Six Edits That Restart Meta's Learning Phase (And Six That Don't)
You launch a campaign. Day three, results look poor. You nudge the budget, or tighten the audience, or add a better ad. Perfectly reasonable, and you have just thrown away three days of the algorithm's learning and started again at a higher cost per result.
The frustrating part is that some edits are completely safe. The difference is not obvious, and Meta does not warn you.
These restart learning
- Budget changes of 20% or more, measured against where the ad set started rather than against your last edit.
- Several smaller budget changes that add up past 20%. This is the one that catches people.
- Any targeting change. Audience, placement, geography, exclusions.
- Changing the bid strategy or the optimisation event.
- Adding a new ad to a live ad set, even a better one.
- Pausing an ad set for 7 days or more.
These are safe
- Budget changes under 20% of the starting amount
- Renaming a campaign, ad set or ad
- Pausing or activating an individual ad inside the ad set
- Adjusting the ad schedule
- Pausing for less than 7 days
- Rotating between ads already in the ad set
The cumulative rule almost nobody knows
Here is the trap. You raise the budget from $100 to $110, a 10% change, safely under the threshold. Two days later, $110 to $121. Two days after that, $121 to $133.
No single change was over 20%. But Meta compares against the baseline for the current learning window, not against your previous edit. From $100 to $133 is 33%, and learning restarts.
This is why "just make small changes" is incomplete advice. You have to track the running total, which means recording the baseline and measuring every proposed change against that number rather than against today's budget.
What this means in practice
Load every ad at launch
Adding an ad later restarts learning, so anything you might rotate in has to be there from day one. Upload five or six, activate two, leave the rest paused. Rotation between existing ads is free; introducing a new one is not.
Change nothing for seven days
The hardest instruction in advertising, because a campaign spending money with unclear results is uncomfortable to leave alone. But the numbers during learning are inflated by definition. Reacting to them means seeing the same inflated numbers again next week.
Scale in steps, with gaps
Once learning completes, raise by up to 20% at a time and wait three or four days between raises. Watch the cumulative total against the baseline.
Pause ads, not ad sets
Turning off an underperforming ad inside the ad set is safe. Touching the ad set is not. This gives you a real lever during learning, which is worth knowing when the urge to intervene becomes overwhelming.
Count your resets
If a campaign has restarted learning three or four times, the problem is not the creative or the audience. Each reset costs roughly a week at inflated cost, and a campaign reset that often never reaches its steady state at all. The entire budget goes into re-learning.
The fix is not a better edit. It is fewer edits.
MyNexusBiz tracks the baseline and the running total, and answers one question before you touch anything: can I change this today.