Legacy: How Does Facebook Learn and Optimize Campaigns?

Last Updated: September 27, 2023

🔍 Please note: this article is for customers on Standard Bundle, Enrichment, or Clearbit Platform plans.

It may sound counter-intuitive to say that higher daily budgets will lower costs; however, Facebook's machine learning algorithms rely heavily on a high volume of conversion signals at the beginning of a campaign lifecycle to become "efficient."
When campaigns first start (generally during the first two weeks of campaign launch), Facebook spends more aggressively and delivers ads more broadly in an attempt to "learn" which combination of ads and users will lead to the lowest cost conversions. They call this the "Learning Phase."

During this period, CPAs are higher and conversions lower. Once Facebook can "exit" the learning phase, you can expect a jump in conversions and a drop in CPA. That's because Facebook's Machine Learning algorithms have identified patterns that they can use to predict where the next conversion will come from and show the right ads to the right people at the right time.
When budgets are too small, Facebook doesn't have enough signal to exit the learning phase. That leads to a long-term persistent problem of high CPAs and low conversions. Instead of optimizing, Facebook is stuck using a "spray and pray" method of showing ads to everyone in the audience and hoping someone converts.

Lost Efficiency

Being stuck in Learning Limited isn't a minor issue. Facebook relies heavily on its ML delivery optimization to generate low-cost, high-quality conversions.

According to Facebook, “advertisers with ~20% of spend in the learning phase (2nd decile) see 17% more conversions and 15% lower CPA than advertisers with ~80% of spend in the learning phase (6th decile).”

How To Make the Most of a Limited Budget

If your budget is limited, and you do not have the spend required to give each ad set sufficient daily budgets, I would recommend a few different approaches:

  1. Double down on campaigns that are performing, and defer others - If your offer resonates with the audience (meaning more conversions at a lower cost), reallocate your budget to that audience. It's great to test multiple audiences, but if your budget is spread too thin, you will end up with poor results across all audiences (instead of excellent results in one).
  2. Consolidate ad sets - If you are running more generic ads (that resonate with all audiences), consolidate them into one ad set and budget. This strategy will allow you to get through the learning phase faster and give you economies of scale.
  3. Optimize for up-funnel events - Worst case, you can optimize for up-funnel events like Traffic. Facebook's bidding algorithms factor in the Estimated Action Rate when pricing impressions. Bidding up-funnel will encourage Facebook to discount your CPMs. Traditionally it is best to use the event you want to optimize for, but if the volume is too low, you shouldn't lose much (if any) efficiency with this method.
  4. Reserve a % of the budget for testing - From an organizational perspective, I encourage performance marketers to ask finance for an "experimental" budget that is not factored into your ROAS calculations. This is important because tests need breathing room to run to completion, without the fear that they will drag down your blended ROAS.