A testing matrix is an overview of the ads you want to compare and the differences between them. It makes production easier to plan and the result easier to read. The number of cells matters less than the question each cell is meant to help answer. If your best ad is wearing out, a matrix helps you turn what it proved into genuinely new concepts instead of more copies of the same one.
Choose the size by your test question and capacity. Four variants can hold a simple comparison, while 35 can cover several products or markets. Neither number is a platform limit or a guarantee of learning, and the larger plan is only useful when the extra cells have a clear job.
Start from one decision
Choose what the team wants to know. Should the product be shown in use or on its own? Is a practical argument more relevant than an argument about looks? Write the question down before you spread ideas across cells.
Separate a new idea from a small change. A new text colour is rarely the same as a new selling argument. Both can be worth testing, but they answer different questions and should be named accordingly.
A new concept changes the reason to buy or the story that shows it, so it asks a new question. A new crop, opening line or caption on the same story is a variant: useful for keeping a working ad fresh, but it still tests the same idea. When your best ad is wearing out, write down what it seems to prove before briefing anything new. Then either keep that reason and tell it through a different situation, or test the next reason to buy. The guides to spotting creative fatigue and to a weekly creative cadence cover the timing.
To see which reasons to buy your live ads already give, paste them into the free Creative Blind Spot Map. It shows which buying questions they answer, which they repeat and which of your priorities no ad answers yet.
Use a simple structure when it is enough
A hypothetical example is two product arguments with two openings for each. That gives four variants. It is an illustrative structure, not a recommendation for every account. If both parts change at once, be careful about explaining which part caused the difference.
A larger plan can hold more products, languages or situations. Expand only when the extra difference is relevant and there is capacity to produce, approve and test it. Otherwise more cells can spread a limited budget without giving a clearer answer.
Agree measurement before delivery
Keep the reference ad, variant names, period and any significant changes along the way. Do not let today's best number rewrite the decision rule afterwards.
Before launch, agree four things with whoever launches the ads: which observations count as comparable, who approves each variant before it runs, what else changes in the campaign at the same time, and who reads the result and decides the next brief. Exact settings and thresholds belong in the internal test plan, not in the matrix.
Record raw counts and differences in delivery. If some variants are barely shown, the matrix has not necessarily answered anything about them. "Unresolved" is a useful conclusion when it is accurate.