Meta Pixel Learning Phase Explained

In this tutorial, we will take a look at everything you need to know about Meta ads learning phase.

The primary goal when running a Meta campaign is to exit the learning phase.

The quicker you achieve this, the more benefit you see from a performance perspective, because this means that Meta has “learned” what are the common characteristics between the people within the audience that who are likely to convert. Consequently, it can prioritise delivering future ads to people that match those characteristics.

Let’s start at the beginning!

Starting the learning phase #

According to the official documentation,

When we start delivering our ad set, whether at the start of a campaign or after you edit it, we don’t have all the data necessary to deliver it as stable as possible. In order to get that data, we have to show ads to different types of people to learn who is most likely to get you optimisation events. This process is called the learning phase.

Meta Business Help Centre

In detail:

1. The learning phase is the period during which the Meta delivery system gathers data and experiments with different delivery options to understand how users respond to your ad set and optimise its performance based on factors like:

   • What time of day produces the best results
   • What day of the week produces the best results
   • How many times showing your ad produces the best results
   • Which ad placement produces the best results
   • Which ad creative produces the best results
   • And more!

   All these various elements, combined with your chosen objectives, go into working out how to get you the best possible results.

2. The learning phase begins when you create a new ad or ad-set or make a significant edit to an existing one.

   Find here the list of what is considered significant edit in Meta.

   Note: You cannot opt-out the learning phase process, regardless of the campaign objective, however Campaign Budget Optimisation or broader targeting can potentially shorten or streamline it.

3. CPA performance volatility during this period

   Before you panic and set off any alarm bells, keep in mind that unstable results can be completely normal during the learning phase, as the Meta delivery system is exploring the best way to deliver your ads – actively trying different audiences, placements and more.

   The more an ad is shown, the better the delivery system becomes at optimising the ad’s performance, so this learning process should lead to a stable results for the most efficient cost.

Exiting the learning phase #

According to the official documentation,

Ad sets exit the learning phase as soon as their performance stabilises. Typically, performance stabilises after an ad set receives around fifty optimisation events since its last significant edit.

Meta Business Help Centre

In detail:

1. The optimisation event depends on your campaign objective.

   When you select a campaign objective you need to tell Meta WHICH conversion event you want.

   

   

   If your campaign objective is conversion and the conversion event is purchase, then the ads will have to generate 10 sales in a week to exit the learning phase.

   Of course, for the conversion campaign and catalog sales campaign objectives it is a lot harder to exit the learning phase. Because, it’s harder to get 10 sales than to get 10 people to your website or 10 video views.

   So, choose your conversion event wisely:

   • If you’re an eCommerce business, you’ve selected the Purchase event, and you’re not getting enough optimisation events, try changing the event to one that’s back a stage or two in the checkout process, such as Initiate Checkout or Add to Cart.

   • If you’re a service-based business and you’re generating Leads or Subscribers, you could create a ‘View Content’ event on button clicks or landing pages to increase the number of conversion events.

2. The threshold is 10 conversions within 3 days (previously was about 50 in a week).

   Jul 2024 Update: Meta (Facebook and Instagram) introduced a new threshold. Instead of requiring 50 conversions in a week, advertisers now need just 10 conversions within 3 days to exit the learning phase.

   This is much easier to achieve and allows advertisers to work with smaller budgets. Although the timeframe is shorter, the lower number of conversions makes it simpler for advertisers to optimize their campaigns.

   You can track your learning phase progress in the delivery column of the Ads Manager.

   If you hover over the “ learning” text, you can see a progress bar showing how long it’s been since you made a significant edit, as well as how far you are from reaching the 10 action threshold.

   

Previously was:



The 50 optimisation event is only a general guidance, the actual number required could vary based on its specific characteristics and/or the market conditions at the time it’s running.

I’ve had conversion campaigns that at 30 conversions I’ve been able to exit the learning phase and I’ve also had campaigns that had over the 50 conversion threshold and haven’t exited.

So, it also depends on the quality of the signals which that’s at facebook’s algorithm discretion.

Don’t forget about your attribution window.

If you choose a one-day click conversion window that means that if the conversion happends 3 days after clicking on your ad, it won’t be counted in the optimization data whereas if you choose a seven-day window will be counted, so setting to 7 days click facilitates the successfully exit the learning phase.

After exiting the learning phase #

Once the learning phase has ended, your ad sets can either go:

1. From Learning To Active

   The delivery status “Active” means the ad set has gathered enough data to effectively optimise based on your chosen objective. You can expect more stable performance and potentially improvements in efficiency.

2. From Learning To Learning Limited

   The delivery status “Learning Limited” means the ad set has not reached the 50 optimisation events (or encountered an obstacle preventing it from learning effectively).

   According to the official documentation,

   If your ad set isn’t getting enough optimisation events to exit the learning phase (or if the delivery system predicts that it won’t receive enough optimisation events in the future), the Delivery column reads “Learning Limited.”

   Meta Business Help Centre

   Learning Limited isn’t a penalty – it’s an indication that your budget isn’t being spent effectively because the ad delivery system can’t optimise performance with your current setup.

   An ad set becomes Learning limited when it is unlikely to receive around 50 optimisation events in the week after your last significant edit.

   Generally, an ad set becomes Learning limited when the ad set is limited by small audience size, low budget, low bid or cost control, high auction overlap, an infrequent optimisation event or other issues such as running too many ads at the same time.

   Meta Business Help Centre

In detail:

1. Learning Limited it’s not a penalty

   Obviously it’s not ideal, but it’s not necessarily something to really worry/panic about.

