2022 Guide to Facebook’s Learning Phase

In Social Media, Paid Media3 August 202212 Minutes

If you’re looking to build your business using social media, there’s no larger audience than Facebook’s, which boasts 2.93 billion monthly active users.

When launching an ad campaign, initial performance can fluctuate once it has been pushed live. Campaigns go through this instability as Facebook collects data to improve performance, which is known as the Facebook learning phase.

This guide has been created to help you understand why campaigns enter the learning phase, and how you can speed up the process to drive the results you are looking for from your social media marketing campaign.

So, what exactly is the Facebook learning phase?

When creating a new ad set or making a significant edit to an existing ad set, the initial period of delivery is known as the learning phase.

While Facebook emphasises that the delivery system never stops learning, the learning phase is generally the period when the bulk of the learning about an ad set is done.

During the learning phase, Facebook is establishing how to best deliver the ads within the ad set in question. Facebook experiments with the different aspects of ad delivery: who within your audience to deliver ads to, when to show your ads and which placements to show your ads in. Facebook warns that performance can be unstable in the learning phase, meaning the CPA, CPM and CPC can fluctuate.

Facebook is testing to see which combinations of delivery criteria most frequently produces the event the ad set is optimised towards. This can be anything from purchases, link clicks or landing page views to other available pixel events such as registrations, add to cart and initiate checkout.

Working out the most effective delivery criteria for achieving the desired ad set event ultimately means optimised ad sets. However, to achieve this, Facebook requires each ad set to achieve enough of the chosen event to gather enough learnings.

How long does the learning phase last?

The learning phase ends when 50 of the chosen optimisation event have been generated; Facebook requires 50 to be achieved within 7 days of launch to complete learning.

For example, if the optimisation event is purchase, 50 purchases will need to take place within a 7-day window for learning to complete. The same rule applies for other optimisation events, such as link clicks or landing page views: 50 events within a 7-day period.

If the ad set fails to generate 50 optimisation events within the learning window, the ad set will end up with learning limited delivery.

During the learning phase, edits to the campaign, ad set or ads themselves should be avoided. By implementing an edit, the learning phase is reset, delaying the delivery system’s ability to optimise.

Why does the learning phase matter?

To put it simply, if the learning phase isn’t completed (resulting in learning limited delivery), Facebook’s delivery system has not learnt how to optimally deliver the ads within the ad set.

It’s worth noting here that, while it’s possible to generate good results from campaigns with learning limited delivery, the chances are that performance will be worse than if the learning phase was completed.

The implication of learning limited delivery is lower performance, likely in the form of higher CPMs (cost per 1,000 impressions) and CPAs (cost per acquisition).

Getting campaigns out of the learning phase means that the delivery system has gathered enough data to deliver ads in the most optimal way.

What is learning limited?

Learning limited is the diagnosis applied to an ad set when the learning phase isn’t exited. It is applied when 50 optimisation events aren’t recorded during the learning phase period.

Facebook underlines that learning limited delivery isn’t a “penalty”, but rather a way of indicating that budget being used by the ad set isn’t being spent effectively, as the current setup within the ad set is hindering the system from delivering optimised performance.

An ad set becomes learning limited when it fails to record 50 optimisation events in the 7-day learning phase window.

As mentioned, it’s possible to generate good results from campaigns with learning limited delivery, but the chances are that the performance will be worse than it could be.

Re-entering the learning phase: what is a ‘significant edit’?

Edits to the campaign, ad set, or ads will trigger re-entry into the learning phase, resulting in any learnings around delivery gathered by Facebook being overwritten. These are known as significant edits.

Edits that will move an ad set back into the learning phase include the following:

Campaign level edits
  • Budget (increased/decreased by more than 20%)
  • Bid amount
  • Bid strategy
Ad set level edits
  • Targeting
  • Placements
  • Optimisation event
  • Creating a new ad
  • Bid strategy
  • Bid amount
  • Budget (increased/decreased by more than 20%)
  • Pausing the ad set for over 7 days or longer
Ad edits
  • Any change to the ad creative (visual, ad copy, headline, description, call to action, landing page URL)

Significant edits during the learning phase should be avoided, if possible, as restarting the learning phase can mean a longer period of unstable performance with the potential for worse results.

