The ROAS review sheet: what to fill in, how to read it, what to change
Updated 2026-08-15 · Risk-term list version 2026-08-15c
The review answers one question: did the angle win or the hook. Both hooks on an angle up — the angle. Only one — the hook. The test is mechanical; no judgement required.
What changes afterwards is your own library: which angle moves up, which hook joins the reusable set, which rejected line joins your own flagged list.
The value is that the ledger stays in your store: an agency cannot take it when the contract ends, and neither can a media buyer who leaves. Below: the eleven columns, the four verdicts, and exactly what to change afterwards.
1. The eleven columns
One more column is not in the header and is the most valuable: the verdict. It has exactly four values — angle won, hook won, both lost, not enough spend. "Okay" and "let us see" are not permitted.
| Column | What goes in | Required? | Note |
|---|---|---|---|
| Cell | A / B / C… | Yes | Match the A/B matrix labels; do not renumber |
| Angle | Pain / Solution, Myth-bust… | Yes | One angle occupies two rows (two hooks) |
| Hook | The first three seconds, verbatim | Yes | This is the primary key. "Hook A" is useless |
| Platform | TikTok / Meta / Amazon… | Yes | Cross-platform differences are signal |
| Language | Chinese / English | Yes | Question three depends on it |
| Spend | Actually spent | Yes | Cells below the minimum get no verdict |
| Impressions | Count | No | Fill if available; it feeds the rates below |
| 3s view-through | Percentage | Yes | The only metric that judges the hook |
| CTR | Percentage | Yes | Judges the strength of the middle |
| Orders | Count | Yes | |
| ROAS | Number | Yes | Compare against the target you set before the flight, not against anyone else |
2. The four verdicts
Note the difference between "both lost" and "not enough spend": underspend is a state, not a verdict. Recording it as a verdict is deciding on noise.
| Situation | Verdict | Next |
|---|---|---|
| Both hooks on an angle hit target | Angle won | Extend the angle to the other language and another placement; produce two fresh hooks on it |
| Only one hook hit target | Hook won | Port that hook onto other angles. The angle itself is undecided — give it one more round |
| Neither hit target, both reached minimum spend | Both lost | This angle did not work in this market this round. Question three decides whether it is a category or a market problem |
| Minimum spend not reached | Not enough spend | No verdict. This is the most common false execution — calling it "the angle failed" throws away an angle that might work |
3. Cross-language: the same angle in both markets
The step most often skipped and the densest in information. The gap between an angle’s Chinese and English performance separates a category problem from a market problem.
| Chinese | English | What it means | Next |
|---|---|---|---|
| Won | Won | The angle works in this category | Raise its prior; extend it to the other sub-ranges |
| Won | Lost | Market difference, not an angle problem | Look at the English script: was it translated? Pace, CTA convention and first-person register all differ |
| Lost | Won | Same, reversed | Check whether the Chinese script sounds like an ad — the first three seconds cannot include throat-clearing |
| Lost | Lost | The category does not want this angle | Lower the prior. This is the only case where the angle library may move |
4. The rejected line is the most valuable cell
If a cell never ran — rejected, throttled, taken down — that information is worth far more than its ROAS.
Record the exact sentence in section 4 of the sheet (not "policy violation" — the actual line). Three reasons:
- It seeds your own flagged list. External lists, ours included, are generic. The line that got rejected in your category, your market, your account is more accurate than any of them.
- It will recur. The same sentence usually already exists in the listing, in three other assets and in a creator’s script. Fixing only the rejected asset just delays the next rejection.
- It transfers. On the next SKU it goes straight into the do-not-say section of the brief (template, section 8).
5. What to change afterwards
What not to change: the thresholds of the round that is currently running. Changing them after seeing the data is self-deception — the same point the A/B article makes, because it is the shared foundation of both sheets.
| What | When | How much |
|---|---|---|
| Angle prior | Only when both hooks lost and both reached minimum spend | A little at a time. Six cells in one round do not support a large move |
| Reusable hook library | Any hook that hit target | Store it verbatim with angle, language, platform and that round’s numbers |
| Your own flagged list | Any rejection | The exact line, plus platform and date |
| The product page (not the creative) | CTR fine, CVR weak | Size chart, review count, main image, price. Do not make new creative this round |
| Test parameters (per-cell minimum, window) | When most cells came in under minimum spend | Before the next round — never mid-flight |
6. Why the ledger stays with you
An agency delivers footage and campaigns, and the experience leaves with the contract. A media buyer delivers execution, and the experience leaves with the person. The review sheet delivers a record the next person can read: which angle works in which market, which hook won, which sentence got rejected.
That is why the sheet is in every plan and also a downloadable template: it is the one part of this product that gets more valuable the more you use it — and the valuable part belongs to you.
Related
FAQ
Do I fill it in only after the flight?
Thresholds before, data after. Changing thresholds mid-flight is the one prohibition on this sheet, because what you are changing then is not a standard but your interpretation of the result.
How small is too small?
If a cell did not reach the minimum you set beforehand, mark "not enough spend" and give no verdict. That is not caution, it is accuracy: deciding on noise throws away an angle that might have worked, and that loss is invisible.
Is this the same sheet as in the exported plan?
Same structure. Section 5 of the exported plan is this sheet, pre-filled with your angles and hooks. The template is for people who have not run a plan yet.
Can I move the angle priors afterwards?
Only when both hooks on an angle lost and both reached minimum spend, and then only a little. Six cells do not support "this angle does not work in this category". A prior is a starting point, not a scoreboard.
Do you collect this data?
No. The review sheet is a document you fill in yourself; nothing is uploaded. If an account system later uses review data to improve the category knowledge base, it will be in de-identified, aggregated form and it will be stated in the terms.