Skip to content
Chandler Nguyen
Leave

Opening the floor…

The operating model

Day one at the agency. The course's rule of thumb: AI handles 75–80% of ongoing operations, while people keep business context, client intelligence, strategic direction and quality judgment. It is a rule of thumb, not a count of cards.

Place each piece of your agency's work, then name who checks what AI does. Each colleague can check only 2 pieces of work this week.

Automate, augment, protect or decline?

  • Automate: AI does it end to end, a person reviews it from time to time
  • Augment: AI drafts, a person approves before it goes out
  • Protect: a person does it, AI helps with inputs
  • Decline: not automated, even if it could be
  • Pulling weekly performance data from five platforms into one report
  • Updating the budget pacing tracker
  • First-draft channel summaries
  • Persona research
  • The brand guide
  • Ad copy
  • Campaign setup
  • The launch go or no-go
  • The client's monthly review
  • Crisis communications
  • A culturally sensitive market entry

Who checks it? The course: validation needs platform depth, not seniority.

  • Megan, account lead: the client relationship
  • Marcus, paid media: in the platforms daily
  • Emily, research
  • Jake, planning
  • Lauren, analytics: in the platforms daily
  • Ryan, measurement: in the platforms daily

AI numbers reach the client unless someone in the platforms daily (Marcus, Lauren or Ryan) reads them first. A report reaches the client with numbers nobody in the platforms checked. The course's own example of what slips through: an AI reported a UK supplement cost per click of $0.02 with complete confidence, and it was wrong (a course example, not GreenLeaf data).

An AI-configured campaign needs someone who can verify targeting and check tracking. A campaign goes live with targeting and tracking nobody in the platforms verified. The course: every AI-configured campaign needs someone who can catch platform edge cases.

Every check by a platform specialist is one they cannot spend elsewhere this week. Someone who lives in the platforms reads the AI's numbers before they go out: the person who can tell when one cannot be right. The course: that validation layer is often your most technical people, not your most senior.

The course names this as work AI should handle: keeping it human leaves recapturable time unused. Your people spend the week on work the course names as AI's: pulling platform data, updating the pacing tracker, drafting channel summaries. The course suggests most teams find 15–25% of their hours in this bucket, time that could go to the client.

The course: AI checks; humans decide. Launch day, and AI takes the go or no-go. The course splits it: AI checks tracking, budgets, creative and geography; a person decides.

If this goes wrong in public, an AI-drafted reply is a reputational risk. GreenLeaf is in the news for the wrong reasons, and the first reply is AI-drafted. The course puts crisis communications in Decline: AI-generated responses are a reputational risk.

AI can be culturally wrong in ways no model detects. The UK launch's culturally sensitive work goes through AI. The course's own lesson: AI translations were culturally wrong in ways no model could detect.

The monthly review is where a person learns what makes the client look good. The client's monthly review arrives as an AI-written deck. The course lists client relationship conversations under Decline: the human connection is the value.

Keeping this human means a person who knows the client owns it every month. The client's monthly review stays a conversation with a person who knows what makes the client look good.

The course's position

  • Pulling weekly performance data from five platforms into one report: Automate, checked by Lauren. One of the course's three first automation candidates for GreenLeaf.
  • Updating the budget pacing tracker: Automate, checked by Ryan. One of the course's three first automation candidates for GreenLeaf.
  • First-draft channel summaries: Augment, checked by Lauren. The course names first-draft summaries as work AI should handle; as drafts, a person approves them before they go out.
  • Persona research: Augment, checked by Emily. The course has AI build behavioural personas; deciding which audience matters stays with a person.
  • The brand guide: Protect. The course: the quality of AI creative is directly proportional to the brand guide you feed it.
  • Ad copy: Augment, checked by Marcus. AI drafts creative from the brand guide; a person approves it before it runs.
  • Campaign setup: Automate, checked by Marcus. The course says stop selling setup as premium labour, and still have someone verify targeting and tracking.
  • The launch go or no-go: Protect. The course: AI checks; humans decide.
  • The client's monthly review: Decline. The course lists client relationship conversations under Decline, where the human connection is the value; this experience treats the monthly review as one.
  • Crisis communications: Decline. The course lists crisis communications under Decline: AI-generated responses are a reputational risk.
  • A culturally sensitive market entry: Decline. The course lists culturally sensitive market entry under Decline: AI can be culturally wrong in ways no model detects.

Your operating model decides which errors you catch and where your people's hours go. The course puts checking with whoever lives in the platforms, and adds Decline for work that should stay human even when AI could do it. It suggests looking for 15–25% recapturable time in the work AI should handle: an audit opportunity, not a promised saving.