When you run ads, it’s not always clear which parts of the creative are driving results. I recently helped an org evaluate their Meta ad creative, and it felt pretty helpful for the time it took.
The basic approach was:
- Pull a report that breaks performance down by each image/video asset (asset-level breakdown) instead of by ad, then export it to CSV.
- Manually tag each asset using binary tags (yes/no) based on major creative levers, not minor design details. For example: human present, eye contact, CTA presence, urgency language.
- Use that tagged sheet to compare cost per result for creatives that have each element vs those that don’t.
I uploaded the CSV to ChatGPT to quickly calculate cost per result and summarize the findings, then spot-checked the math.
A few things I did to avoid misleading conclusions:
- I focused on assets with enough volume so I wasn’t basing conclusions on tiny sample sizes.
- I split the analysis around a major campaign change (budget structure) and emphasized patterns that held up in both periods.
- I treated everything as correlation, not causation, until it’s A/B tested properly.
I'm curious if other orgs have done analyses like this that go beyond what's readily available in an ad platform.