0h
Founder / week — out of late-night admin
03 · Case study · TUW + MILK.
240 hours a week. In one month.
A five-person F&B operations team running a national distribution business + a customer-facing dairy brand. Twelve workflows. One Mac Mini. Week-one numbers.
01 · Headline
The whole team got a second self.
Five staff, twelve workflows, one Mac Mini. Numbers measured against a real baseline — not vibes.
0h
Manual hours removed every week
0×
ROI vs running cost of the AI
0
Workflows running on autopilot
0
Staff with their own integration
02 · The business
Two brands, five staff, no margin for waste.
A national keg supplier and dairy operations consultancy, plus its customer-facing premium milk brand. Both run out of the same warehouse, with the same five-person team handling logistics, accounts, customer service, marketing, and back-office for both.
The brief, in the founder's words: "I'm working until 1am. Something has to change before someone breaks."
03 · The build
Five agents. Twelve workflows. One month.
We mapped the work, ranked it, and built one custom agent per team member — each role-shaped, each living inside the tools the human already used (Outlook, WhatsApp, ordering portal, warehouse spreadsheets).
- → Founder agent — morning briefing, inbox triage, sales pipeline summary, monthly P&L view, anomaly flags
- → Ops agent — daily warehouse standup, supplier alerts, stock low triggers, picking sheet drafts
- → Accounts agent — weekly invoicing run, debtor chase emails, payment reconciliation, BAS prep
- → Customer agent — quote follow-ups, delivery status replies, complaint triage, brand inbox
- → Marketing agent — LinkedIn post drafts, weekly content radar, case-study writeup support, partner comms
Each agent runs on the team's own Mac Mini — no cloud subscription, no per-seat fees, no data leaving the warehouse. Every agent has guardrails, a kill switch, and a monthly impact report.
04 · Week one results
Per-role hours back.
All numbers are measured baseline minutes per task × frequency, before vs after. Detailed methodology available on request.
0h
Ops / week — daily standup + supplier work
0h
Accounts / week — invoicing + reconciliation
0h
Customer / week — quote follow-up + inbox
Marketing agent added another ~34h/week once the LinkedIn workflow stabilised in week three — making the headline figure of ~240 hours/week across the team conservative by month two.
05 · The economics
Roughly 9× ROI vs the running cost of the AI itself.
The actual API + infrastructure cost to run the five-agent system sits well inside one part-time salary equivalent. The hours reclaimed across the team — when valued at conservative replacement rates — comes back at roughly nine times that figure, every week.
And those hours don't go back into more of the same. They go into the work that actually grows the business: customer relationships, product, strategy. The boring stuff is what the AI is for.
// Detailed cost breakdown shared under NDA on request.
06 · What it actually means
The founder stopped opening the laptop at 11pm.
That's the whole pitch. The rest is implementation detail.
Want this in your business?
30-minute discovery call. We'll map your top recurring workflows in real time and tell you which to integrate first.