When should we remove an image, video or text version from a multi-asset Meta ad, now that Ads Manager reports results for each asset?
Our answer
Don't cut a version from a multi-asset ad for low spend alone. Cut on cost per purchase once it has about 50 purchases, and swap idle versions for stronger, distinct ones in one edit.
- Confidence
- Low
- Strongest proof
- 1 controlled test
- 3 Meta rules, 4 practitioners, 1 opinion or claim
- Sources
- 6 sources
- 9 statements
- M
PALRPG
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentShare of spend shows Meta's delivery, not a fair test. Meta describes the Media breakdown as how each image or video was delivered. In a live campaign, Pal and Susarla measured that Meta's delivery shifts who sees an ad in response to its text.
Low spend is not proof a version is weak. It may only mean Meta deprioritized it, so Common Thread Collective waits for at least 50 purchases before cutting on cost per purchase alone. Webtopia makes the same point about low impression share.
Swapping versions has a cost, so do it once. Webtopia replaces idle slots when one asset takes the bulk of impressions, and Meta's rules count any change to ad creative as a significant edit that can send an ad set back into learning.
Our judgment
- Treat a version's share of spend as Meta's choice, not a verdict, and don't remove a version for low delivery alone. This combines Meta's description of the breakdown, Pal and Susarla's experiment and the warnings from Common Thread Collective and Webtopia.
- Wait for about 50 purchases rather than using the faster rule of cutting at $500 spent and a cost per purchase 2 times the campaign average. Common Thread Collective's own page says low-spend assets often lack statistical significance, and Webtopia advises judging concepts across several ads before writing them off.
- When Meta puts the bulk of impressions on one or two versions, replace the idle ones with stronger, distinct versions rather than only deleting them, and give a new concept its own ad (Multiple images or videos per ad). Webtopia advises replacing and rotating rather than deleting, and in Nick Boddington's ad Meta delivered one of about eight assets.
- Make removals and swaps on one ad in a single edit, not one at a time. Whether removing one version counts as a change to ad creative is our reading of Meta's page, not something Meta states.
Our judgment, built from Meta, Pal and Susarla, Common Thread Collective, Webtopia and Nick Boddington: our reading of these sources, not a finding from any one of them.
What the sources say
9 statements from 6 sourcesCommon Thread Collective says Meta's per-asset breakdown reflects delivery choices, not a fair split test, so wait for at least 50 purchases per asset before removing one on CPA alone.
“This reflects Meta's delivery optimization, not a fair split test.”
Common Thread Collective says Meta now reports each image or video in multi-asset ads, but the split reflects Meta's delivery choices, not a fair test. It suggests cutting assets over $500 spent at 2x CPA before BFCM.
“Low-spend assets often lack statistical significance”
Meta's Ads Manager can now break down results by each image or video and each text variation in a multi-asset ad, but not for collection or carousel formats or some conversion elements.
“If you select Media, you can see the performance of how your images and videos delivered as single media”
Meta's Insights API offers a creative_automation_asset_id breakdown to group results by the image or video delivered. Dynamic Creative asset breakdowns support only a few metrics, like impressions, clicks and spend.
“Use this breakdown to group results by the image or video that was delivered.”
Meta says targeting, creative or optimization event changes and adding a new ad are always significant edits, while a budget change may or may not restart learning, depending on its size. No safe percentage is given.
“if you increase your budget from $100 to $101, that isn't likely to cause one or more ad sets to reenter the learning phase.”
Nick Boddington says adding multiple creatives to one ad unit lets the Breakdown tool show performance for individual images and videos.
“we've got that intelligence.”
Webtopia, an agency, says if one asset takes 80 to 90% of impressions in a Meta multi-media ad, replace the others, but low share alone doesn't prove an asset weak. Keep single-asset test cells for clean reads.
“Low impression share is not proof an asset is weak, it may simply have lost early auctions before gathering enough data.”
Phil Graham says to change only one variable at a time when reworking an underperforming ad, so you know what caused the result.
“I only want to have one variable in my change so that I know whatever the result was, the change I made caused that to happen”
A Meta field experiment by Pal and Susarla found the algorithm alone shifted female impression share by 2.27 points, so two-arm creative tests mix up a creative's effect with Meta's targeting.
“a conventional two-arm test would attribute the entire delivery shift to the creative.”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer