It seems that every day brings a new feature or upgrade from Facebook Ads Manager, and upon logging in today – after last checking on the account Friday – I was greeted with an in-app message informing me that our account would now feature the “Learning” label in the status field for newly active ads, a feature that was first seen in September of last year but, like many of Facebook’s updates, is presumably still rolling out to advertisers.

This label, according to Facebook, signifies that your ad is still in what they call the “Learning” phase of the ad delivery process. The label itself doesn’t signify any actual change to Facebook’s delivery method – it is simply another way for advertisers to follow the progress of their campaigns more precisely.

What Is The Facebook Ads “Learning Phase?”

This “learning phase,” which presumably has always existed in Facebook’s algorithm, is the period of time during which Facebook gathers data about how users are responding to your ad, so it can best optimize for ad delivery. As their documentation explains, this helps stabilize performance and costs over time, as ad performance will naturally fluctuate for a while until the “right” users are consistently shown the ad most relevant to them.

What We Know About Automated Ad Delivery & Optimization

With any type of paid ad campaign, it’ll take a period of time to gather the data needed to optimize for best results. In the days of purely offline advertising, that could take a very long time, and it was difficult if not impossible to obtain a completely accurate picture of performance (with direct engagement, usually a sale, or anecdotal evidence as the only “metrics”).

In a platform like Google Ads, of course, data is accumulated almost instantly – and this applies to nearly any other type of campaign that can be tracked within Google Analytics. But the process of using this data to optimize for better results can be fairly manual at times, since advertisers need to make judgments about which ads are performing best based on the key performance indicators (KPI’s) most valuable to them. (There are of course some automated features involved, with Google optimizing ad delivery based on flexible bidding strategies, etc. – and with Google’s recent investment in machine learning, we can expect the role of automation in Google Ads will only increase with time.)

How Facebook User Data Helps Improve Ad Performance

What sets Facebook apart here, though, is the degree to which its algorithm actively optimizes ad delivery during the course of a campaign, finding the users that “perform” best, not just the ads that resonate across a given audience. With most other platforms, that just isn’t possible – you have to set your audience parameters and see what you get once you’ve run the campaign for a while.

But because Facebook has such a (notoriously) vast amount of information about users’ interests and behavior on their platform, their ad platform can go a step further, building on the rules you set and finding the exact patterns and related behaviors or interests that correlate with a strong response to your ad.

How Does Facebook Ads’ “Learning Phase” Work?

Facebook will still serve the best performing ad, and will optimize for ad delivery in other ways as well, just as platforms like Google Ads do. But as time goes on, the data you accumulate (or rather, that Facebook collects) about your audience’s responsiveness will produce a not-entirely-transparent refined audience that will almost certainly perform better than the original groups you targeted – or at least, is most likely to based on the available data.

And while you can’t necessarily “see” this audience – it would be difficult to do so without getting dangerously close to user-identifiable information – you can get some idea of its selectivity by using the Lookalike Audience feature, which chooses a new audience based on who responded the most to your previous ad set. This is almost certainly using the same set of data about performance, and you may be surprised just how much better your ads perform, even with the same creative, when delivered to a Lookalike Audience.

How Long Will The Learning Phase Last?

As with any data-based solution, these sorts of decisions must be made with an adequately large (and ideally diverse) sample in order to produce the most accurate results. For this reason, Facebook specifies that the duration of their learning phase will be determined according to one of two criteria: either the results after the first 50 “optimization events” – that is, the objective of your campaign, such as conversions or clicks, or after “enough time passes” without reaching the 50 event mark. (Their documentation also does add that the number here is “general guidance” and may actually vary between ad sets based on specific attributes and markets.)

In the latter case, Facebook suggests pursuing some additional manual optimization changes, because achieving fewer than 50 events typically indicates poor or at least below-average performance. These could include changes to targeting, placements, or creative.

What Changes Will Affect Facebook’s Learning Phase?

Being aware of this stage in the ad campaign process might make advertisers reluctant to make any modifications to their ads, lest Facebook decide to “reset” the learning phase based on the new information. This is not entirely unwise, as Facebook does say that edits they consider “significant” will actually cause the learning phase to be reset. These include any changes to targeting, creative, or the selected optimization event. (Pausing the ad set or campaign will also reset the phase.)

However, less significant edits should not trigger a full reset of the learning phase – though Facebook may slightly increase the sample size threshold before concluding it. Those less significant edits include bid cap, target cost amount or budget amount. (It’s worth noting that the “significance” of these changes will depend on the magnitude of them – so a massive increase in budget will probably trigger a reset, because it pretty significantly changes how your ads will be delivered and in what quantity.)

In Conclusion

Ultimately, this change doesn’t mean too much as far as the performance or operation of Facebook’s platform is concerned, but it does shed some interesting insights both on how the algorithm measures performance and what best practices they recommend following before making any decisions based on data. As the Ads Manager adds more and more features over time, we can most likely expect to see a lot more functionality come about from machine learning – and what that means for the future role of advertisers.