Case study · Creative agency
Hours handed back, with the judgement kept human
An independent creative agency brought AI into daily operations across both of its businesses in three months, and drew a hard line around the judgement its clients pay for.
Headline result
13–16%
of repeatable-task time returned in 90 days, against a 10% target
- Client
- Independent creative agency, two business units, 19 people
- Engagement
- Agency-wide AI integration program
- Stages
- Discover, Design, Deliver
- Where it got to
- Handed over after three months, with 18 of 19 people using it every week
The situation
The agency runs two businesses under one roof: a client-services practice covering PR, social, creative production and new business, and a talent management practice. The creative work was strong and the client relationships deep. Almost every recurring process ran by hand.
The brief had two halves. Return at least 10% of the time spent on repeatable tasks. And lift the quality and consistency of everything that reaches a client, so a report or a pitch stopped depending on who wrote it and how much time they had.
The real constraint wasn’t technical. This is an agency that sells human craft, and a program that read internally as “replace the people” would have failed in its first week, whatever it saved.
“We are down the very far end of how important humans are to what we sell.”
The approach
We spent a day on site running six structured sessions across creative, social, talent, new business and leadership, and came out with a costed map of eight processes worth automating.
Before building anything, we wrote down what the system could own and what only a person could. Then we built in three layers: a shared foundation everyone uses daily, four workflows aimed at the largest drains on time, and the governance and people structure that made both stick.
Discover
One day on site, six sessions, and a costed map of eight processes. The biggest drain was reporting: 16 hours per client, per month.
What the client keepsA priced view of where the hours go, before any build spend
Design
A Rules of Engagement framework with a signed position on environmental impact and responsible use. A 40-prompt library, client workspaces with brand voice pre-loaded, and a house tone-of-voice skill.
What the client keepsGovernance issued over the Managing Director’s name, and a foundation every new hire starts from
Deliver
Outreach pipeline, weekly status, monthly newsletter and client reporting rebuilt. Four role-based training sessions, AI Champions given ownership, daily tips for three months.
What the client keepsAll 16 signed deliverables plus nine builds outside scope, documented for a successor to learn without us
What’s next
A scoped next phase. The team has already started writing its own skills, including lead-finding and email-drafting tools nobody asked for.
What the client keepsThe choice of what to build next, and the people to build it
What it exposed
The processes looked healthier than they were
Reporting meant scrolling each post, reading figures off the screen and typing them into a spreadsheet, then typing the same figures again into a client deck. The data couldn’t simply be piped across. The analytics platform exported PDF and images only at the agency’s tier, extracted numbers came back garbled, and the social platforms block automated collection outright.
Two senior staff raised environmental impact, creative integrity and what this means for junior development early on. Their concerns shaped the Rules of Engagement directly, and the agency issued it in its own name.
Where we drew the line
Written down before anything was built: what the system may own, and what stays with a person.
The system may own
The date and sender of a reply
Read directly off the email
A figure visible in a screenshot
Extracted, then flagged if uncertain
A category matched to a known company
Resolved from existing records, never invented
A first draft in house style
Applying rules already written down
Only a person may own
Whether a lead is warm, cold or dead
The only commercial judgement on the sheet
Whether a creator suits a brand
Weights set by the account team
Whether a story runs this month
One named editor, one editorial pass
Anything that reaches a client
Every output is a draft until someone approves it
Results
- Repeatable-task time returned, against a 10% target
- 13–16%
- Hours returned in 90 days, about nine working weeks
- 349
- Team members active every week
- 18/19
- Multi-step working sessions in 90 days
- 1,300
Every signed deliverable shipped, along with nine builds the team asked for along the way.
Seat use hit 100% within three weeks of kickoff, and new seats were in use within days of being added.
The team now writes its own tools, which is the clearest sign the capability stayed when we left.
How we measured this
Adoption and time figures come from platform analytics, June to September 2026. Each output type carries a deliberately low minutes-saved assumption, and about a thousand direct actions in connected tools are left out of the total. The percentage assumes repeatable work is a quarter to a third of an 8,900-hour quarter.
What transfers
Decide what the system isn’t allowed to decide
Say it out loud, in writing, before anything is built. In a business whose product is human judgement, that line is what lets AI in at all, and it answers the question the sceptics are really asking.
Contact
Something worth changing?
Tell us what’s slow, expensive or stuck. If we’re not the right people, we’ll say so and point you somewhere better.