# Meta Creative Testing

> Goal: find the winning creative with the smallest budget, and keep a steady flow of new ones. Structure decisions come second; most accounts lose more money on too few creatives and too few events than on the "wrong" test structure.

- Skill: `ua5er/meta-creative-testing` (Agent Skill)
- Install (CLI): `npx get-skill add ua5er/meta-creative-testing`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ua5er/meta-creative-testing/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Finance & Business
- Author: ua5er (https://skillmd.com/u/ua5er)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/ua5er/meta-creative-testing

---


# Meta creative testing framework

Goal: find the winning creative with the smallest budget, and keep a steady flow of new ones. Structure decisions come second; most accounts lose more money on too few creatives and too few events than on the "wrong" test structure.

## 1. Pick creatives from history
1. Export ads with date range "Maximum": ad name, delivery, spend, purchases, CPA, ROAS, CTR, quality ranking, creative thumbnail.
2. Rank by purchases, then CPA, then ROAS. Keep ads with a meaningful sample (≥ 5 purchases) or a CPA well under target on a small sample.
3. Drop: rejected ads (cannot be duplicated, re-upload an edited version instead), "- Copy" duplicates of an ad already picked, ads with spend ≥ 2× target CPA and 0 purchases, same visual already covered.
4. Write a one-line reason for every included and every excluded creative. Stakeholders ask for both.

Output table: creative | purchases | CPA | ROAS | in/out | reason.

## 2. Choose the test structure
| Option | When it fits | Cost of an answer |
|---|---|---|
| **One broad ad set (CBO), all creatives + auto-rules** (default) | Account below ~100 purchases/week; mixed static and video; budget-limited | Lowest: weak ads stop at ~2× target CPA |
| **ABO, fixed budget per creative** | High volume (each test ad set can reach ~50 events/week) or a hard requirement to compare concepts evenly | High: N creatives × (≥ 3 to 4× target CPA) before any verdict |
| **Meta Creative Testing tool** (2 to 5 ads inside an existing ad set, equal split, then back to normal delivery) | Need a fair read on 2 to 5 specific new ads without a new ad set | Medium: the test budget is spent evenly regardless of early signal |

Explain the default with this logic (use the account's numbers):
- With a fixed budget, a weak creative spends its full budget even when it is clearly losing on day 1, and a strong one is capped and starves.
- 11 creatives at $150 each = $1,650 for one answer. Previous fixed-budget tests that got $250 to 320 each and 0 to 2 purchases answered nothing.
- In one ad set, Meta shifts spend to creatives with early signals; the stop rule caps the loss on a bad creative at ~2× target CPA.
- Each creative is a separate ad with a clear name, so per-creative metrics stay visible.
- Testing tools are also built to increase test spend. The advertiser's goal is the cheapest winner.

Mixing static and video in one ad set is fine: one format taking most of the spend is itself the answer.

## 3. Winner and kill criteria
- **Kill:** spend ≥ 2× target CPA and 0 purchases; or spend > 4× target CPA and < 3 purchases.
- **Winner:** ≥ 5 purchases at CPA ≤ target over ≥ 3 days, while still getting spend share.
- **Keep watching:** everything else. Do not judge on CTR alone.
- **Fatigue:** frequency > 3 with CTR falling and CPA rising over 5 to 7 days.
- Restart rule for late attribution: a paused ad that later shows 2+ purchases in 3 days at CPA under target goes back on.

## 4. Creative pipeline
- New batch every week, 3 to 6 creatives, into the same ad set. Concepts, not small variations: different angle, persona, hook, format.
- Mix: proven winner iterations (new hook on the same idea), new angles, UGC video, static with a strong headline.
- Name every ad: `PROJECT | TEST | Format | Concept | src: file_or_original`.
- Optional screen before production: run the batch's concepts through `creative-preflight-panel`. Go items and fixed rework items enter the batch; the panel never decides kill, winner or budget, the criteria in section 3 do.

## 5. Brief template for designers and UGC creators
```
Product / offer:
Audience and their main pain (one sentence):
Angle (why now, what changes for them):
Hook (first 1 to 3 seconds / headline):
Proof (number, review, demo):
CTA:
Format: 9:16 video 15 to 30s / 1:1 and 4:5 static / both
Must avoid: income promises, before/after claims, platform policy risks for the vertical
Reference ads (links from Ad Library or account winners):
Deliverables and deadline:
```
Generate 3 to 5 filled briefs from the winners' patterns and the funnel's selling points. For UGC, add a shot list and a script with the hook in the first line.

## 6. Reporting
Weekly table: creative | spend | purchases | CPA | ROAS | status (winner / watch / killed / fatigued) | next action. One paragraph on what the batch taught us and what the next batch tests.

## Rules
- Never declare a winner below 5 purchases unless the client accepts the risk in writing.
- Do not change creatives and budget on the same day; batch changes weekly.
- Keep claims in ad copy policy-safe for the vertical (employment, health, finance).

## Author

Vlad Yasinski, AI growth consultant for app and subscription businesses. Telegram: @ua5er (https://www.roimobi.com/tg?src=skill-meta-creative-testing). Site: https://www.roimobi.com/?utm_source=github&utm_medium=skill&utm_campaign=user-acquisition-skills

