AnalyticsTestingCheatsheetPlaybook

Win/Loss Creative Readout Sheet

A scoring sheet for creative tests: sample-size floors, win/loss/inconclusive bands by metric, what to scale vs kill vs iterate, and the one-page readout the next sprint actually uses.

Updated 2026-08-2410 min read

LanguageEnglish

A test without a written readout is just spend. This sheet is the win/loss scorecard: sample-size floors, metric bands, the decision each cell earns, and the one-page format you paste into next week’s sprint so ‘ad 3 felt good’ never survives Friday.

Wins are defined before spend, not after the CPA looks pretty

If the bar is ‘better than last month’s vibe’, every noisy day produces a new hero. Write the win definition on the brief: primary metric, sample floor, and the action if it misses. For cold purchase tests the primary is usually CPA or ROAS against a 14-day baseline; hook rate and hold are leading indicators that explain the CPA, not a substitute for it. A 4% CTR with a 2× CPA is not a win. A 1.1% CTR at target CPA with a 35% hook can be a win you scale while you hunt a better opener. The readout exists so those two ads do not get the same sentence in Slack. Losers are equally specific: spend ≥2× target CPA and zero purchases is a kill, not a ‘needs more time’ story you tell yourself on day nine.

Inconclusive is a legal status, not a personality

Most ‘this one is interesting’ ads are under-sampled. Below ~1,000 3-second views you cannot rank hooks; below ~1× target CPA in spend you cannot rank purchase CPA; below ~10–15 conversions you cannot crown a scale candidate. Calling those ads losses trains the team to kill too early. Calling them wins trains the team to scale noise. The sheet forces a third bucket: inconclusive, plus the exact extra spend or extra days required to exit it. If the budget cannot fund that extra spend, you did not run a test — you ran a lottery. Cut the number of concurrent ads until each cell can actually graduate. Volume is not the same as learning.

Read the cell, not the celebrity ad

The point of a 3×2 grid (hook × avatar, or hook × offer) is to learn a family, not to pick a file. If all three hooks on Avatar A beat all three on Avatar B at the same sample, the readout is ‘face’, not ‘hook 2’. If one hook wins on both faces, the readout is ‘angle’. If everything is flat, the readout is ‘offer or page’, not ‘we need wittier copy’. Teams that only ship the single best file throw away the only transferable lesson and then wonder why the sequel flops. The locked tables tag each outcome with the next-sprint action: scale the winner, iterate the family, rebuild the angle, or fix the landing page. That sentence is the whole deliverable.

How to run a Friday readout in 20 minutes

Export last test window at ad level: spend, impressions, 3s views, hook, hold, CTR, CPC, CPA, purchases, frequency. Drop each ad into win / loss / inconclusive using the sample floors and metric bands. Then fill the one-page template: what won, why (hook vs face vs offer), what dies, what iterates, what the next 6 cells are. Production for next week is that list, not a new brainstorm. When the next cells are hook variants of a proven body, a product-URL pass in Klip Kanvas is how the readout becomes files before Monday’s ABO, instead of a slide that nobody renders.

5 tables inside: sample-size floors by question, win/loss/inconclusive bands by metric, family-level (hook vs face vs offer) diagnosis, next-sprint action codes, and the one-page Friday readout template with required fields.

Sample-size floors (do not score below these)

Hook ranking (3s / impressions): Floor 1,000 3s views per ad · directional 1,000–2,500 · stable 2,500+ · ignore hook on <400 3s
Hold / ThruPlay ranking: Floor 1,000 3s views · need hook ≥20% or hold is a starved sample · do not rank hold before hook clears
CTR ranking: Floor 2,000 impressions · outbound CTR needs the click URL live · ignore CTR if hook already <18%
CPA / purchase ranking: Floor spend ≥1× target CPA · crown a winner at 10–15 conversions · do not scale on 2–3 purchases
ROAS ranking: Same conversion floor · plus value variance: AOV CV >0.6 needs 20+ orders before you trust ROAS
Frequency / fatigue call: Need 7 days live AND ≥2,000 impressions · day-2 frequency is not a readout
Under-floor label: Mark INC · write the $ gap to floor · if gap > remaining test budget, kill the cell as untestable, not as a loss
Concurrency cap: Max ads = floor(daily budget ÷ (0.5–1.0 × target CPA)) · extra ads are guaranteed INC

