Facebook uses Machine Learning to find people that are more likely to care about your brand and to buy your product or your services.
Facebook categorizes the delivery path in 3 different stages:
- Learning Limited
We'll talk about these stages in more detail in the next few sections of this article.
When you create a new ad set or make a significant edit to an existing ad set or ad, Facebook's Machine Learning will explore and learn about the best combination of people to target, times of day, creatives, and placements to show your ad. This is the Learning Phase.
During this phase, ad sets are less stable and usually have a higher CPA and lower ROAS because Facebook hasn't yet identified the best prospects to see your ads or which one of your ads will resonate the most. It's also possible that not all of your budget will be well spent at this stage.
Ad sets exit the learning phase as soon as performance stabilizes, usually after around 50 optimization events since its last significant edit. A smaller share of budget spent in the learning phase typically results in a higher share of budget spent on stable performance and a lower cost per results.
For this reason, Facebook recommends that you set up each ad set to get 50 conversions in 7 days. The faster your ad set gets out of this phase, the better; this is why it is important to take into account the Learning Stage when optimizing your campaigns.
If your ad set does not generate enough data for the Facebook Delivery System to learn what it needs to learn, the Learning stage will change from Learning to Learning Limited.
During this stage, you may see low performance. We recommend that you widen your budget, redefine your audience, or switch to a lower event in the funnel. Please contact your Marin Account Representative for guidance specific to your Business.
Contrary to the Learning Limited stage, if the Machine Learning was able to gather enough data from its testing, it will use the winning creative combinations (if you have several ads within the same ad set, or if the Asset Customization is activiated) with the winning placement and focus on delivering your ads to the people more likely to convert
Test different attribution settings or optimization events to reach the 50 conversions recommended by Facebook.
Use the Marin Ad Study feature for accurate testing and learnings.
If performance is low, wait for your ad set to be out of the learning phase to optimize. The Machine Learning is still trying to find the best people within your audience to whom to show your ads.
Avoid unnecessary edits that cause ad sets to re-enter the learning phase.
Avoid splitting your audience into many ad sets and consolidate as much as possible. This gives the algorithm more space to deliver, explore, test, learn, and be successful at its job.
In the coming months, Learning Stage will be available to select from amongst the list of performance metrics in the Marin Rule Conditions. For example, you’ll be able to exclude ad sets that are still in the Learning Phase from your Automated Bid and Budget Optimization.
Stay Tuned for more information!