Round-Robin Media Buying

A single shop running paid media for solar operators, 50+ accounts in the book, together spending somewhere between $300K and $400K a month. Their customers paid for leads. The ads themselves were the agency's problem, and the operators mostly stayed out of the accounts.

Six media buyers, each sitting on their own set of accounts. That arrangement breaks in ordinary ways. A buyer takes a week off and their accounts drift. A buyer gets good at one slice of the portfolio and nobody else learns it. And the work that actually needed doing on a given day wasn't shaped like account ownership at all. It was a scatter of small corrections spread across the whole book.

Wire the corrections to the data and route them to whoever is working. Triggers watched every account for deviation from its own average and for anomalous results, and each trip generated a task. Around 80 task types: scale an ad set up 10%, optimize at the ad level for a specific customer, tweak a campaign's optimization settings. Make.com built the list. The list was then randomized and distributed across the six buyers. No buyer owned an account, and whoever was on shift took the next task.

• Make.com automations
• Supermetrics
• BigQuery
• Google Data Studio
• Facebook Ads

First time the shop had run round-robin rather than dedicated ownership, and it made the floor much easier to manage. Time off stopped being an event, because the queue didn't care who picked a task up. The playbook underneath stayed simple: compare the current day against the previous three and the previous seven, kill what's getting worse, keep what's getting better.

Account performance stopped being a function of which buyer got assigned to it. The work turned into a queue with a known shape, which is the kind of thing you can staff, cover, and hand to somebody new. It also gave the agency an answer to what was happening across the portfolio, because every task in the queue came from a trigger that had already read the data.

Assigning accounts to people looks like ownership and behaves like a single point of failure. Route the tasks instead and a buyer's day off costs you nothing. The routing only works if the triggers are good, though. Distributing useless tasks fairly still wastes the day.