After a significant change, Smart Bidding re-gathers data before it can bid confidently. During that window performance is unreliable and, crucially, not representative of anything.
How long
Roughly two to four weeks for an account of normal volume. That is a working estimate from running accounts, not a figure Google publishes. Higher conversion volume shortens it; very low volume can mean the bidding never really leaves it.
What triggers it
- Changing bid strategy or target significantly
- Large budget changes
- Restructuring campaigns or asset groups
- Changing the conversion action or its value
Note the last one. Fixing broken tracking is worth doing and it will disturb the bidding. That is a price worth paying — but do it deliberately, not in the same week as three other changes.
The habit that keeps accounts stuck
Change everything. See bad numbers during learning. Panic. Change everything again. The account is now permanently learning and never performing, and every change is justified by data produced during the previous change.
One significant change at a time, then wait long enough to read the result. Boring, and it is most of what separates accounts that compound from accounts that thrash.
What is folklore
Exact percentage thresholds for budget changes, "never touch a campaign on a Friday", magic conversion counts that unlock performance. None of that is published and most of it is one person's experience generalised. What is reliably true: volume shortens learning, instability lengthens it, and nobody can promise results faster than the platform can relearn.