Win / loss / inconclusive bands (cold purchase, vs 14d baseline)

WIN — scale candidate: CPA ≤0.90× baseline · ≥10 conversions · hook ≥25% (or ≥ week-1 if a sequel) · freq <2.5 cold
WIN — iterate, do not scale yet: CPA 0.90–1.10× baseline · hook ≥25% · <10 conversions · buy to 10–15, then decide
LOSS — kill: Spend ≥2× target CPA and 0 purchases · or hook <18% at 1,000+ 3s · or CPA ≥1.40× for 3d
LOSS — never-worked concept: Hook never cleared 25% across 2+ variants · rebuild angle · do not ‘refresh’ a concept that never won
INC — more spend: Spend <1× CPA · metrics inside ±20% of baseline · write $ to floor · do not call it a winner in Slack
INC — mixed signals: Hook ≥30% and CPA ≥1.30× · page/offer suspect · hold the creative, fix LP, re-read 72h
Watch — fatigue, not a loss: Was WIN · now hook −15 to −25% · freq 2.5–3.5 cold · action = refresh, not kill-and-forget
Lead vs lag: Hook/hold explain CPA · never crown on CTR alone · never kill on CPM alone if CPA is on target

Family diagnosis (what actually won)

Hook family win: Same opener wins on ≥2 avatars or ≥2 offers · next sprint: 4 more first-3s in that family, keep body
Face / avatar win: All hooks on Face A beat Face B at matched sample · next: 3 new hooks on A, retire B for this SKU
Offer win: Same film, two offer lines, CPA gap ≥20% at ≥10 conv each · next: keep winner, test stack vs discount
Format / ratio win: 4:5 Feed CPA ≤0.80× 9:16 or reverse at matched spend · next: export the winner’s ratio first, not ‘all ratios’
Length win: 15s vs 30s at matched hook · if 15s CPA ≤0.85×, cutdowns are the scale unit · keep 30s as warm
Page / offer failure: Hook ≥30% and CTR ≥1.2× baseline and CPA still ≥1.3× · do not iterate hooks · LCP, match, checkout first
No family signal: All cells INC or all within ±10% CPA · too many ads or too small budget · cut grid to 4 cells and restage
Write the sentence: Required format: ‘[Hook|Face|Offer|Page] won because [metric + sample]. Next: [N] cells.’ · no adjectives

Next-sprint action codes

S — Scale: WIN scale candidate · +20–30% budget · 48–72h cooldown · keep 10–20% on a backup cut
I — Iterate hook: Family win on opener · 3–5 new first-3s · same proof/offer · ABO test set, not dumped into CBO
R — Refresh fatigued winner: Was S · now freq 2.5–3.5 · hook −15%+ · 4–8 replacement cuts · original to backup
B — Rebuild angle: Never-worked · new angle from the offer grid · not a synonym hook · 2 offers × 2 hooks min
P — Page/offer fix: Mixed-signal INC · pause new hooks 72h · LCP ≤2.5s · claim match · then re-read same ads
K — Kill: LOSS kill rule hit · archive · no sequels · log the reason code so it does not return as v7
H — Hold as control: Baseline winner still healthy · do not ‘improve’ it · it is the control the new cells must beat
Mix rule for the next 6 cells: Iterate week: 60% I / 25% R / 15% B · Rebuild week: 70% B / 30% I · never 6 unrelated ideas

One-page Friday readout (required fields)

Window: Dates · platform · funnel temp · daily spend · # ads launched / # that hit sample floor
Scoreboard: Count of S / I / R / B / P / K / H / INC · spend $ in each · % of spend in S+H should be 40–70%
Winner sentence: One line from the family table · metric + sample · file token · landing URL
Loser sentence: One line · reason code (hook / claim / page / policy / under-sample) · do not list 12 files
Fatigue flags: Any S-ad now in 2.5–3.5 cold freq · hook vs its week-1 · replacement due date
Next 6 cells: Token list · action code per cell · owner · render-by date · kill rule restated
Budget next week: Test ABO $ · scale CBO $ · freeze if INC share >50% of ads · do not add cells until floors are fundable
Artefact: This page in the sprint doc + CSV export attached · Slack is not the archive · same tokens as ad names

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