The agency was competent. It was also slow, vague on reporting and expensive to manage. The retainer landed every month, whether the campaigns were winning or losing.
The job
The goal was clear: drive parent enquiries that turned into enrolments. The workflow had access to Google Ads, Meta Ads, the website and the CRM.
The agent ran the campaigns end to end. It paused underperforming creative, scaled winners, tested audiences, adjusted bids against conversion data, wrote copy and built landing pages. It also sent a weekly performance summary without waiting for someone to ask.
The book's test was enrolments, not how much activity the system produced.
The economics
After three months, the result was consistent. ROAS was identical to the agency's result. Conversion from enquiry to enrolment was unchanged. There was no uplift, and there was no decline.
The workflow also produced work the agency had not included in the retainer: new landing pages for audience segments, copy variations, tests and reporting.
What stayed human
The book draws a hard boundary around brand work. AI was useful at the high-intent end of the funnel, where a customer was already looking for the service or needed a final nudge.
Brand identity, founder story and creative direction still required human taste. Those decisions shape how a company feels over time. The performance workflow did not remove that job.
The split was practical: use the agent for measurable bottom-of-funnel work, and keep people on the work where taste and distinction carry the result.
Run the test
List every retainer that produces a measurable number. For each one, ask whether an agent could run the work against a clear goal and real source data.
Then separate the measurable workflow from the judgement work. If taste is the reason you pay the retainer, keep the human. If the work is repetitive optimisation against a number, test the system.