The problem
The craft was real, but it lived in a PDF the client never opened.
Three clients had started drafting posts in ChatGPT and asked, at renewal, what the retainer was for. The agency's answer, on-brand, fact-checked, consistent, was true but invisible: nothing the client could point at and say 'that's what we're paying for'.
Internally the cost was ~30 hours a week of an account manager rewriting drafts to sound right, at roughly €30/hour, about €3,600/month of invisible labour that never showed up as a line item.
The approach
Productise the agency's standards, then let the client work inside them.
Phase 1: a voice profile per client built from their ~200 best posts, so the tool writes in the client's register, not generic agency-speak, without anyone hand-writing a style guide.
Phase 2: an approved-claim library, the tool will not assert a claim that isn't in it, plus channel formatting and length rules baked in, so a draft comes out shaped for where it's going.
Phase 3: a white-label workspace, one per client, under the agency's brand. The client drafts; the agency approves. Version history makes tone-drift visible instead of a surprise at publish.
We stopped defending the retainer and started onboarding clients into something. The fee conversation flipped from 'why so much' to 'can we add another seat'.
What was messy
Where the first version leaked.
The first voice profiles were built from too small a sample and drifted toward a flat, safe tone. Widening to ~200 posts per client and hand-picking the strongest 40 as anchors fixed it.
The claim library was the tedious part: legal-sensitive clients needed every approvable claim entered before the tool was safe to hand over. That was a week of unglamorous setup per such client, not a model problem.
A couple of clients tried to use it as a full autopilot. We kept the agency-approves gate hard: nothing publishes from the client's seat without an agency sign-off.
The outcome
The retainer expanded instead of eroding.
- Rewriting hours per week~30 h → ~6 h−80%
- New recurring revenue (brand-system seat)€0 → ~€5.5k/mo11 seats
- Drafts shipping without a full rewrite→ ~80%agency-reported
- At-risk clients moved up a tier3directional, n=18
How we measured it
Baseline, sample and method, so the numbers above are checkable.
Baseline: the 6 months before the pilot, rewriting hours from the account team's time tracking, retainer values from the agency's billing.
Pilot: 8 weeks, tool live across the 18 retained clients; 11 of them took the new 'brand system' line at €499/month.
New recurring revenue = €499 × 11 seats ≈ €5,489/month; rewriting-saved = the drop from ~30 to ~6 hours/week at ~€30/hour. Both are agency-reported, not independently audited.
The retention effect (3 at-risk clients moved up a tier) is directional over one 8-week window, not a controlled study.
What we did not automate
The agency stayed the owner of quality.
Nothing publishes from a client's seat without an agency approval. The tool drafts inside the rules; the agency signs off.
We did not train models on client data, voice profiles are prompt-level, built from the client's own public posts.
Strategy, campaign ideas and anything net-new stayed with the agency team. The tool covers drafting, not thinking.
What's next
The same white-label pattern, sold as a product line.
Once the first workspaces existed, the agency started pitching the 'brand system' seat as a standard part of onboarding, not a rescue for at-risk accounts.
The same shape, voice profile + approved-claim library + approval gate, is being extended to email and short-form video scripts for the same clients.