What is the Meta learning phase?
What happens during the learning phase?
You launch an ad set. Meta's system does not yet know which people are most likely to respond, so delivery is still settling. Meta shows this as a status on the ad set: learning, exited, or not getting enough results to exit.
In our view, the practical point is simple. While the ad set is still learning, its numbers tell you less about whether the ads work. Judge them with that in mind.
Meta sets the exit conditions. We do not quote numbers here. Ask for the current ones for your account.
Does a new ad reset it?
Meta says significant edits send an ad set back into learning. Whether adding one new ad counts is disputed. One practitioner's blog post questions that rule after testing it on his own account1. A blog post is not Meta's documentation, so treat it as a reason to ask, not as an answer.
For ZFD: Our view is that the useful question is not "is it learning?" but "what are we deciding while it learns?"
Easily confused with
Learning limited
A related status, not the same phase. See the FAQ below.
A drop in an ad's results after the audience has seen it too often. Different problem, different fix.
The rules a platform uses to set bids, from manual bids to automated targets such as target ROAS or target CPA.
Meta's automated campaign type for sales, in which machine learning manages audience, creative, budget, optimisation and placement.
Meta's system for the retrieval stage of ad recommendation, which narrows a very large pool of candidate ads to a shortlist.
The channel choices around all of this are covered in paid media strategy for ecommerce.
Buzzword alert
"It's still in the learning phase"
Usually means: results are poor, and the explanation is waiting for the calendar.
Ask: what will we change if results are still weak once learning ends?
"We can't touch it, or it will reset"
Usually means: a rule of thumb, applied to everything.
Ask: which edit exactly resets it, and where is that written down?
FAQ
How long does the Meta learning phase last?
It ends when Meta's own conditions are met, not after a fixed number of days we can promise. Ask which condition the ad set is missing.
What does learning limited mean?
It is the status for an ad set that is not getting enough results to exit learning. Ask what is holding it back and what will change.
Is the learning phase bad?
No. It is how the delivery system starts. It becomes a problem when it is used to explain every weak week.
Can I judge ads during the learning phase?
You can look, but with care. In our view, results from an ad set that is still learning are a weak basis for a decision.
Sources
- New Ads and the Learning Phase (Jon Loomer)Blog or media
Let’s find what’s actually working.
Next step. Check what a platform-reported number means for your margin with the Is your ROAS real? calculator.