Definition
The learning phase is the period where Meta's delivery system is still gathering data to figure out who converts on a new or significantly edited ad set. It's typically exited after roughly 50 optimisation events per week.
Also called: Learning Phase, Learning, Learning periodTest the message before scaling the budget.
Everyday colour.
One variable per comparison makes the learning easier to interpret.
01Define what the customer should understand or want.
Why 50 conversions is the threshold
Meta recommends around 50 optimisation events per ad set per week to exit the learning phase. Below that threshold, CPA swings wildly and the algorithm can't find a stable pattern. It's the single most important reason to consolidate budget into fewer ad sets rather than spread it thin.
“Learning Limited” means an ad set can't reach those 50 conversions and therefore never leaves learning. The fix is almost always to consolidate: fewer units, broader audiences, more budget per set.
Significant edits reset learning
A significant change (budget shifts over ~20%, a new audience, new creative, or a changed optimisation event) sends the ad set back into the learning phase. That's why we dose changes, make them early in the week, and avoid daily tinkering. Here, patience is a technical parameter, not a virtue.
Test the message before scaling the budget.
Everyday colour.
One variable per comparison makes the learning easier to interpret.
01Define what the customer should understand or want.
Frequently asked questions
How many conversions are needed to exit the learning phase?
Roughly 50 optimisation events per ad set per week. Fall short and delivery runs unstable (Learning Limited), the core argument for consolidating budget into fewer units.
What resets the learning phase?
Significant edits: a budget change over ~20%, a new audience, new creative, or a changed optimisation event. Minor tweaks don't. Dose significant changes so you don't hold the ad set in constant learning.
From insight to action
See how it applies in practice.
Growth also needed a better purchasing process
We built Matraws' trade-in system with submission, assessment, offers and handling in one backend. The work addressed supply as a constraint.
Read the casePriorities and collaboration
Discuss your next step
Tell us where your business stands. We clarify the work needed and whether we are the right fit.
Choose a time30 minutes · No obligation
About the author
Nicklas Segatz Mortensen
Growth hacker and fractional CMO with 10+ years' experience and hundreds of millions in managed ad spend behind him. Background from larger Danish and international scale-ups, and from the agency world.
Meet the team