PPC automation with approval: delegate, keep control
How to automate PPC work without handing over the keys: approval by default, limits per change type, monitoring after every change, and honest restore.
On this page
Automate the work, not the decision. Let software monitor, check and draft changes, and keep a person approving what actually changes in the account until a change type has earned trust.
Why "fully autonomous" is the wrong goal
Ad platforms already automate a lot: Smart Bidding sets bids per auction, PMax picks placements, AI Max widens matching, and auto-applied recommendations change settings on their own. Another fully autonomous layer on top gives you two systems deciding about your money, and neither explains itself.
What business owners and marketers actually want is narrower:
- Someone (or something) watching the account every day.
- Problems caught early, with the money at stake.
- Fixes drafted, so they don't have to build them.
- The final say on anything that changes spend or targeting.
- A record of what changed, why, and what happened next.
The four parts of approval-based automation
- 1Proposal
- 2Approval
- 3Monitoring
- 4Restore
| Part | What it means | What to look for in a tool |
|---|---|---|
| Proposal | The change is drafted before anything happens | Current vs. proposed values, the evidence, the expected effect and what's uncertain |
| Approval | A person decides, per proposal or per change type | Approve, edit, or keep current settings, with roles so not everyone can approve |
| Monitoring | After the change, a protected metric is watched against a baseline | A defined window, a threshold, and enough data before judging conversions |
| Restore | Supported settings go back if monitoring calls for it | Which change types support restore, and a clear statement that spend isn't recovered |
A good proposal answers seven questions on its own: what will change, why, the evidence, current and proposed settings, the expected outcome and its uncertainty, whether it can be restored, and how to approve or decline. If you have to open three reports, it isn't done.
Delegating one change type at a time
| Change type | Typical risk | When to consider auto-apply |
|---|---|---|
| Adding exact-match negatives for clearly irrelevant terms | Low | Early, with a simulation of what they would have blocked |
| Pausing an ad that's disapproved or links to a broken page | Low | Early |
| Small daily budget adjustments for pacing | Medium | After a few weeks of approved proposals |
| Changing Target CPA or Target ROAS | Medium to high | Later, with tight limits |
| Structural changes (new campaigns, match type changes) | High | Usually keep manual |
For each type you enable, set limits: maximum change size, maximum changes per day, protected campaigns that are never touched, and hours when nothing runs (for example, Shabbat or a holiday).
Monitoring needs patience. Spend can be judged within hours; conversions can't. They arrive late, and Smart Bidding needs time to adjust. A sound rule waits for a mature click cohort and a minimum lag, and checks that tracking is healthy before blaming the change.
A budget increase that doesn't hold
Worked example+20% budget, restore if CPA exceeds 1.5× baseline
The numbers are illustrative, not real account data. A campaign spends $500 a day at a $50 CPA. You approve raising its budget to $600. CPA is judged after a 72-hour click cohort has matured and at least five days have passed.
| Day | Event |
|---|---|
| 0 | Budget raised from $500 to $600; previous settings snapshotted |
| 1–7 | Spend rises; conversions lag, as expected. No conversion-based judgment yet |
| 8 | Cohort is mature. CPA is $78, which is 1.56× baseline. Tracking healthy, no other changes |
| 8 | Budget restored to $500; you're notified with the reason and a link to the change log |
Restore stops the higher spend going forward. It doesn't recover the extra roughly $100 a day spent during the test, or guarantee the campaign returns exactly to its earlier performance. A good tool says both plainly.
Common objections
"Approving everything is slower than doing it myself." Bulk approval for low-risk proposals, plus auto-apply for change types you trust, keeps the queue short. What you skip is the investigation and the drafting.
"Google already offers auto-apply." For Google's own recommendations. The difference is who sets the limits, whether you see before and after, and whether there's monitoring and a log. Either way, know which auto-apply types are on.
"What if the tool is wrong?" Sometimes it will be. That's why proposals show evidence, monitoring exists, and restore is defined in advance.
A checklist for any automation you adopt
- Approval is the default; automatic action is something you enable.
- Limits are per change type, not one global switch.
- Every change is logged with who approved it and through which channel.
- Monitoring windows fit the metric (spend: hours; conversions: days).
- Restore support is stated per change type, and the tool never implies it recovers spend.
- Your manual changes and the platform's changes show up on the same timeline.