Ask a media buyer what creative they tested against this hook three months ago and the honest answer is often "I would have to look". The spend happened, the learning did not get recorded, and the same losing concept gets briefed again by somebody who was not there.
A testing log is not reporting. Reporting tells the client what happened. The log tells the team what has already been tried, so nobody pays twice for the same lesson.
Why the learning evaporates
The result lives in the platform, the reason lives in someone's head. The platform keeps the CPA. It does not keep "the hook was too abstract for a cold audience".
Tests get judged on one number. A creative that lost on CPA but produced the highest quality leads is recorded as a loss, and the insight is discarded with it.
Nobody writes the verdict. The test runs, the numbers come in, the team moves on. Two weeks later the finding is gone, because a number without a sentence is not a learning.
The log
One row per creative test. Nine columns.
| Field | Why it exists | |---|---| | Date and client | The two filters everyone uses | | Hook, written out in full | Not "video 3". The actual opening line, so it can be compared | | Format and length | A 6s cutdown and a 30s explainer are different products | | Audience and placement | The same creative wins and loses across audiences | | Spend | Without it, the result has no denominator | | Result, against the agreed metric | Whatever that client's metric is, named | | Verdict | Winner, loser, or inconclusive. One word, required | | The learning, in one sentence | Why it did what it did. This is the asset | | Reusable? | Can this hook be adapted, or is it dead |
The rules that make it worth keeping
A test without a verdict is not finished. Write the verdict within a week of the spend ending, or the row gets closed as inconclusive and the budget was a donation.
Inconclusive is a real verdict. Under-spent tests that cannot reach significance are common. Recording them honestly stops the team re-running the same underpowered test hoping for a different answer.
Record the loser's reason. "Hook too abstract for cold traffic" is more valuable than the winning creative, because it prevents a whole family of future mistakes.
One variable per test, written down before launch. If the hook and the audience changed together, the result teaches nothing.
What to automate
Automate: the pull of spend and results per creative, the join to the test row, and the flag when a test has ended without a verdict.
Do not automate: the verdict and the learning. Those are judgement, and a model guessing at why a creative worked produces confident nonsense that the team will then act on.
How to tell it is working
- You can answer "have we tried this hook" in under a minute.
- Losing tests are referenced in briefs, not just winning ones.
- New briefs start from the log rather than from a blank page.
- The number of tests per quarter goes up while the spend stays flat, because briefing is faster.
A testing log converts spend into knowledge. Without it, you buy traffic and forget what you learned, which is the most expensive way to run ads.