What runs without a person
The work that can run unattended shares three properties: it is bounded, it is reversible, and it is judged against a target somebody already set.
- Bid and bid modifier changes against an agreed efficiency target
- Budget moved between campaigns inside a fixed total
- Negative keywords added from search term reports
- Ad groups and keywords paused when they spend past a threshold without converting
- Device, location and schedule adjustments where the data is unambiguous
- Disapprovals, tracking breakages and sudden delivery drops, caught and fixed or escalated
None of these decisions require a view about the business. They require attention, applied consistently, at a frequency no person sustains. That is the part worth handing over.
What still needs a person
- The budget ceiling. How much the business is prepared to lose while learning.
- What a conversion is worth. Margin, lifetime value and which orders you actually want more of.
- What the account is for. Prospecting and efficiency pull in opposite directions, and only the business knows which quarter it is in.
- Brand and claims. What may be said, and about what.
- Anything with a contractual or regulatory consequence.
These are not harder computations. They are decisions with no correct answer inside the account data, which is exactly why a system cannot be trusted to make them. A platform that sets its own target and then reports that it hit the target is marking its own homework.
Where the line sits
| Decision | Who decides | Why |
|---|---|---|
| Raise a bid 12% on a converting keyword | System | Bounded, reversible, judged against a set target |
| Move £400 from a weak campaign to a strong one | System, inside a fixed total | The total is the constraint, not the split |
| Increase the monthly budget by £4,000 | Human | It is a spending decision about the business |
| Set target ROAS at 4.2 | Human | Encodes margin, which is not in the ad account |
| Pause a campaign at 3x the target CPA | System, with notification | Stops the bleeding, easily reversed |
| Launch a campaign for a new product line | Human | A commercial bet, not an optimisation |
Google's own automation is not the same thing
Smart Bidding and Performance Max automate the auction. They do not manage the account. Both optimise towards the target they are given, and both will hit a target that is wrong with complete confidence.
Nothing inside Google Ads notices that the campaign with the best ROAS is selling the product with the worst margin, because Google does not hold your margins. The useful autonomy sits above the auction, where the ad data meets the order data, and it needs the systems read together rather than one at a time.
When the answer is genuinely no
Autonomous management does not help, and can hurt, when:
- Conversion tracking is unreliable. Every decision downstream inherits the error, faster than a person would have made it.
- Conversion volume is thin. Below roughly 30 conversions a month per campaign, most of what looks like a signal is noise, and acting on it weekly is worse than acting on it quarterly.
- The calendar is not in the data. A promotion, a product launch or a stock outage changes what good looks like, and nothing in the account announces it.
- The category is regulated. Where copy needs approval, the approval is the bottleneck and automating around it is the wrong fix.
How to check whether a system is actually autonomous
Four questions, and the answers should be specific rather than reassuring:
- What does it change without asking?
- What does it hold for approval?
- What can it never do, whatever happens?
- Can it show you what it would have done, before it is allowed to do anything?
The fourth is the one that separates a real system from a demo. If it cannot run in observation mode for a fortnight and hand you a list of the decisions it would have made, you have no way to judge it except by watching it spend. What happens when one of those decisions is wrong is a separate question worth asking.