ProductUse cases
Learn
BlogAnalyses, trends & playbooks GuidesIn-depth market guides
PricingFAQ
DE · EN · FR · IT · ES · ROStart now

The DACH supplement ads playbook · Chapter 11 of 12 · 4 min read

The architecture of tests: what you test without cloning

Here is where the playbook parts ways with the screenshots folder: everything you've extracted from the market — hook patterns, psychological tensions, offer mechanics — is worth exactly as much as the tests it produces. And between analysis and test there's a step haste always skips: formulating the hypothesis.

The difference between an account that learns and one that burns budget isn't creative luck. It's the discipline of knowing, at every test, what question you're asking — and of not mixing the questions.

What you'll be able to do after this chapter

  • separate the two test levels — positioning and execution — and stop mixing them in a single experiment;
  • turn the patterns from the previous chapters into written hypotheses, with the prediction before launch;
  • read the competitors' Test History as a journal of other people's experiments.

The principle

Any ad test asks, at bottom, one of two questions — and it's crucial to know which:

  • The positioning test asks whether the angle is right: the chosen problem (ch. 3), the psychological tension (ch. 6), the angle (ch. 4). You vary the angle, hold the execution constant.
  • The execution test asks how an already-proven angle is best delivered: the concrete hook, the format (ch. 5), the first seconds. You hold the angle constant, vary the delivery.

Mixing them is the classic error: two ads differing in angle AND format AND hook — whatever result comes, you don't know what you learned. And the variants in the Ad Library show mature brands making exactly this separation: the same promise in many executions is an execution test at scale; different angles in parallel executions, a positioning test.

The chapter's second source is the funnels' Test History: the journal of the changes each competitor made on the page — new offer, new headline, new price — with what they kept and what they abandoned. Every row is someone else's experiment, readable for free: the quick abandonment of a mechanic says as much as keeping it.

The workflow in RavenBI

  1. From the inspiration collection ranked in chapter 10, take the pattern with the strongest signals and write its positioning hypothesis: "For audience A, angle U on problem P will beat my current angle." The prediction — including what failure would mean — is written before launch.
  2. Build the test on a single level: if you're testing the angle, the executions stay as close as possible; if the execution, the angle is nailed down.
  3. Before launch, check the thread from chapter 9: does every variant have its continuation on the page?
  4. In parallel, read the Test History of two or three arena veterans: what they changed on their pages in recent months, which mechanics appeared and disappeared. What they abandoned quickly removes from your list hypotheses others seem to have burned already.
  5. After the test, return to the written hypothesis and confront it with the result — in writing. The account that learns is the one with an archive of judged hypotheses, not a folder of creatives.

How to read the signals

  • A competitor with many variants on the same angle = an execution test at scale — their angle is already confirmed internally; for you, a positioning signal.
  • A competitor with parallel angles at the same time = a positioning test in progress — watch which survives; it's their answer to your question.
  • A Test History with quick abandonment of a mechanic = an experiment its owner didn't like — that hypothesis moves down your list or drops out entirely.
  • A Test History empty for months under long-running ads = a stabilised construction — the competitor has finished searching; what you see is their final answer, the cleanest object of study.

The typical mistakes

  • testing without a written hypothesis — the result then gets interpreted by taste, which is to say not at all;
  • mixing the levels — positioning and execution in the same experiment means nothing was learned for sure;
  • repeating tests the competitors' Test History already shows as burned;
  • reading your own variants by a different standard than others' — the signals from chapter 1 apply identically to your account.

Exercise

Write your first complete test in the chapter's format: the hypothesis (a single level — positioning or execution), the prediction with its success threshold, the judgment date. Check it against the veterans' Test History: if one of them has already burned the hypothesis, rewrite it. Only then produce the creatives.

A good test is an event; a system of tests is a rhythm. The final chapter ties the whole playbook into a weekly routine — what you watch, when you react and, just as importantly, when you ignore the noise.

Guide overview