The line between its calls and yours
An ads agent is an AI agent that runs your paid advertising day to day. The hard part of using one is not the connection to your ad accounts. It is the boundary of the brief, meaning which decisions the agent makes on its own, which ones it brings to you and which ones it never touches.
Get that line right and the agent takes real work off your plate without putting your money or your name at risk. Get it wrong in one direction and you are approving every small tweak. Get it wrong in the other and the agent is making business choices you never meant to hand over.
This piece is about drawing that line, not the daily work of an ads manager agent.
Three kinds of decisions
Every decision in your ad accounts falls into one of three groups. Sorting them before the agent starts is most of the work.
- The agent owns it. Small, frequent, reversible changes inside limits you set. It acts and tells you afterward in its summary.
- The agent proposes it. Changes that matter more. It does the research, makes a recommendation and waits for your yes.
- You keep it. Choices about what your business is, who it serves and what it promises. The agent may give you data, but the decision stays with you.
The rest of this article is a way to decide which group each decision belongs in.
Four questions that sort a decision
When you are unsure where a decision belongs, ask these four questions of it.
Can it be undone quickly?
Pausing an ad can be reversed in a minute. A post that went out with the wrong price has already been seen. The easier a change is to undo, the more safely an agent can own it.
How much money does it move?
Shifting a small amount between two ad sets is different from doubling a campaign’s daily spend. Set a size. Below it the agent acts. Above it the agent asks.
Does it speak for your business?
Bids and budgets are mechanics. Ad copy, offers and images are your voice in public. Anything a customer reads deserves a different rule from anything only the ad platform sees.
Is it strategy or execution?
Deciding to sell in a new city is strategy. Keeping the campaigns for your current city in good shape is execution. Agents are well suited to execution. Strategy is a business owner’s job.
A decision that is reversible, small, invisible to customers and part of execution belongs with the agent. A decision that fails all four belongs with you. Most sit in between, which is where approvals come in.
A boundary for a typical small business
Here is how those questions might sort the common decisions for a small business spending a modest monthly budget on search and social ads. Your answers may differ. The point is to write yours down.
| Decision | Agent owns | Agent proposes | You keep |
|---|---|---|---|
| Pausing an ad whose cost per result is past your limit | Yes | ||
| Moving budget between live campaigns inside a fixed total | Yes, up to a set amount | Above that amount | |
| Adjusting bids within the range you set | Yes | ||
| Adding negative keywords to stop wasted clicks | Yes | ||
| Testing new headlines in your approved style | Yes | ||
| Writing copy that mentions a price, discount or guarantee | Yes | ||
| Raising the total monthly budget | Yes | ||
| Launching a new campaign for an existing product | Yes | ||
| Targeting a new region or a new kind of customer | Yes | ||
| Choosing which products to advertise at all | Yes | ||
| Deciding what a new customer is worth to you | Yes |
Notice what sits in the right-hand column. These are the choices that would still be yours if you hired a person to run your ads. An agent changes who does the daily work, not who owns the business.
One campaign, start to finish
The table above is easier to trust once you see it in motion. Here is one campaign cycle for a performance marketer at a subscription software company, launching a spring promotion on Google and Meta with a fixed budget for the quarter. Three agents take part. The Ads Manager runs the accounts, the Data Analyst works out what the results are worth, and the Chief of Staff turns it into the update leadership reads.
The cycle at a glance
| Stage | Ads Manager agent | Performance marketer | Other agents |
|---|---|---|---|
| 1. Brief, week 0 | Reads the brief and the last quarter’s results, and proposes a campaign structure and test plan | Sets the goal, budget, audiences and claims allowed, then approves the plan | Data Analyst confirms the trial-to-paid rate the goal is based on |
| 2. Setup, week 0 | Builds campaigns, ad sets and tracking in both accounts as drafts | Reviews the drafts and approves the launch | None |
| 3. A/B test, weeks 1 to 2 | Runs two headlines and two landing pages, splits budget evenly, watches for a clear winner | Approves the creative and copy before any version goes live | None |
| 4. Daily optimization, every day | Pauses losing ads, moves budget inside the total, adds negative keywords, flags broken pages | Reads the morning summary, answers anything held for approval | None |
| 5. Analysis, end of week 2 | Hands the test results to the Data Analyst | Reviews the analysis and picks the winning direction | Data Analyst joins ad data with trials and paid conversions, and checks whether the winner holds for paying customers |
| 6. Reporting, every Friday | Sends spend, results and changes to the Chief of Staff | Adds context and a recommendation | Chief of Staff writes the leadership update and lists the budget decision for Monday |
| 7. Next round, week 3 | Proposes the next test and a budget shift toward the winner | Approves, adjusts or says no | Data Analyst sets up the measure for the next test |
Read down the middle column and the performance marketer’s job becomes clear. They set the goal, approve what speaks for the brand, judge the analysis and decide where the money goes next. Everything in the left column happens without them having to open an ad account.
