One Video, Five Line Items

A creative team wants one answer from the data: what should we make next? Ads Manager can't give it to them, because it has no idea what a creative is. It knows ads, ad sets and campaigns.

Why the answer stays buried

Results get split. Run the same video in five campaigns and Ads Manager reports it as five separate line items. Putting it back together means exports and pivot tables.¹

Hand-tagging doesn't last. Comparing hooks or offers means someone labels every ad, then renames every ad each time the team wants to track something new.¹ It gets done once, then it stops.

Creatives wear out fast. Meta's own analysts found that by the fourth repeated exposure, the chance of a conversion drops by about 45%.²

The warning comes late. Meta marks a creative as fatigued once its cost per result is about twice what it used to be.³ By the time the flag shows up, the money is already gone.

What we built

• A nightly sync. Every ad and its results, pulled from Meta's Marketing API.
• One asset, one row. Duplicates are merged into a single creative using the video or image ID plus visual similarity.
• Tagged once, properly. A model that watches video and reads images tags each creative on a fixed list: hook, format, persona, offer, angle, call to action, creator and length.
• Tags joined to money. Every attribute carries its spend, CTR, CPA and ROAS.
• A weekly brief for the creative team. Which attributes are winning, which ads show rising frequency and falling CTR before Meta flags them, and three gaps worth briefing next.

What it can't do

• Patterns aren't proof. Meta decides who sees each ad, so an attribute that wins may be riding a better audience.
• Small spend is noise. Ads below a spend floor are left out of the read.
• Tags need checking at first. We audit the first 50 to 100 by hand before the team relies on them.