If the learning phase has been successfully completed, a best practice is to make any future changes in bulk. As just one of the changes will trigger re-entry into learning anyway, it’s best to make all changes at once so that the ad set can be left alone to go through the learning period again.

If the learning phase has resulted in learning limited delivery, making a significant edit will re-enter the ad set into learning, restarting the process, and hopefully starting the journey to completed learning and active delivery.

Getting stuck in the learning phase

Ad sets can sometimes end up continually switching between the learning phase and learning limited delivery in a cycle, as they are unable to achieve the required number of optimisation events to exit the learning phase.

In most cases, it will be ad sets optimised towards more significant actions such as purchases or add to carts, rather than actions that happen in greater volume such as link clicks and landing page views.

Typically, ad sets that get stuck in the learning phase cycle are those targeting a conversion action as the optimisation event (purchases, registrations etc), with the conversion not happening frequently enough to reach the requisite 50 optimisation events. This can be the situation for products or offerings that don’t return high volumes of conversions.

Ultimately, the learning phase cycle is often an unresolvable situation if conversion ad sets are a necessary part of the overall Facebook marketing strategy.

The effect of changes to Facebook audience targeting on the learning phase

The Apple iOS 14 update, launched in 2021, has resulted in less third-party data available to build Facebook custom audiences (all audiences based on pixel events), as well as the potential for performance reporting limitations. This is due to the update allowing users to limit the data being recorded about their activity.

In a lot of cases, these changes have resulted in the reduced effectiveness of audience targeting based on third-party data. In practice, this has meant traditional remarketing campaign structures have seen decreasing returns and lower performance.

The knock-on effect of this is that, with smaller and less effective custom audiences, campaigns can be less able to achieve 50 optimisation events, resulting in more learning limited campaigns.

The changes resulting from iOS 14 have meant that the audience targeting and optimisation events used within certain campaign types have had to be rethought.

Tips for exiting the learning phase

When reviewing campaigns continually cycling between learning and learning limited, the 2 reasons we most frequently see are insufficient audience sizes and budgets. These factors ultimately provide ad sets with too little to work with, meaning they don’t reach the 50 optimisation events within 7 days.

Getting out of the learning phase means providing ad sets with enough audience volume and budget means to achieve 50 optimisation events in 7 days. It also means optimising campaigns towards an event that is likely to happen at least 50 times in 7 days.

With this in mind, here are our top tips for exiting the learning phase:

Change your optimisation event

Consider choosing an optimisation event that occurs earlier in the conversion journey or more frequently. For example, moving from purchases to add to basket, initiate checkout or even page view.

This is our top recommendation. In practice, it’s often the case that campaign budgets are set in stone and audiences can’t be increased in size, meaning changing an ad set’s optimisation event is best route to generating enough events to exit the learning phase.

Expand the ad set audience

The larger the audience, the more opportunities for different people to complete the optimisation event. Audience expansion can be achieved by expanding the audience itself or by combining ad sets/campaigns.

The challenge here is that expanding an audience can often make it less targeted, which could end up being a hindrance to achieving 50 optimisation events. Certain audience types (e.g. website visitors to a certain page) can’t simply be expanded.

Increase budget, raise bid or cost control

A budget can be too low to generate enough traffic to reach 50 optimisation events, meaning the ad set won’t exit the learning phase.

More spend means more traffic, resulting in more opportunities for the optimisation event to be completed.

As with audience expansion, increasing budgets isn’t always possible, in which case – consider raising the bid control.

This can help because if bid or cost control is too low to receive 50 optimisation events, the ad set won’t exit the learning phase.

Facebook’s learning phase should now be something which you know how to tackle. However, if you’re having issues with it, or your social media marketing in general, get in touch today to see how we can help.