Meta wants your ad sets combined. Keep the comparison you split them for

In Meta's own worked example of opportunity score, fixing an audience fragmentation issue may improve the score by 35 points, against 5 points for applying an Advantage+ placements recommendation. The same page says the score, including a high score, does not itself reflect your actual or future performance.

This post is our argument about what to do with that, built on Meta's own help pages, and the parts that are ours are labelled as ours.

Meta also says you should not turn off campaigns with a low opportunity score, because the score represents how well optimised a campaign is, not whether it is performing poorly, and that you should judge performance based on your business goals.

Our reading: the score is telling you Meta would combine your ad sets, and telling you on the same page not to read that as a verdict on how they perform.

Meta's case starts with audience size. Its example is three ad sets with the same objective and creative, one for people interested in dogs, one for dog toys and one for dog collars, and it says splitting them could decrease performance because the total audience size is split across multiple ad sets.

Then learning. Meta says that when you run too many ad sets at the same time, each one gets fewer opportunities to learn and therefore fewer results, so ad sets may spend more time in the learning phase.

Then overlap. When two or more ads from the same advertiser enter the same ad auction, Meta says it chooses the ad with the highest total value to compete, and the advertiser's other ads are not considered in that auction. That ensures your ads will not bid against one another, but Meta says it can prevent an ad set from spending its full budget or achieving enough results to exit the learning phase. Meta says running ads from separate ad accounts may not help you avoid overlap, and that auction overlap can occur across these accounts.

Our reading, not Meta's: almost nobody fragments an account by accident. Three ad sets for three interests, or one per country, or one per age band, were set up to answer a question, which of these works, and the separate rows in Ads Manager were the answer.

Meta's own advice is aimed at exactly that habit: instead of fragmenting audiences for reporting purposes, it says, consider using breakdowns in ads reporting.

The breakdowns come with conditions, on Meta's own breakdowns page. Conversions it isn't able to attribute are shown as Uncategorised. For web and iOS app offsite conversions, it supports nine breakdowns only: age, gender, age and gender, country, impression device, platform, platform and device, placement and audience segments. And if you used Advantage+ audience suggestions for age, gender or detailed targeting, Meta says it will show ads to people matching your suggestion and other audiences when it's likely to improve performance.

Our reading: check that the breakdown you need exists for your conversion event before you merge, not after. If the dimension you split by is not one Meta can break your conversions down by, merging gives up a comparison you cannot get back.

Riibon's audit rule is that a recommendation which applies to more than one entity is checked one level below the level it is made at, and has to hold in every bucket before it is rolled out. When a merge removes the ad set as a level, we do not report the campaign as unsegmentable. We move the check to placement, demographic and creative breakdowns. That is our method, not Meta's.

Meta's own rules shape the merge itself. Broadening one ad set to absorb the others is a change to targeting, and Meta lists any change to targeting as a significant edit, the kind that causes an ad set to re-enter the learning phase. Where Meta suggests turning off overlapping ad sets instead, it suggests considering the ones that are learning limited or have the fewest results, and moving their budget to the active ad set to maintain a similar or greater number of optimisation events. A change to budget amount, Meta says, may or may not be significant, depending on the magnitude of the change.

So, in our practice, each ad set's results are written down by the breakdowns that will survive before the merge, and the merged ad set is read only after it has left learning. Meta says that during the learning phase your results aren't necessarily indicative of future performance, so comparing a merged ad set's first week with the last month of three settled ones is comparing a number Meta says is not necessarily indicative of future performance with one it does not caveat that way.

Our reading again: when the question was never which segment did better, but which audience strategy wins, neither split ad sets nor breakdowns settle it. Meta's A/B test shows each version to a segment of your audience and ensures that nobody sees both, and that is the tool for a verdict.