Should you let Meta's AI assistant change your client's budgets?

·5 min read·Meta Ads

Meta's business assistant in Ads Manager is being tested with the ability to change budgets, update targeting and generate creative from a chat prompt; on an agency's own account that is a convenience, but on a client's account every change has to be explained to the client afterwards, so use the assistant to find and draft changes and keep the final click with a named person on your team.

At Advertising Week New York on 6 October 2026, Meta showed the next version of the business assistant inside Ads Manager. It remembers more about the business (what was tested last quarter, seasonality, stated goals), and Meta is testing what it calls agentic actions: generating creative, updating targeting and changing an ad's budget from a conversational prompt. Scheduled tasks, such as recurring performance reports and alerts, are in test too.

These actions are being tested, not rolled out to everyone. But the direction is clear, and agencies should decide how they'll use it before it arrives on their clients' accounts rather than after.

Your own account and a client's account are different questions

For a brand running its own ads, letting an assistant raise a budget is a judgement about its own money. For an agency, the money belongs to the client, and every change will be discussed with them: why did spend go up on Tuesday, why did the audience change, who decided. "Meta's assistant did it" is not an answer a client accepts.

On a client's account, the question is never only whether a change is good. It is whether someone on your team can explain it on the next call.

What goes wrong when changes happen on their own

  • Learning resets. A significant budget or targeting change sends an ad set back into Meta's learning phase. Several small automatic edits can keep it there. See every edit restarts the clock.
  • Spend that runs ahead of the client's plan. A client with a fixed monthly budget can be spent out by the 20th if increases compound.
  • Optimising for the wrong thing. An assistant judging success on cost per lead will scale a campaign whose cheap leads are junk. See cheap leads cost more.
  • No record a client can follow. Ads Manager keeps an activity history, but it shows what changed, not the reason, and the reason is what the client asks for.

How to use it without losing control

  • Use it to look, not to act. Asking it what changed this week or which ad set is fatiguing is useful and changes nothing. The same goes for outside assistants: if you connect Claude or ChatGPT to an ad account, limit what it can change first.
  • Let it draft, then approve the change yourself. Read the suggested change, decide, and make it, so a person owns every edit on a client's account.
  • Set an account spending limit on every client account, so no change, by anyone or anything, can spend past what the client agreed.
  • Keep a short change log per client with the reason for each change. It turns the monthly report from a defence into an explanation.
  • Judge changes after Meta's learning has settled, not the next morning. See how much to increase a Meta ad budget.

Admetriq watches for exactly this across every client account — delivery drifting outside the brief, one placement quietly taking the budget, a cost per lead that has fallen for the wrong reason.

What Admetriq doesStart free trial