There are situations in which the best decision is to step back from automation and return to manual control: scarce data, abrupt business changes, erratic behaviour or brand risks. Defining the criteria for switching AI off in advance avoids emotional decisions and lets you switch it back on with method.
The essentials in a definition
Knowing when to turn AI off is part of supervision: there are situations where automation does not serve and manual control is preferable. Clear criteria, defined beforehand, avoid emotional decisions. NIST’s approach of measuring, reviewing and having contingency plans applies, and controls of campaigns such as smart bidding and Performance Max should include a fallback plan.
When AI stops helping
When there is little information for the system to learn from, when the business changes quickly (prices, products, promotions), when conversion data is wrong or when campaign behaviour becomes unpredictable and costly. Also when there are legal or brand restrictions the system cannot respect.
In such cases, staying on automatic can cost money and trust.
Clear criteria
Define thresholds: cost per result above X for Y days, spend without conversions above Z, share of irrelevant terms above W, compliance failures. When a threshold is crossed, a review is triggered and, if necessary, a move to a more manual mode.
The criteria should be written before the problem arises.
Stepping back from automation with method
- Define step-back criteria before the problem
- Reduce automation in steps
- Fix the cause
- Test again and compare
How to step back without losing everything
You can reduce the level of automation gradually: use simpler bidding, restrict terms, fix budgets, limit audiences. Keep structures and history. Instead of deleting campaigns, pause and document. Note what changed, to compare.
Avoid stepping back and forward repeatedly: each change restarts learning.
Knowing when to switch AI off is part of knowing how to use it.
How to automate again
Resolve the cause (data, tracking, structure, offer). Define a test phase with a limited budget. Compare with manual mode in the same period. Only delegate fully again when results are stable and understandable.
Record the decision and the criteria for the future.
Decision table
| Signal | Interpretation | Action |
|---|---|---|
| Corrupted conversion data | The AI learns from junk | Move to manual control until fixed |
| Abrupt change of offer or price | History no longer serves | Temporary manual limits |
| Brand or legal risk | The AI may breach rules | More control and approval |
| Persistently worse results with no cause | Possible mismatch | Experiment with a manual version |
Common mistakes to avoid
- Turning off from frustration. Use criteria and tests, not impulses.
- Having no fallback plan. Without a ready manual setup, time is lost.
- Never testing the alternative. Compare periodically with manual control.
How to measure the result
Record each activation of the fallback plan, its cause, resolution time and effect on results.
Limits and exceptions
Returning to manual control also has costs: worse performance, more work and loss of learning. In small accounts, manual may be the default choice for a while. The aim is balance: use AI when it helps and step back when it does not serve. Review criteria with experience and avoid impulsive fallbacks.
Illustrative example
Illustrative example: a company defines written criteria: if recorded conversions depart more than 25% from real sales, if there are policy disapprovals or if cost per customer rises 40% in two weeks without cause, it moves to manual control until fixed. When a tracking error occurs, it applies the plan and recovers within days, without serious losses.
Frequently asked questions
Is switching AI off a failure?
No. It is a management decision when conditions do not favour automation.
How do I avoid oscillating?
With written criteria and minimum periods before changing.
Do I need to delete campaigns?
No. Pause, document and preserve the history.
How do I maintain a fallback plan?
With ready, documented manual campaigns or settings, and clear criteria to activate them.
Is going back to manual a step backwards?
No; it is a temporary safety measure while the cause is fixed.
Define written criteria to fall back to manual control (corrupted data, abrupt changes, brand risk or anomalous results) and keep a fallback plan ready.
Sources and recommended reading
- NIST: AI Risk Management Framework
- Google Ads Help: Smart Bidding
- Google Ads Help: Performance Max campaigns
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.