Smart Bidding learns from the recent past. That works as long as the future looks like the past. When you know it will not — a three-day sale, a television promotion, a week when measurement was broken — you have two tools in Tools → Bidding strategies → Advanced controls: seasonality adjustments and data exclusions. They are powerful and easy to misuse.
Seasonality adjustments
You tell the algorithm: "between these dates, expect the conversion rate to change by X%". It adjusts the bids during that period without "learning" that the world has changed for good, and when it ends it goes back to normal without dragging the spike along.
What it is for: short events (1-7 days) with a sharp, predictable change in the conversion rate: flash sales, Black Friday, a weekend discount code, a mention in the press. Google recommends it for periods of less than a week.
What it is NOT for:
- Long seasonality (the Christmas run-up, the summer): the algorithm already picks that up from its own history and Google's aggregated data. A month-long adjustment distorts more than it helps.
- Changes that started days ago: Smart Bidding has already seen them.
- "Pushing up" conversions that are not going to go up: if you adjust +50% and the rate does not move, you pay more expensive clicks for the whole period.
How to set it up: exact dates, affected campaigns (or the whole account), device if relevant, and the estimated percentage change in the conversion rate (not in conversions, and not in traffic). If in the last sale your rate went from 2% to 3%, the adjustment is +50%. Be conservative: better to fall short than to overshoot.
Data exclusions
You tell the algorithm: "between these dates the conversion data was wrong; do not learn from it". Smart Bidding ignores that period when calibrating its models.
What it is for: periods with broken measurement — the tag stopped firing, conversions were duplicated, a redesign took down the thank-you page, the cookie banner blocked everything for a week. Without an exclusion, the algorithm will have learned from a hole (or from a false spike) and will take weeks to recover.
What it is NOT for: periods when the business genuinely did badly (that is information), or to "delete" an experiment you did not like, or as an alternative to fixing the measurement.
How to set it up: the exact dates of the problem (the change history and the conversion diagnostics tell you when it started and ended), affected campaigns, device if the failure was limited to one. It can be created after the period: it is retroactive.
The procedure when measurement breaks
- Detect it (conversions at zero with normal clicks, duplicates...).
- Fix the tag and verify with a real conversion.
- Create the data exclusion for the exact period of the failure.
- Do not touch targets or budgets during the learning period that follows (the strategy goes into "learning" when you exclude).
- Note it in the change log: what failed, when, what was excluded.
Common mistakes
- Seasonality adjustments covering whole weeks "just in case".
- Forgetting to remove a badly set adjustment (it can be edited or deleted; do not wait for it to expire once you can see it was wrong).
- Data exclusions used to cover up a bad commercial week.
- Setting the percentage on traffic or sales instead of on the conversion rate.
- Not telling anyone: another user sees an odd bid change and "fixes" it.
💡 Ninja trick: the Suite's Smart Bidding (SBNS) includes the anti-anomaly threshold that prevents exactly this kind of damage from a broken period: if a day's CPA shoots above what is reasonable (in the account's currency), it does not take it as a signal to move targets. And the Agent warns you the same day about a drop in conversions with normal clicks, which is the moment to create the exclusion — not three weeks later.
What you should remember
- Seasonality = a short, predictable event; a percentage on the conversion rate; conservative; remove it if you get it wrong.
- Data exclusion = broken measurement; exact dates; retroactive; it comes after fixing the tag.
- Neither of them replaces measuring properly, and neither is there to dress up a bad week.