fatigue — including the cold-frequency 2.5–3.5 zone — so you scale, refresh, or wait on numbers instead of a mood.

A CPM number is a symptom, not a diagnosis

A $22 CPM is not a verdict. It is a symptom that can come from three different machines. Auction inflation: the same people, same placement, same creative, paying more because more advertisers showed up or you asked for a rarer event. Mix-shift: spend slid into Feed, the US, ages 25–34, or Purchase while cheaper Reels or a cheaper country fell, so blended CPM rose even though nothing ‘broke’. Fatigue: frequency on cold climbed through 2.5–3.5, unique reach share collapsed, and the algorithm is bidding harder to find leftover people who still have not seen the ad. If you cut budget on mix-shift you punish a cheap placement that shrank. If you refresh creative on pure auction inflation you waste a production cycle. Fork on frequency first, then mix, then auction.

The 2.5–3.5 band is a fatigue window, not a CPM cap

Teams treat frequency as if it had one number for the whole account. It does not. A 3.0 frequency on cold prospecting is a warning; a 3.0 on 7-day add-to-cart is barely warmed up. The band you were asked to watch on cold is 2.5–3.5: below 2.5, do not call a CPM spike ‘fatigue’; inside 2.5–3.5, diagnose with hook rate and unique-reach share; above 3.5 on cold, plan a replacement even if CPA has not blown up yet. CPM can rise inside that window because you are paying to re-find the same people, or it can rise with frequency still at 1.4 because the auction or the mix moved. Those two cases get opposite actions. This sheet is the fork; the weekly fatigue audit is the production follow-up once you have already decided it is creative, not auction.

Read mix before you read the auction

Blended CPM is a weighted average. If Advantage+ or a CBO pool shifts 20 points of spend from Reels into Feed, blended CPM jumps and every ad looks ‘more expensive’ even though unit prices per placement did not. Country mix, age mix, and objective mix do the same trick. Before you declare inflation, lock last 7 days versus prior 7 at placement, country, age, and objective — not at campaign total. If the cells that grew are the expensive cells, you have a mix problem (structure, exclusions, creative that only wins on Feed). If every cell is up 20%+ at constant mix, then you have an auction or a fatigue problem, and frequency tells you which. Skipping this step is how people ‘optimise’ a winner that merely stopped getting cheap inventory.

How to run the diagnosis in 15 minutes

Export last 7 versus prior 7 at ad set: CPM, frequency, unique reach, hook rate, CTR, CPA, and spend share by placement and country. Plot each ad set on the CPM × frequency grid in the first locked table. Then run the mix-shift detectors. Only the cells that still look like inflation after mix is constant go to the isolation tests (hold creative, hold geo, split placement). The last table is the action: scale, refresh, wait, or roll back. When the verdict is fatigue in the 2.5–3.5 cold band, the next job is replacement volume — a URL-to-ad pass in Klip Kanvas is how those hook variants exist on Tuesday instead of in three weeks.

5 diagnostic tables inside: CPM × frequency decision grid, cold/warm/hot frequency bands (fatigue zone 2.5–3.5 on cold), mix-shift detectors, auction-vs-creative isolation tests, and the scale / refresh / wait / rollback action sheet.

CPM × frequency decision grid (7-day, cold first)

CPM +15–25% · freq 1.2–2.0 cold: Not fatigue · check mix then auction · do not refresh · hold budget unless CPA +25%
CPM +25%+ · freq 1.2–2.0 cold: Auction or mix · run isolation tests · refresh only if hook already −15%
CPM flat · freq 2.5–3.5 cold: Early fatigue · queue 3–5 new first-3s · cap spend share of that ad at 40%
CPM +15–30% · freq 2.5–3.5 cold: Fatigue primary · unique reach share usually <0.40 · replace this week · do not raise budget
CPM +30%+ · freq 3.5+ cold: Replace now · original to 10–20% backup · no vertical scale until freq <2.5 on the new ad
CPM down · freq 2.5–3.5 cold: Cheap inventory on tired people · still refresh · low CPM is not permission to hammer
CPM +20%+ · freq <2.0 · hook −15%+: Creative miss, not auction · kill/rebuild on hook · CPM is a side effect of a weak 3s
Need 2 coordinates: Never act on CPM alone · pair with 7-day frequency · confirm with unique reach share <0.45 watch / <0.35 act

Frequency bands (the 2.5–3.5 rule in context)

