Case study
An AI Opportunity Audit for a US marketing agency1.2 people they were paying for, and never hired.
Twelve account managers. Two project management systems. And in between, people carrying the same information from one to the other by hand. This is how that became a number, and what the audit recommended.
01 The ask
They knew it was slow. They didn't have a number.
Leadership already knew the two-system setup was inefficient. What they did not have was a number, an order to fix things in, or a view of what would break if they changed it.
They had also already decided that AI was part of the answer. They wanted help finding where.
So before anything was built, the operation got diagnosed.
02 What already worked
It opened with what was already right.
Most audits open with a crisis. This one opened with a list of what the agency had already built, because none of it needed replacing.
- In placeNaming conventionsAgreed and used
- In placeFolder structuresFinalized
- In placeAn estimate workbookIt existed, and people used it
- In placeInternal review standardsWritten down
The information was organized. It was organized in places that could not talk.
03 Where it broke
Four findings, all from their own systems.
Not discipline, and not people. Answering one question about one project meant opening each platform in turn and putting the answer together by hand.
- 0systems that could read across all their toolsEvery answer was assembled by a person.
- 2project management systems doing one jobOne for client work, one for internal delivery.
- 10 to 20%of account management time on adminBy their own time study, not client work.
- About 5%of estimates reached the repository meant to hold themNothing routed anything into it.
10 to 20 percent is easy to nod at. So it got turned into people.
04 The number
Ten percent of each week, from each of twelve people.
12 account managers × 10% of each week
The floor of their own range. At the top of it, 20 percent, it is 2.4. I led with the floor, because the low end of a client's own number is the one nobody can argue with.
Spread thinly enough that nobody ever saw it leaving.
05 The map
Then their chain, mapped in their words.
I mapped how one piece of work moves through the agency with their team, in their vocabulary, before naming a single tool. Nine steps, from the client's problem to the financial records.
Before approvalMost of the friction
- 1Client problem
- 2Estimate
- 3Approval
After approvalProject management
- 4Project creation
- 5Specification
- 6Delivery
- 7Time entry
- 8Invoicing
- 9Financial records
Three of nine steps held most of the friction. All three come before the client says yes.
Everything after approval was project management, not decision-making. So the work was scoped to the first three.
06 Where AI stops
Quoting depended on who was doing it.
Every estimate carried judgment: what the client's business was worth, what they could pay, the history of the relationship, how much account management it would need. Leadership did not want that judgment removed. They wanted it visible, and consistent.
And they were clear on one thing: nothing with money attached would happen without a person reviewing it. A tool that replaced judgment instead of supporting it would not get used.
The AI
Makes the judgment visible
- Brings out the factors experienced staff already weigh
- Records which factors informed each quote
- Shows a reviewer the reasoning behind it
The person
Keeps the judgment
- Judges what the client's business and history are worth
- Adjusts the quote, with the reasoning in front of them
- Decides anything with money attached
07 The plan
Three phases. Free to stop after each one.
An open-ended AI project is what owners fear most. So the roadmap was built in an order, with the reason for the order, and a clean place to stop after each phase.
- Phase one
The data layer
Which tools stay, and how each one is really used. A database the agency owns, so changing AI models later does not mean rebuilding. Every step written down first.
- Phase two
The estimate layer
Which parts of quoting can be automated and which cannot, and a responsibility map that puts AI inside it, not beside it. Financial decisions stay with people.
- Phase three
Implementation scope
A minimum build that leaves room to grow, with every assumption stated, so a change in scope can be spotted instead of argued about.
08 What they kept
What they walked away holding.
- Their operating chain, written downEnd to end, including the parts that had only ever existed in conversation.
- The cost, counted in people1.2 full-time equivalents, from their own time study.
- Clear AI and human layersAgreed before any tool was built, not discovered afterwards.
- A priced first phaseSomething leadership could approve or decline, not a recommendation to think about.
The most valuable thing the audit produced was not a list of things to automate. It was a clear account of which parts of the business should stay exactly as they are.
The agency's name, revenue, total headcount, platforms and location are withheld. Every figure on this page came from their own systems.
Your agency's version
Every agency has a 1.2 somewhere.
The AI Opportunity Audit finds yours, from your own numbers. It starts with a free 30-minute call and a straight answer on whether the audit is worth buying.