Learning Phase · 20% Deep-Dive
What Meta really learns.
What the Learning Phase Actually Is
The learning phase is not a marketing buzzword. It is an algorithm status with clear rules. An ad set needs 50 conversions in 7 days, otherwise it stays in Learning. While in Learning, performance can fluctuate because Meta has not yet determined who to serve the ad to.
01
50
Conversions / 7 days for stability
02
+20%
Max jump without reset
03
+25%
Trigger for reset
The Reset Mechanic
If you increase the budget by more than 25% per day, Meta resets the learning phase. The next 50 conversions are in Learning, with unstable performance. With an active scaling plan, every reset costs 5 to 10 days of performance stability.
Example
47 accounts in Q1 2025: a +20% cadence kept ROAS loss below 5%. A +30% cadence produced 22% ROAS loss per jump. The math is trivial.
How to Run the Rule in Practice
- Daily budget increase of max +20% per ad set
- 24-hour observation per jump, with ROAS stability as the gate
- If ROAS drops below 0.9x target, pause for one day
- With each jump, activate 4 to 6 new hooks in parallel
„Scaling is patience plus discipline. Those who are aggressive burn budget and learning phase alike.”