A solo media buyer approves agent-proposed Meta budget changes before they are applied

The task

One person handled all paid acquisition for the Canadian business. Every day that meant going through each channel, campaign, ad set and creative, reading the numbers and deciding where budget should move and which creatives to pause or scale. As the account grew, a single person no longer had enough hours to review everything by hand.

What they did

The buyer first built a custom agent in Cursor, connected to the Converge attribution data and to Meta. It produced daily reports and creative-testing recommendations with its reasoning, but according to the source it misread data at times, lost older instructions and had no guardrails, so the buyer still checked and carried out every change himself. The buyer then moved to the Converge media buying agent. It reads the account history, current pacing, channel split and product targets before it suggests anything, and it uses the same attribution figures the buyer already works with. When the buyer gives it a rule, such as a threshold for stopping a weak creative, the agent stores that rule in its memory and applies it on later runs.

What the source reports

According to the source, the agent now carries out more than 90 percent of Meta budget changes, and performance analysis takes more than 10 fewer hours per week. The source states that ad spend for the Canadian business grew by more than 130 percent over the year, and the buyer estimates that about 80 percent of the time once spent on media buying now goes to other growth work, according to the source.

Where the approval sits

The agent presents each proposed change together with the data and reasoning behind it. The buyer checks the proposal against the same attribution views used for a manual decision and approves it. Only then does the agent apply the change in Meta. The source describes this approval layer as the condition for letting the agent work on a large budget.

Worth knowing

The source is a customer story published by Converge, the vendor of the agent, so the figures are the vendor's and the customer's own and were not checked independently. The page carries no date. It describes one buyer's setup at a large brand, and the earlier self-built agent shows what can go wrong without guardrails: misread data and forgotten instructions. Whether every single change is approved one by one is not spelled out in detail.

More in this category

• Media and ad operations

  Google Ads suggests AI-generated ad assets that the advertiser reviews before they serve

  Google Ads documents a review space where Google AI proposes new headlines, descriptions, images and sitelinks for a campaign. The advertiser adds or dismisses each suggestion, and nothing from this list runs in an ad until it has been added.

  Read the entry
• Media and ad operations

  A Dutch agency has Claude draft an out-of-home media plan and two people check it

  A digital agency in the Netherlands used Claude to turn a client's campaign goals into a plan for digital out-of-home screens. A buyer approved the plan before the booking went ahead, and a second person checked the buyer's work.

  Read the entry
• Media and ad operations

  A Google Ads agency drafts account audits with Claude and has staff edit each section

  A New Zealand Google Partner agency built an audit tool in which Claude writes a 17-section report from Google Ads API data and the agency's own best-practice file. Staff review the output and can edit every section before the audit is exported as a client PDF.

  Read the entry