Creative-testing log: the column set that keeps a test honest

Most creative-testing "systems" collapse into a folder of screenshots and a memory of what felt like it worked. Six weeks later nobody can say why the winning hook won, whether it was ever really tested, or what to try next. The fix is not a tool. It is a sheet with the right columns and three rules about when you are allowed to write in them.

This is the log the Dispatch keeps for paid-social creative. It records what you tested, holds you back from calling a result before it has earned a verdict, and turns every test into a sentence the next test can build on. It changes nothing in any account; it is a discipline, not a script.

Before you start

• Decide the ONE primary metric you will judge on before the ads go live, and write it in the column. For most prospecting tests that is cost per result (purchase, lead, install); for a top-of-funnel hook test it may be cost per three-second view or hook rate. Pick one.
• Decide your spend gate: the amount an ad must spend before its verdict column is allowed to be filled. Below the gate, the verdict stays blank. A common starting gate is roughly three target CPAs, or whatever spend gives you a handful of results.
• Agree who owns the sheet. A shared log with no owner rots; one person writes the verdicts.

The columns

Column	What it holds
Concept	The idea being tested, in a few words ("founder piece to camera", "UGC unboxing").
Hook	The first three seconds or the opening line, quoted.
Format	Static, video, carousel, aspect ratio, length.
Audience	Where it ran (broad, interest stack, lookalike, retargeting).
Date live	When it started delivering.
Spend gate	The spend this ad must reach before a verdict is allowed.
Primary metric	The single number this test is judged on, declared up front.
Verdict date	The earliest date the ad has cleared the gate and can be judged.
Verdict	Win, hold, or kill. Blank until the gate is cleared.
Learning	One reusable sentence about WHY (see the rules below).
Next action	Scale, iterate the hook, retire, or park for later.

The decision rules that make it work

Three rules do the actual work; the columns just hold them.

1. A spend gate before any verdict. The Verdict column stays empty until the ad has spent past its gate. This is the rule that stops the sheet lying to you. An ad judged at half the gate is judged on noise, and noise usually flatters whatever launched first.
2. One primary metric, declared before launch. You write the metric in the column when the ad goes live, not when the results come in. Picking the metric afterwards is how a losing ad gets re-scored on the one number where it happened to look good.
3. Learnings written as reusable sentences. "Video 3 won" teaches nothing. "A face in the first second beat a product shot for cold prospecting" is a hypothesis the next test can carry forward. Write the learning so it would still make sense pasted into a different test's row.

A worked example

Two rows, as they would read after the gate is cleared:

Concept	Hook	Format	Primary metric	Verdict	Learning	Next action
Founder to camera	"I almost shut this company down"	9:16 video, 22s	Cost per purchase	Win	Founder-confession opener beat the polished promo on cold broad.	Scale, cut a 15s version.
Studio product spin	Silent product rotation	1:1 video, 10s	Cost per purchase	Kill	Silent openers under-delivered on cold; no clear reason to test again soon.	Retire.

Set it up

1. Copy the eleven columns above into a fresh sheet, one row per ad (not per campaign).
2. Fill Concept through Primary metric when you launch. Leave Verdict, Learning and Next action blank.
3. Set the Spend gate and Primary metric per row before the ad delivers a single result.
4. Each review, only touch rows that have cleared their gate. Write the verdict, then the one-sentence learning, then the next action.
5. Once a month, read the Learning column on its own. That column, not the win rate, is the actual output of the system.

How to read it, and what it will not do

The log is a record of what you decided and why, not a significance test. It does not tell you a result is statistically real; a spend gate of three CPAs is a practical floor for "worth a verdict", not proof. On small budgets even a cleared gate can hide a lot of variance, so treat a single win as a lead to follow, not a law. It also cannot judge creative you never logged: an ad launched off-sheet is invisible to it, and the most common failure of this system is simply not writing the row. Its whole value is that in six weeks the Learning column answers "why did that win", and the next test starts from a sentence instead of a screenshot.