Cold prospecting (broad / LAL / interest): Healthy 1.2–2.0 · Watch 2.0–2.5 · Fatigue zone 2.5–3.5 · Replace 3.5+ · 7-day window
Warm (50%+ viewers, 14–30d engagers): Healthy 1.8–3.0 · Watch 3.0–4.0 · Fatigue 4.0–5.5 · Replace 5.5+ · 7-day
Hot ATC / IC 7d: Healthy 2.5–4.5 · Watch 4.5–6.0 · Fatigue 6.0–8.0 · Cap 8 · higher frequency is expected
Purchasers (upsell / replenish): Healthy 1.5–3.0 · Watch 3.0–4.0 · Fatigue 4.0+ · 14-day · do not weekly-hammer 180d buyers
7-day vs 30-day: Diagnose on 7-day · 30-day 5–8 on cold = long-tail fatigue even if 7-day reads 2.2 · check both
Unique reach share confirm: Watch: unique ÷ impressions <0.45 · Act <0.35 · Kill <0.25 on cold · same people, more times
Spend-speed modifier: $50/day: +~7d to hit 2.5–3.5 · $500+/day: −~5d · high spend enters the zone faster, not at a higher cap
CBO mixing temperatures: If one CBO mixes cold+hot, frequency is unreadable · split · diagnose at ad set, never campaign blend

Mix-shift detectors (last 7 vs prior 7)

Placement mix: Flag if Feed spend share +10 pts or more · Reels/Stories −10 pts · explains +15–40% blended CPM with stable unit CPM
Country mix: Flag if US/UK/AU/DE share +8 pts · IN/BR/MX −8 pts · US Feed Purchase is several× MX Reels Views
Age mix: Flag if 25–44 share +10 pts on a brand that was winning 18–24 · 25–44 CPM often +20–50% vs 18–24
Objective mix: Flag if Purchase share rose vs Traffic/Views · Purchase CPM 1.5–3× a cheap objective on the same geo
Advantage+ / ASC drift: Flag if ‘Advantage’ inventory or Audience Network +15 pts of spend · unit CPM may fall while CPA rises — not a win
Creative-driven mix: 9:16-only file that only wins Reels · when Reels caps, Feed gets the letterbox and CPM+CPA jump · export 4:5
Constant-mix test: Reweight prior-7 CPM to this-week placement×country weights · if gap closes to <8%, it was mix, not inflation
Pass threshold: Mix-shift if ≥1 detector flags AND constant-mix gap <8% · else proceed to isolation tests

Auction vs creative isolation tests

Hold-creative geo split: Same ad, two geos · if both CPM +20%+ at freq <2.2, auction · if only the high-freq geo, fatigue
Hold-geo new-hook test: Ship 3 first-3s, same body, same ad set · if new hook CPM −15%+ in 3d, it was creative · if all match, auction
Placement split: Force Reels-only vs Feed-only 48–72h · if Feed CPM 1.6–2.5× Reels, mix not fatigue · restore 4:5 file
Time-of-week control: Compare Tue–Thu vs Tue–Thu · Mon/Sun CPM noise ±10–20% is not inflation · need 3 matched weekdays
Seasonal overlay: Q4 / BFCM / Prime / payday weeks: expect CPM +15–60% at constant freq · do not refresh solely on that
Learning-reset false positive: Edit, duplicate, or +50% budget in 24h can lift CPM 10–30% for 48–72h · wait one cooldown before diagnosing
Hook-rate tie-break: Hook ≥ week-1 and freq <2.2 and CPM up → auction/mix · hook −15%+ and freq 2.5–3.5 → fatigue even if auction is also hot
Sample floor: Do not diagnose on <2,000 impressions or <1× target CPA in spend · variance masquerades as inflation

Action sheet: scale / refresh / wait / rollback

Scale (vertical +20–30%): Freq <2.2 cold · hook ≥ week-1 · CPA ≤ 14d baseline · mix stable · cooldown 48–72h after last raise
Scale sideways (new geo / placement): Winner at freq 2.0–2.8 cold · CPM rising from reach cap · copy to a new country or 4:5 Feed · do not +50% same cell
Refresh hooks: Freq 2.5–3.5 cold · hook −15 to −25% · hold stable · 3–5 new first-3s · same offer · no budget raise that week
Wait 72h: CPM +15–25% · freq <2.2 · mix flagged OR learning reset in last 48h · no edit · re-read on matched weekdays
Rollback last raise: After +20–30% step, CPA +25%+ for 3d AND freq still <2.5 · return to prior budget · not a creative funeral
Kill, do not refresh: Hide+decline >1% · hook never cleared 25% · CPA +40% with hook <20% · off, no sequels
Budget freeze rule: Any cold ad set in 2.5–3.5 without a replacement in draft = freeze vertical scale · production is the CPM control
Re-entry: After refresh: new ad must show freq <2.2 and CPM within +10% of its own day-3 by day 7 · else rebuild angle

Locked

Unlock the full cheatsheet

Enter your email and we'll instantly unlock the complete cheatsheet — plus send you fresh ad playbooks every month.

No spam. Unsubscribe anytime.

Ready to put this into practice?

Create your first AI UGC video ad in minutes — no filming, no actors, no editing.

Try Klip Kanvas free