It’s common for advertisers to micromanage their ads, turning off the versions that get the worst results to leverage top performers. The thought is that by forcing the algorithm to only show the best ads, you’ll be more efficient.
Here’s what I think…
Countering the Algorithm
Meta’s delivery algorithm updates in real time to get you optimal results. A primary ad version will be selected quickly, and the other versions will get fewer impressions.
But that doesn’t mean those versions are worthless. If Meta still shows them, they are being shown for a reason. They may still contribute to your results, even if they don’t get direct credit.
Distribution Changes and Ad Fatigue
Also, the distribution of your ads isn’t fixed. Just because the algorithm prefers one ad this week (and it is getting the best results) doesn’t mean that the trend will continue. The rate at which one version is shown over another can change from week to week.
This also helps prevent creative fatigue. Your top performer will eventually get overexposed. Not only might another version overtake it, but simply having variations shown can help prevent that main ad from becoming noise.
Exceptions and How the Algorithm Works
That said, there are always exceptions. If you spot that a large percentage of your budget is spent on a poorly performing ad while ignoring a better performer, sure. Turn that bad ad off and see if you can get better results.
But this technically shouldn’t happen if the way you measure success is consistent with the performance goal. Meta’s delivery algorithm is constantly making updates with the focus of getting more of what you want. Meta should favor the ads that satisfy that goal — assuming there’s an adequate sample size to prove that.
Stopping ads to force the algorithm is like only using news feed placements for conversion ads. Sure, those may be the most effective placements, but that doesn’t mean that the other placements aren’t helping. Doing this often leads to worse results.
My Advice
Don’t micromanage your ads as a rule. Any of these adjustments that counter the algorithm, particularly when the performance goal accurately defines what you want, should be the exception.