Change a bid strategy and Google enters a learning period. Performance gets worse before it gets better, and the instinct is to intervene. Intervening restarts the clock, which is how accounts end up permanently learning and never improving.
What is happening
A Smart Bidding strategy sets a bid for every individual auction using a model. When you change the strategy — or change something the model depends on — that model has to be rebuilt against the new conditions. Until it has enough data, it bids cautiously and unevenly.
Google shows this as "Learning" in the campaign status column. It typically lasts one to two weeks, longer on low-volume accounts, which describes most grant accounts.
Low conversion volume makes learning longer, not shorter. A campaign with four conversions a month is gathering evidence very slowly. If your account is in that range, expect three weeks rather than one, and plan the review date accordingly.
What it looks like from the outside
- Cost per click jumps around, often sharply upward.
- Daily spend is uneven — nothing on Tuesday, double on Thursday.
- Conversions dip in the first week.
- Impression share moves in both directions.
Every one of those is expected. None of them is evidence the change failed.
What restarts it
This is the part worth memorising, because several of these look harmless:
| Action | Restarts learning? |
|---|---|
| Changing bid strategy | Yes |
| Changing a Target CPA or ROAS value substantially | Yes |
| Large daily budget change | Usually |
| Changing which conversions are primary | Yes |
| Adding or pausing a few keywords | No |
| Editing ad copy | No |
| Adding negative keywords | No |
So routine optimisation continues during learning. What stops is anything touching the strategy, the target, or the conversion definition.
The rule that saves accounts
Make one strategy change, then do nothing to the strategy for three weeks. Write the review date in a calendar at the moment you make the change, because three weeks of poor numbers creates real pressure to act and a pre-committed date is what resists it.
The pattern that destroys accounts: switch to Maximize Conversions, panic on day five, switch back to manual, panic again on day twelve, switch to Target CPA. Three learning periods, no stable data, and a conclusion that "Smart Bidding does not work here" drawn from an experiment that was never allowed to run.
Judging it properly
At three weeks, compare against the three weeks before the change — same day count, and ideally avoiding a seasonal spike on either side.
- Conversions. The number that matters. Up is success.
- Cost per conversion. Useful context, secondary on a grant where the money is donated.
- Spend. On a grant account, higher spend after lifting the $2.00 cap is the intended outcome.
- Cost per click. Ignore it. CPC rising is the mechanism, not a problem.
If conversions are flat and spend is flat after three weeks, the bid strategy was not the bottleneck. Look at impression share next rather than trying another strategy.
Timing the change
Do not start a learning period immediately before something that matters. A giving-season appeal, an awareness day, or a campaign launch deserves a settled account.
Practically: make bid strategy changes at the start of a quiet month, so the learning period and the review both complete before anything is at stake. On most nonprofit calendars that means not in November, and not the week before a major appeal.
The short version
One change. Three weeks. No strategy edits in between. Judge on conversions, not on cost per click. That discipline is most of what separates accounts where Smart Bidding worked from accounts where it "did not".