I get asked this a lot, and the number depends on the account. What I can show is how I found it on one account, so you can run the same check on yours.
I took one Meta account, $1.2 million in spend over 22 weeks, and paired every ad with itself the week after, which gave 722 pairs of ad weeks. Then I grouped the pairs by how many purchases the ad had in the first week and checked how well that week’s ROAS predicted the next.
The group with 50 or more purchases came out on top. Its correlation between this week’s ROAS and next week’s was 0.44, the strongest of all the groups, and it had the smallest average swing from one week to the next.
Below 20 purchases the numbers jump around instead of rising steadily: 0.20 for 1 to 4 purchases, 0.05 for 5 to 9, 0.09 for 10 to 19. With that few purchases, a couple of big buyers can move an ad’s ROAS enough to look like a trend when nothing has changed. Between 20 and 49 purchases the correlation was 0.255.
0.44 is still moderate
Even with 50 or more purchases, a good week only makes a similar week more likely. I still wait for a second good week before I move budget on the strength of the first.
A few caveats on the data. This is one account, so the 50 is not a benchmark for anyone else. An ad can contribute several pairs and neighboring pairs share a week, so 722 pairs are not 722 independent tests, and ads paused before their second week are not in the sample at all.
The 50 also has nothing to do with Meta’s learning phase. Three different 50s appear on this site, and they are easy to mix up:
| The number | What it counts | Where it comes from | What it is for |
|---|---|---|---|
| About 50 optimization events per ad set within 7 days | Optimization events, per ad set | Meta’s learning phase guidance | Below it, the ad set sits in Learning Limited |
| 50 optimization events a week per campaign | Optimization events, per campaign | My floor, from how many campaigns an account can carry | How many campaigns one budget can carry |
| 50 or more purchases in one week, per ad | Purchases, per ad | This account’s analysis | At or above it, an ad’s ROAS predicted its next week best, at a correlation of 0.44, which is still moderate |
The first two are floors for learning. Only the last one is about trusting a result, and it comes from one account.
Find the threshold in your own account
Fifty is this account’s number. Yours depends on the business model, how much purchase values vary, the attribution setup and the spend level, so it could be 30 as easily as 100.
The analysis is simple to repeat:
- Export ad results by week: spend, purchases and ROAS.
- Keep the ads that ran in two consecutive weeks and pair each week with the one after it.
- Group the pairs by purchases in the first week.
- Inside each group, correlate this week’s ROAS with next week’s.
Where the correlation jumps and the week to week swings shrink, that is the purchase count above which you can start to trust an ad’s ROAS in your account. Check it against a few later weeks before you rely on it, because a small group can look convincing by chance. If you are not sure how few purchases your budget decisions rest on, the growth audit is where I check this for an account, and it is bookable on its own at any spend level.
Run the analysis on your own data instead of borrowing my 50.