Creative, audience, and page as one matrix
Multivariate creative testing here is a nested experiment tree, not one giant A/B. Each layer gets an ID, a purpose, and a stop rule so the stack stays readable.
Published 2026-09-09 · Search job: multivariate creative testing
Multivariate creative testing on a small paid program is a nested tree, not one test that changes creative, audience, and page at once. The June 2026 plan is explicit: many campaigns, many targeting variants per angle, many creatives per campaign, many page styles, and many copy variants inside each style. The discipline is that every layer gets an ID and a purpose so we can read the stack later instead of blending everything into one confusing average.
The layers
- Angle. One strategic promise. Six keys in the first portfolio.
- Campaign cell. The spend and delivery container. Separate Adprime Zero campaigns, backed by separate exchange campaigns, when we need clean budget and targeting separation.
- Targeting variant. Content category, catalog-backed segment, inventory, or geo. Targeting changes belong at the cell level, not hidden inside creative rotation.
- Creative and messaging variant. Headline, body, CTA, visual, format. A backlog plus a smaller active rotation.
- Page style. Layout and persuasion model, separate from the copy. Short-direct, pain-led, interactive-planner, and the rest.
- Copy variant inside the template. Outcome-first, pain-first, proof-first. Same skeleton, different words.
- CTA and capture. Account CTA versus lead magnet versus planner.
- Nurture. Direct activation versus education follow-up, inheriting the same IDs.
If an ad, page, or targeting cell cannot be assigned to one angle, it should not be in the first-wave paid test.
How to read a winner
The first readout should pick creative winners inside each campaign cell, then compare those winners across targeting cells. Avoid comparing every creative against every other creative across unrelated targeting. That hides whether the creative or the audience caused the result.
Do not declare a copy winner if the winning copy only ran on the best targeting cell and the losers ran on weaker supply.
Do not change two layers in the same cell on the same day if you still want a readable cause. The buying-side version of that rule is One trafficking change per day. The ideas-side version is: isolate the question in the name. Name the campaign so a stranger can read the stack.
This is operator-judged method from the June 2026 plan. It is not a published matrix of winners. The IDs exist so a later Lab piece can say which layer moved.