Meta gave one ad most of the spend. That is not a test result

If you have multiple ads in your ad set, Meta says it will show the ad that's most likely to achieve the lowest cost per optimisation event for the given person, and that each of your ads won't necessarily be delivered the same number of times.

Meta also says that sometimes the ad or ad set that gets the most results isn't the one that received the lowest cost per optimisation event, because the delivery system uses predictions of future performance to determine where to deliver next, not each ad set's past performance.

Our reading, not Meta's: the split of spend between your ads is the system's forecast, made before the results are in. An ad that got a tenth of the delivery has not been shown to lose. It has been predicted to, and the prediction behind it rests on less delivery than the prediction behind the ad that took the budget.

Meta says that each time an ad is shown, the delivery system's predictions of relevance become more accurate, and that the more an ad is shown, the better the delivery system becomes at determining where, when and who to show it to.

Its ad relevance diagnostics, the quality, engagement rate and conversion rate rankings, aren't available for ads with fewer than 500 impressions.

Our reading: an ad with little delivery has a thin prediction behind it and, under 500 impressions, no ranking to read either. Calling it a loser from its own handful of results is reading noise.

Meta says that when an advertiser runs too many ads at once, each ad delivers less often, which means fewer ads exit the learning phase and more budget is spent before the delivery system can optimise performance. In its words, too many ads can result in worse performance.

Its fixes start with fewer ads, not less creative. Meta recommends decreasing ads per ad set but maintaining diverse creative assets per ad set: one ad can contain multiple, up to ten, creative assets. It says one ad with multiple text optimisation is usually more effective than multiple ads without it, and that dynamic creative lets the delivery system find the best creative to deliver for a given user while using just one ad.

Meta's ad volume page names no number of ads per ad set, so we use our own. In the audits we run, Riibon flags a campaign as fragmented when it has more than eight active ads and fewer than a quarter of them have passed 1,000 impressions. Those thresholds are Riibon's working rule, not Meta's, and stricter than the 500 impressions Meta needs before it shows a ranking. Our fix is to keep the three ads that have taken the most spend, so each gets enough delivery to judge, or to fold the variants into one ad and let Meta's own creative tools choose between them.

Meta does not recommend testing informally, such as by turning ad sets or campaigns on and off manually, and says this can lead to inefficient ad delivery and unreliable test results.

Its A/B test shows each version to a segment of your audience and ensures that nobody sees both. Meta recommends using the same budget for both versions to ensure a fair comparison, and says the test can then measure each strategy on a cost per result basis or a cost per conversion lift basis.

Adding a new ad to your ad set is on Meta's list of significant edits, the changes that send an ad set back into the learning phase.

Our reading, not Meta's: dropping a challenger into a live ad set buys two problems at once. The ad set goes back into learning, and the challenger then competes for delivery against ads the system has already seen more of, so the split you get back is another forecast. If the question is which creative sells, put it to an A/B test with equal budgets and judge it on cost per result.