How long does Meta's learning phase take, and what should we do when an ad set is stuck in Learning limited?
Our answer
An ad set leaves learning after about 50 results in the week after its last significant edit, often within a week. Don't edit it meanwhile. If it shows Learning limited, combine ad sets or raise budget before changing the event.
- Confidence
- Medium
- Strongest proof
- 4 Meta rules
- 5 practitioners, 4 opinions or claims
- Sources
- 6 sources
- 13 statements
- M
TS
New here? Take the 1-minute tour
Why this is the answer
4 reasons and our judgmentLearning ends on results, not days. Meta says an ad set usually leaves learning after about 50 results in the week after its last significant edit. Jetfuel puts that at a few days to a week for a well-funded ad set.
Every significant edit restarts the clock. Meta counts adding a new ad, changing targeting, creative, event or bid strategy, and pausing for 7 days or longer. It advises waiting to edit until the ad set is out of learning.
Learning limited means too little volume, and that costs money. Meta's fixes start with combining ad sets, widening the audience and raising the budget. By Meta's own figure, ad sets under about 50 conversions a week paid about 28% more per purchase.
The bar is a guide, not a wall. Jon Loomer plans on 25 to 50 optimized events a week and says good results are possible without leaving learning. PPC Land calls the thresholds heuristics set by the seller.
Our judgment
Act on Learning limited only when cost per result misses target, since Jon Loomer gets good results without leaving learning while Meta’s own analysis finds ad sets under the bar pay more per purchase. Expect a well-funded web ad set to leave learning within about a week, and judge it at two weeks, drawing on Meta’s bar of results in the week after an edit, Jetfuel’s few days to a week and Loomer’s two-week reading window.
Our judgment, built from Jon Loomer, Meta and Jetfuel Agency: our reading of these sources, not a finding from any one of them.
What the sources say
13 statements from 6 sourcesMeta says an ad set becomes learning limited when it is unlikely to get about 50 optimization events in the week after its last significant edit. Fixes include combining ad sets and raising budget.
“move from purchases to add to cart.”
Charley Tichenor says low volume is about conversions, not spend: $1,000 a day with 30 sales a week is low volume. Start with one ad set and add one only when it can get out of learning.
“Low volume isn't about spend. It's about conversions. $1,000/day with 30 sales a week is low volume. Start with one ad set.”
Meta says to give each ad set about fifty conversions a week, and that in its analysis ad sets under that ran about 28% more expensive per purchase. It gives a formula for how many ad sets a budget supports.
“Roughly fifty conversions a week.”
Meta says ad sets in the learning phase are less stable and usually cost more per result, so wait to edit until learning ends and avoid frequent budget changes, but do not try to avoid learning entirely.
“During the learning phase, ad sets are less stable and usually have a higher CPA.”
Meta says targeting, creative or optimization event changes and adding a new ad are always significant edits, while a budget change may or may not restart learning, depending on its size. No safe percentage is given.
“if you increase your budget from $100 to $101, that isn't likely to cause one or more ad sets to reenter the learning phase.”
Meta says cost per result goals work best with at least 50 to 100 conversions per ad set per week. Run them 3 weeks before scaling, judge on weekly averages, and wait 7 days after changing the goal.
“Cost per result goal works best when your ad set gets at least 50 to 100, or more conversions per week”
Jon Loomer says Meta campaigns can get good results without exiting the learning phase.
“you can get good results without exiting learning.”
Jon Loomer says Meta's learning phase still aims at about 50 events in 7 days but it is a soft rule; he plans for 25 to 50 optimized events per ad set per week.
“I’d recommend applying our typical volume threshold of between 25-50 optimized events per week to exit the learning phase.”
PPC Land says Meta says an ad set usually exits learning after about 50 optimisation events within 7 days of the last significant edit. Lower vendor-reported thresholds have no public Meta documentation.
“changes what the system buys: an ad set optimising for add-to-cart learns to find people who add to carts.”
Jetfuel, an agency, says Meta's learning phase lasts until an ad set gets about 50 optimization events, often a few days to a week. It says Learning Limited is not a penalty and is fixed by consolidating.
“It is not a penalty. It is Meta telling you the ad set does not have enough conversion volume to optimize properly.”
Jetfuel, with no client data, works an example: at a $40 target cost per purchase, about $2,000 of weekly spend is needed for 50 events. Too little budget for the target event causes Learning Limited.
“If your target cost per purchase is $40, that ad set needs to be able to spend in the range of $2,000 over the week to clear learning cleanly.”
The Social Outline, an app ad agency, says Meta's learning phase ends at about 50 optimization events in seven days. Plan 7 to 10 days for iOS installs, and fix learning limited by consolidating.
“Learning limited is a structural problem, not a waiting problem. The fix is consolidation, not patience.”
Jon Loomer says to judge Meta results on seven-day windows at least two weeks in, not daily swings, since short dips are often noise, and that manual intervention usually does more harm than good.
“Evaluate seven-day windows at least two weeks in rather than reacting to daily swings.”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer