The Learning Phase Is Just a Label

So I had a bit of a breakthrough moment recently, and it was related to what the learning phase is and possibly isn’t. It’s got me rethinking all of this. Let me explain.

First, just a quick summary of how Meta defines the learning phase in the first place. When you publish a new ad set, it will typically begin with the “Learning” label in the delivery column. That means that the algorithm is trying to sort things out — which ads to show to which people, and all the many layers of complex details that go into all of that.

While learning, Meta says your results will be less stable. You could see wild swings in performance from day to day, and this is what generally freaks advertisers out. We don’t like instability and a lack of predictability.

To exit the learning phase, advertisers generally go with the benchmark of 50 optimized events or conversions in a seven-day period. Originally, Meta set this as a hard rule. You would not exit the learning phase until those 50 conversions happened, and if you didn’t get them in seven days, you’d be stuck in “Learning Limited.”

Of course, exiting learning and going into “Active” didn’t guarantee anything either. If you make significant edits, you can send an ad back into the learning phase. What qualifies as significant edits has evolved over the years, but it’s generally been things like publishing new ads, making changes to ads, updating targeting, or significantly increasing the budget.

None of these things are black and white. You can increase the budget more than you could in the past without reentering learning. In fact, if an ad set is classified as high performing — and Meta will notify you of this — you can increase the budget quite a bit without triggering learning again.

Advertisers debate all the time about whether any of this matters. Yes, you want to exit the learning phase to get optimal results, but the inability to exit learning isn’t a death sentence for your ad set either. There may be times when it’s simply not reasonable to get 50 or 25 or whatever number of conversions in a week to satisfy the algorithm.

So we’re already at a point where the learning phase is more of a guide than a hard and fast rule. But something happened recently that has me rethinking it even more.

Here’s what happened. I had an active ad set that was doing really well, but there was a new ad that I really wanted to create. I knew that if I published that new ad to the existing ad set, it would—or should—restart the learning phase. At the same time, I don’t fear restarting the learning phase, so I decided to publish the ad anyway.

Based on Meta’s documentation about significant edits, this was supposed to trigger learning again. But to my surprise, publishing that ad did not restart learning.

At first this was a pleasant surprise, but it inspired me to start thinking about what the learning phase actually is and what Meta’s labels even mean.

Because there are two realities when I publish a new ad. First, I want Meta to keep learning at this point. Otherwise, wouldn’t Meta refuse to show that new ad? How else could the algorithm get the data it needs to know who to show that ad to?

And that underscores another important point that we already know about Meta ad delivery — the algorithm is always learning. It might just not be called the learning phase. Whether it’s due to competition, changes in the audience pool, feedback and reactions to your ad, or fluctuations related to your website or your page, the algorithm is always learning and always adjusting. And we want that.

Because of that, does the “Learning” label really mean anything? Does it just reflect that Meta detected more instability in performance than normal? Or is it possible that the label doesn’t reflect anything particularly significant?

It’s just a label, after all, and advertisers have historically run in fear from it. I’d argue that part of the confusion comes from the fact that there’s a “Learning” delivery label in the first place. It suggests that the ad isn’t active, but it is. And when it switches to “Active,” it suggests that it’s no longer learning — but the algorithm is always learning.

If Meta is listening, I’d recommend that the delivery column always display “Active” if the ads are actively delivering. If there’s a problem related to instability or results, create a separate column for that.

So here’s the bottom of the glass. Advertisers have historically run in fear from the learning phase. See it as just a label. You can get good results while learning. The algorithm is always learning anyway.

Its role has mostly been to let you know that things might not be running optimally right now. But let the results tell you that more than anything. And don’t fear making a change that might restart learning. Don’t constantly tinker, but don’t avoid important changes either.