Letting an AI agent run a whole campaign: goals, plans, an overseer and approvals
Letting an AI agent run a whole campaign works when the campaign is framed as a project — a goal, a period, a budget and KPIs — with a plan you review before anything runs and an overseer whose changes you approve. Here is how to brief one, what to check in the plan, and how to read the result.
1. Brief it like a project, not a task
A one-off task answers a question. A campaign is weeks of decisions against a target, so the brief has to carry what a good account manager would ask for on day one.
- The goal and the period — "grow holiday revenue on Google Ads and Meta from 1 November to 31 December", not "optimize the account".
- The budget, if there is one, so pacing has something to be judged against.
- KPIs with a direction and a number — cost per conversion at most 25, ROAS at least 4. A KPI without a threshold cannot trigger a decision.
- The campaigns in scope, so the agent does not wander into the brand account you keep separate.
- A direction: where to optimize, what to protect, what to avoid. "Focus on Performance Max; never pause brand campaigns" is worth more than a paragraph of background.
2. Review the plan before it runs
The plan is the cheapest place to catch a misunderstanding. A good campaign plan has a recognisable shape, and gaps in that shape are worth questioning before you activate it:
- Setup first: tracking checks and a baseline, so later numbers are compared with something real.
- A monitoring rhythm: a daily pacing check, a weekly review with proposals and a weekly report.
- Optimizations tied to the KPIs: budget reallocation, bid changes, pausing chronic underperformers, creative refresh, search-term and negative-keyword hygiene — each one there because a KPI could need it.
- Checkpoints: a mid-period review, a final report and milestones a person marks done.
3. How an overseer should decide
The value of an overseer is not that it acts often; it is that it acts for stated reasons. Useful reasoning looks like a short checklist run on every wake:
- 1Read every channel's numbers for the period and the previous equal period, so a change is a change and not noise.
- 2Compare them with the KPIs and with the budget pace — is spend on track to land inside the budget?
- 3Act only on a sustained gap with enough volume behind it: several days off target, with enough clicks or conversions that the number means something.
- 4Prefer small, reversible moves — shift part of a budget, adjust a bid — over large ones, and keep the total inside the project budget.
- 5Write down what it saw and why it did or did not act, so the next wake and the next person can follow the thread.
4. Decide where approvals stay on
By default every change a project proposes — on an ad platform, an email send, a budget move or the plan itself — waits for your approval card. That is the right starting point while you learn how the overseer reasons.
Turning on a project's auto-approve is an opt-in for ordinary changes, which then apply inside the workspace's spend caps and guardrails. Deletes, live sends, code and other changes a person must decide still wait for a person either way.
5. Read the dashboard, not the activity feed
- Start with the scorecards: each KPI against its target, and spend against budget with a projection to the end of the period.
- Then the per-channel and per-campaign table, with the change against the previous period. Different currencies are shown apart, never added together.
- Treat "partial" as a warning. When a channel could not be read, pacing and KPI judgements should say they are partial rather than pretend.
- Read the overseer's notes for the why behind each decision, and undo a plan change you disagree with.
Frequently asked questions
- Can an AI agent manage a marketing campaign end to end?
- It can plan, run and monitor one when the campaign is set up as a project with a goal, a period, a budget and KPIs, and when a person reviews the plan and — by default — approves each change the overseer proposes.
- How often should an AI campaign overseer check the numbers?
- Often enough to catch pacing problems, rarely enough that each decision rests on real volume. In Agent Planners a project's overseer typically wakes every 6 to 24 hours.
- Should I turn on auto-approve for an AI-run campaign?
- Not at first. Keep the default — every change waits for approval — until you trust how the overseer reasons, then opt in per project; deletes, live sends and code still wait for a person.