A significant edit sends an ad set back into the learning phase, where delivery spends inefficiently on purpose — so an account optimised every few days can spend most of its life in the most expensive state it has, and the weekly tweak ritual is often the thing causing the instability it is trying to fix.
There is a habit almost every agency has, and it looks like diligence. Monday morning, open the accounts, and adjust something. A budget nudged up where it is working, an audience narrowed where it is not, a creative swapped where the numbers dipped.
Every one of those is defensible on its own. Together, on a weekly cadence, they can keep an account in the single most expensive state Meta has.
What the learning phase actually costs
When an ad set is new or significantly changed, delivery does not yet know who to show it to. So it explores — spending money to find out, rather than spending it on the people most likely to convert. Meta documents the exit at roughly fifty optimisation events in a week, though the number matters less than the principle.
That exploration is not waste. It is the price of information, and an ad set that never pays it never gets good. The problem is paying it repeatedly for the same ad set.
Learning is a cost you pay once to get an efficient ad set. Paying it every Monday buys you nothing and looks like diligence.
The arithmetic nobody does
An ad set with modest lead volume can take one to two weeks to accumulate enough events to exit. Edit it every seven days and it may never exit at all. Everything you then observe — unstable cost per lead, results that will not settle — is the signature of an account that is permanently learning, and it reads exactly like an account that needs more attention.
So the tweaking intensifies, which is the one intervention guaranteed to make it worse. This is the most common self-inflicted problem we see, and it is invisible from inside a single week’s numbers.
What actually resets it
Not everything counts. Roughly speaking, things that change who sees the ad or what they see will reset it; things that do not, will not.
- Resets: changing targeting, changing the optimisation event, swapping the creative, changing placements, a large budget change.
- Usually does not: a small budget adjustment, pausing overnight and resuming, editing an ad name.
- Grey area: budget changes are a matter of degree, not kind — a nudge is absorbed, a doubling is a different ad set.
Which means most of the Monday ritual sits firmly in the first list.
The discipline this implies
Not “never edit”. Sometimes an ad set is genuinely broken and the learning cost is worth paying. The discipline is knowing you are paying it, and having a reason that survives the question “is this a real change, or is this last week being noisy?”
- Decide what would have to be true before you touch it, before you open the account.
- Let an ad set finish learning before judging it. Reading a cost per lead mid-learning is reading the price of exploration.
- Batch edits. One change that resets learning once is cheaper than three changes across three weeks that reset it three times.
- When you do edit, write down what you expected — otherwise you cannot tell later whether it worked or whether learning simply ended.