   In an ideal world you would want to be generating enough convert to be able to not be “learning limited” but depending on your budget, depending on your audience size, that may not be possible.

   Let’s say your ad-set is generating roughly 10-15 conversions every seven days. That’s not enough for Facebook in order to optimise your ad-set because are required 50 conversions a week.

   However, as long as the ad-set cost per acquisition is acceptable and if it’s profitable the learning limited it’s not something to worry about.

   So don’t necessarily just see learning limited and think I need to change something (I need to change the campaign objective, I need to change my structure etc etc.). Instead, always look at your primary metric: what’s your cost per lead? what’s your cost per purchase? because if your cost per purchase or cost per lead is profitable nothing else matters, don’t worry about any of your other stuff.

   It’s only if those metrics aren’t working then the learning limited could indicate an issue.

   So, only if you’re not generating good results and you see learning limited that’s a time where you may want to make some adjustments.

2. Learning Limited adjustments recommendations

   

   As you can see (img above) Facebook provides some recommendations:

   • Audience Size.

     The first thing to do if you see learning limited is increase your audience size.

     It’s the first recommendation because you don’t need to spend any more money and it’s usually very easy to do.

     To do that Facebook suggests to consolidate your Ad Sets.

   • Cost Controls

     If you are using a bid cap make sure it’s not too low.

     The bid caps should be double your target cost per conversion which gives you quite a lot of flexibility a lot of leeway.

   • Budget:

     Make sure you have the budget to optimally reach 50 conversions in seven days.

     Depending on the industry, cost per unique acquisition/purchase (CPA) varies significantly. Facebook requires 50 conversions, so assuming your CPA is £60, then you are on the hook for £3,000 for every campaign that you launch.

     This is another reason why Facebook suggests to consolidate your Ad Sets.

     Does CBO influence the learning phase

     Facebook says the Learning Phase will not take any longer for advertisers who have implemented Campaign Budget Optimization (CBO). It also confirmed that ad sets within a campaign will not be forced to re-enter the Learning Phase as the campaign budget is distributed via CBO — and that making a significant edit to one ad set within a campaign will not cause an ad set within the same campaign to re-enter the Learning Phase. If you are using CBO make sure you’re spending enough to support each individual ad set within the campaign — too low of a budget could keep one or more of your ad sets from exiting the Learning Phase.

   • Consolidate Ad Sets

     When running too many ad sets at once, each ad set delivers less often.

     This means that fewer ad sets exit the learning phase and more budget is spent before the delivery system has fully optimized performance.

     By consolidating ad sets, you are increasing the volume of conversions for the ad set. This helps to reach Facebook’s threshold of 50 conversions needed to exit the learning phase.

     Note #1: If you’re just starting on facebook advertising and you’re testing to figure out who your audiences are, it is absolutely fine. I totally understand why you might want to have multiple audiences before you start scaling.

     Note #2 If you’re seeing sales and you’re seeing results great you can make the decision to have multiple ad sets potentially not exit the learning phase but at least get a read at what audience is performing better and then in following campaigns you start to condense and put more money into less audiences that are performing better.

Learning phase impact on CPA #

The graph illustrate the impact of learning phase on Cost per acquisition (CPA).

According to the official documentation,

Advertisers with ~20% of spend in the learning phase (point 1 – second decile) see 17% more conversions and 15% lower CPA than advertisers with ~80% of spend in the learning phase (point 2 – sixth decile).

Reference: Meta guide to the learning phase

The CPA curve gain is not linear. At only 3-5% drop in learning phase (Point A) corresponds a huge gain in CPA (Point #1 – approximately 55% decrease in CPA). However After this point the slope of the CPA curve starts to flatten and the CPA gains are marginal.

For this reason, advertisers should avoid behaviours that prevent ad sets from exiting the learning phase.

Reference: Meta guide to the learning phase

The chart below, from Closedloop, shows that each time new changes was made, a new learning phase (yellow shading) started causing CPA increased (blue line) and leads decreased (purple line).

Of course, every situation is unique and I would recommend doing a cost/benefit analysis to determine if it is worth going into the learning phase or not.

If you believe that making an adjustment that results in the learning phase will generate more of the results that you want and outweighs the time and money spent in the learning phase, then it may be more beneficial to make the adjustment.

Note: Here a video that shows dos & don’ts change during the learning phase. Also, here another video that goes through different changes scenarios.

Facebook Ad Inspect Tool #

• Audience saturation

  The audience saturation graph is part of the Facebook Ad Inspect Tool that helps you understand the relationship between your ad frequency and CPA.

  Looking at the graph below, you can see that as ad sets exit the learning phase, the audience saturation starts to increase. This especially happens if you’re targeting a lookalike or interest-based audience that is generally small.

  What happens is that your frequency starts to increase faster than usual, resulting in ad fatigue and audience saturation. So in conclusion really, exiting the learning phase isn’t a necessity since sooner or later you’re going to need to expand your audience.

  

  Note: the yellow area of the graph represents the learning phase.

• Insight

  The graph below shows how the Facebook algorithm categorizes spend and summarizes how the advertiser’s budget is classified.

  

  The teal section shows that ~2/3 of the spend is falling in the Learning Phase, which means the budget entered another category.

  In this case, the spend was optimized and moved to the purple “Amount Spent” section.

  The purple section represents the efficient spend, and the yellow section shows the small amount of budget that is classified as “Learning Limited.”

  The yellow Learning Limited section is a result of Facebook not gathering enough insights from the ad set to optimize the account.

  Although a bit of spending fell here, the data is a clue to look at the ad set(s) and make changes to help the algorithm graduate from the Learning Phase.