Creative A/B: how much per cell, how long, how to read it
Updated 2026-08-15 · Risk-term list version 2026-08-15c
Default: 3 angles × 2 hooks = 6 cells, $30 minimum per cell, 5-day window, roughly $60 opening daily budget.
The rule that matters most has nothing to do with the numbers: set the thresholds before the flight. Deciding them after seeing the data is self-deception, and it outranks every specific figure here.
The whole design answers one question: did the angle win or the hook. Below: where the numbers come from, how to adapt them to your business, and six ways people fool themselves.
1. Why six cells
Squeezed between two constraints:
- Floor: fewer than three angles tells you nothing about angle. With two, the winner may just be the less-bad one, and you learn nothing about which narrative this category rewards.
- Ceiling: more than three angles thins each cell below significance. The budget is fixed; more cells means less per cell, and past a point the differences between cells are noise.
- Two hooks per angle is mandatory, otherwise "angle or hook" cannot be answered — and that question is the entire reason the matrix exists.
2. No two cells may test the same sentence
This is the matrix’s only hard constraint, and it matters more than it looks.
One hook often fits two angles ("I have worn this for a month, here is the honest version" works for pain-solve and for unboxing). Sharing it across scripts is fine, but a collision in the test matrix destroys the read: two cells running the same sentence cannot tell you whether the sentence or the angle won — which is precisely the information you are paying for.
Our matrix generator falls through to the angle’s next hook on a collision, and if there is none it gives that angle a single cell. Better a matrix short one cell than two cells testing one sentence; that cell’s money buys nothing.
3. Adapting $30 and 5 days to your business
Defaults, not laws. Their job is to guarantee each cell spends enough to separate signal from noise. The logic to reason from:
| Your situation | Adjust | Why |
|---|---|---|
| High price point ($40+) | Raise the per-cell minimum | Conversions are sparser; more spend before an order appears |
| Long conversion lag (days from cart to order) | Extend the window to 7 days | A window shorter than the lag tests carts, not orders |
| Tight budget | Cut angles, not hooks: 2 × 2 = 4 cells | Four cells still separate angle from hook; one hook per angle tells you nothing |
| Angle already settled, optimising hooks | 1 angle × 4 hooks | That is optimisation, not exploration. Not the first round for a new SKU |
| Cold account or new store | Extend the window; do not conclude during the learning phase | Learning-phase data is not comparable to what follows |
4. Kill and scale criteria
Fill this in before the flight and keep it visible. Adjust the numbers using the previous section — but adjust them beforehand.
| What the numbers look like | What it means | What to do |
|---|---|---|
| 3s view-through < 25% | Hook problem | Swap in the other hook on that angle. Do not touch the angle |
| 3s fine, CTR < 0.8% | The middle does not convince | Fix the proof beat with concrete, checkable evidence |
| CTR fine, CVR < 1% | Not a creative problem | Landing page, listing or price. Stop making creative |
| Minimum spent, no orders | Dead cell | Stop; move the budget to the other hook on that angle |
| ROAS > 1.2× target at minimum spend | It works | +50% budget, and immediately produce two new hooks on that angle |
| Both hooks on one angle up | The angle won | Extend that angle to the other language and another placement |
| One up, one dead | The hook won | Port that hook onto other angles |
5. Why a good CTR with a weak CVR means stop making creative
This one saves a wasted production round a year, and almost everyone gets it backwards the first time.
A healthy CTR means the creative did its job: it brought the right people to the product page. They arrived and did not buy, so the problem is downstream — price, review count, size chart, main image, stock, shipping, page speed. None of that is under the creative’s control.
Shooting ten more assets now delivers, at best, more people into the same funnel gap: more spend, same conversion rate. Go look at the product page instead. Is the size chart clear? Are there enough reviews? Does the first image say what this is? Where does the price sit against comparable listings?
One shapewear-specific note: the top return driver here is sizing, not effect. High CTR with low CVR should send you to the size chart and to whether you publish concrete fit data ("I am 5'4", 132 lb, wearing M").
6. Six ways people fool themselves
- Moving thresholds mid-flight. Deciding a standard is "too strict" after seeing the numbers is the numbers persuading you, not you judging. Change thresholds before the next round, never during this one.
- Reading blended ROAS only. A good total can be one strong cell subsidising five that are burning money. Read cell by cell or you cannot tell what to keep.
- Calling underspend a creative failure. Mark cells below minimum spend as "not enough spend" and conclude nothing. This is the most common false execution.
- Changing two variables at once. New hook plus new thumbnail plus new music — a win tells you nothing about which one won.
- Restructuring on one result. Six cells in one round support "this angle and this hook did not run this time", not "this angle does not work in this category".
- Forgetting to record the line that got rejected. If a cell never ran because of review, that information is worth more than its ROAS — it belongs in your own flagged list (section 4 of the review sheet).
Related
FAQ
What if the budget will not cover six cells?
Cut angles before hooks: 2 × 2 still separates an angle failure from a hook failure. One hook per angle tells you nothing — that is not saving money, that is turning the spend into an expense with no information attached.
Is five days too short?
Depends on your conversion lag. The window exists to guarantee enough spend to separate signal from noise; the date is the means, not the goal. If cart-to-order takes three days, five days is testing carts — extend to seven.
Can I test hooks only?
Yes, but that is optimisation, not exploration. A new SKU should settle the angle first — with the wrong angle, a perfectly optimised hook just tells the wrong story more smoothly.
Both hooks on an angle lost. Is the angle wrong?
Not necessarily. First check whether the cells reached minimum spend, then check the same angle in the other language. Lost in both — the category does not want it. Lost in one — that is a market difference. This is question three on the review sheet.
Where do these numbers come from?
They are starting points written into the default matrix, not regressed from campaign data — we do not have that data and will not pretend we do. Their job is to give you a baseline you can argue with and replace, not an authority. Once your review sheets fill up, override them with your own numbers.