What Is Learning Phase?
The Learning Phase is Meta's training period after a new ad set launches or is significantly edited. Meta needs about 50 optimisation events in 7 days to stabilise delivery. Until then, performance is volatile and metrics cannot be trusted for decisions.
Learning Phase in a real launch
A new ad set launches with a 100 dollar daily budget optimising for Purchase. Week 1, CPA swings from 22 to 68 dollars day to day. By day 8 the ad set has 42 purchases, still Learning. The buyer resists the urge to cut budget or kill creatives, waits to day 12, hits 61 purchases, exits Learning, and CPA settles at 38 dollars. Cutting on day 5 would have thrown away a profitable ad set based on noise.
Benchmarks
- Exit threshold: 50 optimisation events in 7 rolling days.
- Typical duration for a well-funded new ad set: 3 to 10 days.
- Cost per result during Learning: often 40 to 100 percent higher than post-Learning steady state.
Why it matters
Every edit that resets the Learning Phase costs money and time. Buyers who tweak daily keep every ad set in permanent Learning and never see true performance. The discipline is to plan the launch, resist the edit, and only judge after the exit.
Common mistakes
- 1.Killing ad sets on day 3 based on CPA. Sample too small to mean anything.
- 2.Editing budget by more than 20 percent in one move. Resets Learning.
- 3.Changing creative, audience, or optimisation event mid-Learning. Full reset.
- 4.Optimising for a bottom-funnel event a small account cannot fire 50 times a week (Purchase on a low-volume brand). Move up-funnel to AddToCart until volume supports Purchase optimisation.
Put Learning Phase to work
Free calculators
Related services
FAQs about Learning Phase
What edits reset the Learning Phase?
Significant budget change (over 20 percent), audience change, creative change, placement change, optimisation event change, bid strategy change, or pausing 7 plus days. Naming and labels are safe.
My ad set says Learning Limited. What now?
Not enough weekly conversions. Fixes: increase budget, broaden audience, consolidate similar ad sets into one, or move optimisation up-funnel (Purchase to AddToCart). Do not just wait; Learning Limited does not fix itself.
Go deeper
Long-form reads that show Learning Phase applied end to end.
Related terms
Ad-set warning: not enough conversions to exit learning; scale, broaden or consolidate.
The single action Meta's algorithm is asked to maximise inside an ad set.
Budget set at campaign level; Meta distributes across ad sets.
Controlled test isolating one variable to prove which version wins.
Meta's AI-driven campaign setup that automates targeting and placements.
Google auto-bid strategy that targets a specific cost per conversion.
Google auto-bid strategy that targets a specific return on ad spend.
Meta's privacy-safe event system after iOS 14.5; caps you at 8 ranked events.