The best way to fix a limited learning phase is to give Meta's algorithm enough room to work, which means simplifying your account structure and consolidating your spend. The goal is to get at least 50 conversions per ad set per week so the algorithm can stabilize and find your best customers consistently. When an ad set is stuck, it's usually because it's starved for data, budget, or creative options. Here’s a playbook to fix it.
- Consolidate Your Ad Account. The single biggest cause of “limited learning” is having your budget spread too thin across too many campaigns and ad sets. The team at Common Thread Collective on an episode of Ecommerce Playbook mentioned that Meta is pushing for more consolidation, and they're moving away from the hyper-segmented structures that used to be popular. Combining ad sets reduces complexity and pools your spend, giving each remaining ad set a much better chance of hitting the 50-conversion threshold needed to exit the learning phase.
- Stop Separating “Testing” and “Scaling.” The old method of running separate 'testing' and 'scaling' campaigns is a major reason ad sets get stuck. The team at Kynship explained on The Bottom Line: Ecommerce Tactics for Profitable Growth that when you move a winning creative from a testing campaign to a scaling campaign, you reset the learning on both. This is a terrible outcome that puts you in a permanent state of learning. Instead, Paul Lynch detailed their approach of having one core campaign where new creative is added to ad sets that have stopped spending, turning them into your new testing ground without disrupting your top performers.
- Go with Broad Audiences. Narrow interest and lookalike audiences can severely restrict your ad set's ability to get conversions at scale. On The Bottom Line, Cody Plofker made the point that Meta's algorithm is so advanced that it now treats these audiences as mere suggestions anyway. Using broad targeting gives the algorithm maximum flexibility to find your customers. If your ads are compelling, as the Hammersley Brothers point out, Meta will find the right people; if they are boring, you'll just get shown to the same warm audience repeatedly.
- Increase Your Creative Volume. Your ad set might be stuck in learning simply because the creative isn't good enough to generate 50 conversions a week. The solution is to give Meta more options. Taylor Holiday from Common Thread Collective made this point on Ecommerce Playbook, explaining that giving Meta a broader array of ads allows it to experiment more and find a winner faster. Instead of just three ads, try loading the ad set with 10 or 20 diverse creatives to increase the odds that one of them will resonate and scale.
- Set a Realistic Budget. The learning phase requires a certain volume of data, and that costs money. You need to ensure your budget is high enough to actually achieve 50 conversions. For example, if your target Cost Per Acquisition is $40, your ad set needs a budget of at least $2,000 per week (or about $285/day) to have a fighting chance. As Kurt Bullock explained on Honest Ecommerce, you have to be willing to spend a certain amount to validate an ad, like spending $300 to test a $100 product.
- Re-evaluate Your Cost Controls. If you're using bidding strategies like cost caps or ROAS floors, they might be too restrictive. Aggressive cost caps are a common culprit for why ad sets fail to spend their budget and get stuck. Cody Plofker put it bluntly on an episode of The Bottom Line, saying that if you set controls at a level you aren't even hitting on your best campaigns, the new ad set has no chance to spend. You may need to ease up on your controls or remove them entirely until the ad set is out of the learning phase.
The one thing you absolutely must avoid is manually turning off ads that are spending money or making frequent, significant edits to the ad set. As Cody Plofker often says, you are competing with Meta's algorithm, which has vastly more data to predict performance than you do. Trust the machine, give it the inputs it needs, and be patient.
