[ 04 — AI AUTOMATION ]
Hand the repetitive work to AI agents that read, reply, file and update for you.
[ 01 — THE PROBLEM ]
What this usually looks like
Before
- Someone spends hours a week retyping information that already exists somewhere else.
- The same twenty customer questions get answered manually every day.
- Invoices and receipts are read by a human and typed into accounting software.
- The Monday report takes a morning to assemble and is out of date by Tuesday.
After
- Systems that talk to each other directly, so information is entered once.
- A first-draft reply generated in seconds, with a person approving anything that commits the business.
- Documents read, extracted and filed automatically, with low-confidence results queued for review.
- Reports assembled and delivered on a schedule, from the live data rather than from last week's export.
[ 02 — WHAT YOU GET ]
Deliverables, not deliverable-sounding words
A process audit
Which tasks are worth automating, what each one costs you now, and which ones to leave alone.
Working automations
Running on your infrastructure or ours, connected to the tools you already use.
Human-in-the-loop controls
Approval steps wherever a mistake would be expensive, with a clear queue rather than a silent decision.
Monitoring and alerts
You find out when an automation fails, rather than finding out from a customer.
A runbook
What each automation does, how to pause it, and what to check when something looks wrong.
[ 03 — HOW IT WORKS ]
The steps, and how long each one takes
- 01
Audit the processes
2–4 daysWalk through a week of real work and count what the repetitive parts actually cost.
- 02
Pick the first one
1 dayHigh volume, clear rules, low blast radius if it gets something wrong. Prove it before scaling.
- 03
Build and shadow
1–2 weeksThe automation runs alongside the human for a week and the outputs are compared before it takes over.
- 04
Hand over control
2–3 daysIt takes the work, with the approval step where it belongs and monitoring on.
- 05
Expand
ongoingThe next process, then the next. Each one is cheaper than the last because the plumbing exists.
[ 04 — EXAMPLES ]
What people use this for
Invoice and receipt capture
Documents arrive by email or WhatsApp, are read, checked against the order and filed into accounting.
Enquiry triage
Incoming messages classified, routed to the right person and answered with a draft that only needs approval.
Reporting on a schedule
Figures pulled from the systems that hold them, assembled and delivered before the meeting rather than during it.
Reading invoices, answering the same enquiry, copying figures between systems, assembling the Monday report: AI does these well and cheaply.
Where AI belongs, and where it does not
AI is good at reading messy input, drafting, classifying and summarising. It is bad at arithmetic you must be able to audit, at rules that must hold every single time, and at anything where a confident wrong answer is expensive. We use it for the first list and ordinary code for the second, and we will tell you which is which before you pay for anything.
A human where it counts
Every automation that commits the business has an approval step. The AI does the work; a person presses send. Over time, the steps that have proven themselves can be loosened — but that is a decision made with evidence, not at the start.
[ 05 — STACK ]
What we build it with
Chosen because they are boring, well supported and easy to hire for. Nothing here will be abandoned next year.
- n8n
- Python
- Claude
- OpenAI
- PostgreSQL
- Docker
- FastAPI
[ 08 — FAQ ]
AI Automation: questions people ask
[ 09 — ALSO IN THIS PILLAR ]
Related services
Let's talk about ai automation.
Tell us the problem. We will tell you what we would build, what it costs and how long it takes — before you commit to anything.