What it looks like in the thread
The day-to-day exchange is short, because the agent does the work before it asks.
Ads Manager, Tuesday morning
Overnight the Meta lookalike ad set spent EUR 410 at EUR 96
per trial, past your EUR 70 limit. Paused it. Moved EUR 300
of today's budget to Google brand search, which is at EUR 41
per trial. Nothing else needs you today.
Ads Manager, Thursday afternoon
Headline B is ahead on trials, 212 to 168, at 95% confidence.
Before I shift budget I have asked the Data Analyst to check
whether B's trials convert to paid at the same rate.
Data Analyst, Friday morning
B's trials convert to paid at 11%, A's at 18%. A wins on
paying customers despite fewer trials. Recommend keeping A
and testing a new B against it.
Performance marketer, Friday morning
Agreed. Keep A, test the pricing-page variant next.The Thursday message is the moment a single agent would have got it wrong. On its own numbers, headline B was the winner. Handing the question to the Data Analyst, which can see paid conversions in the warehouse, turned a click win into the right business decision.
Where people stay in the loop
Across the whole cycle, the marketer is asked to decide only a handful of times, and each is a decision a person should make.
- The brief. Goal, budget, audiences and what the ads may claim.
- Every new message to the public. Creative and copy before they go live.
- The analysis. Which result counts as the winner.
- Money above the line. Any change to the total budget or a shift above the agreed amount.
Everything else, the pausing, the budget moves inside the total, the tracking checks and the reports, is the agents’ work, done every day and recorded so it can be checked.
What the agent should never decide
A few decisions belong with you no matter how well the agent performs. They are worth naming in the brief so they are never in doubt.
- What a customer is worth. The target cost for a new customer comes from your margins and your plans. The agent works toward it but does not set it.
- What you promise. Prices, discounts, guarantees and claims about results are commitments. They come from you.
- Who you sell to. A new audience, a new region or a new product line changes your business, not just your ads.
- How much you spend in total. The agent can argue for more. Only you can say yes.
- What the brand is not. Topics, tones and comparisons you never want in an ad. Write them down, because the agent cannot guess them.
How to write the boundary into the brief
A boundary in your head does not help the agent. It needs to be written in plain words, the way you would brief a new hire. A useful brief covers these points.
- The goal. For example, new customers from search at or under a target cost.
- The limits. A total monthly budget, the most it may move in a day and any campaigns it may not touch.
- What it owns. The list of changes it can make without asking, each with its size limit.
- What it proposes. The changes that come to you, and what it should include when it asks, like the numbers behind the recommendation.
- What you keep. The decisions it should never act on, even if the data points that way.
- What to do when stuck. If results fall sharply or something looks wrong, it pauses the affected spend and tells you rather than trying to fix it alone.
Keep the brief short enough to read in two minutes. A long brief is harder for you to check, and harder to keep current.
Signs the line is in the wrong place
The agent’s record will tell you whether the boundary fits. Look for these patterns after the first few weeks.
| What you see | What it suggests | What to try |
|---|---|---|
| You approve almost every request without changes | The agent is asking about decisions it could own | Move that kind of change into what it owns, with a size limit |
| You often reject or rewrite its proposals | It is reaching for decisions that are yours | Tighten the brief or move those decisions to your column |
| Spend drifts toward one campaign over weeks | Many small moves add up to a big one | Add a weekly limit on how far budget can shift |
| Copy tests keep landing near the edge of your style | The style rules are too loose | Add examples of what is and is not your voice |
| You learn about a change only from a customer | Something it owns should have been a proposal | Move that decision and ask for a same-day note on similar changes |
None of these is a failure. They are the normal way a boundary gets tuned, the same way a new employee’s responsibilities settle over their first months.
Let the line move, slowly
The boundary is not fixed. Most owners start tight, with more decisions in the propose column than they will need later. As the record shows the agent making the calls you would have made, move one kind of decision at a time from proposing to owning.
Move them back just as readily. A change in your goals, a new product or a hard month is a good reason to pull decisions back to you for a while. The brief should follow the business, not the other way around.
The decisions in your own column rarely move. Those are the ones that make the business yours.
