Understanding the Learning Phase
Quick answer
When you launch a new ad set or make a significant change, the platform's algorithm restarts a data-gathering period. Performance is volatile and unreliable during it. Agency AI deliberately holds recommendations until it's over. If your Recommendations page is empty in week one, this is usually why.
Meta
An ad set exits the learning phase after roughly 50 optimization events in any rolling 7-day window. For most e-commerce accounts that takes 3 to 7 days, though many accounts now see 7 to 14 days following Meta's 2026 algorithm changes.
If an ad set can't reach 50 events per week, it lands in Learning Limited and stays there. That's not a bug. It's arithmetic. Daily budget × 7 ÷ your CPA = your weekly conversions. If that's under 50, no amount of waiting fixes it.
During learning, CPAs typically run 20% to 50% above your stabilized baseline. Daily ROAS swings of 20–50% are normal. Do not make decisions on this data.
What resets Meta's learning phase
- Budget changes greater than 20% in a single edit
- Targeting changes
- Bid strategy changes
- Adding or removing ads from an ad set
- Changing the optimization event
- Pausing for 7 or more days
Meta tightened what counts as a "significant" edit in 2026. Changes that used to be safe now trigger resets. Each reset costs real money: the algorithm re-explores, and the next week's performance suffers.
The most common accidental reset is a budget change. Going from $100/day to $125/day is a 25% change. That resets learning. This is exactly why Agency AI caps budget increases at 20% per cycle and enforces a cooldown between increases.
Google Performance Max
Google's learning period typically runs 7 to 14 days, with full stabilization taking longer. Expect elevated CPA during it.
Changes that restart it include budget shifts greater than 20% in either direction, bid strategy switches, new asset groups, and major changes to audience signals.
This is why Agency AI adjusts target ROAS rather than budget on Google. Smart Bidding responds to target changes in real time, and budget moves on a Performance Max campaign are far more disruptive. Your daily budget on Google functions as a ceiling, set once at onboarding.
Stuck in Learning Limited?
Three fixes, in order:
- Consolidate. Five ad sets getting 10 conversions each is worse than one ad set getting 50.
- Raise the budget so the math works. Budget × 7 ÷ CPA must clear 50.
- Optimize for a higher-volume event. If purchases are too rare, optimize for Add to Cart or Initiate Checkout instead. This costs conversion quality. Use it when the alternative is permanent Learning Limited.
Learning Limited is not the same as bad performance. If your 30-day average ROAS is healthy, an ad set can sit in Learning Limited and still make you money. It just won't be stable.
What Agency AI does about it
Agency AI blocks performance rules from firing on ad sets that are in learning. It waits, then evaluates against a confidence score built from spend, time active, conversion volume, and trend stability. Silence in week one is the system working.