Scoped against a baseline
Before anything is built we measure the current process — throughput, error rate, cost per unit. Without that number there is no honest way to claim an improvement later.
Services
AI Automation
Turn a painful manual process into a system your team actually uses.
- Duration
- 4–8 weeks
- Price
- From £35,000
- Basis
- Fixed scope, fixed date
- Runs on
- Your infrastructure
This is the build. You already know which process is costing you — the weekly report nobody wants to own, the intake queue that never clears, the document review that takes nine days. We scope it, build it against a real evaluation set, integrate it with the systems you already run, and hand it to your engineers with the documentation to keep it alive.
A fit when
- Repetitive, admin-heavy work
- Reporting or follow-up that eats time every week
- Intake, onboarding, or support that stays too manual
You leave with
- A production system running on your infrastructure, in your repos and your CI
- An evaluation harness with a graded case set, wired into your pipeline
- A measured before-and-after against the baseline agreed at scoping
- Monitoring, logging and cost instrumentation for the live system
- A runbook, architecture documentation and a recorded handover session
- An on-call window after go-live while your team takes ownership
What's included
How the engagement is put together. Every piece below is in the fixed price. Nothing here is an optional extra we quote for later.
An evaluation harness first
We build the scoring set before the system, from your real cases with the correct answers attached. Every subsequent change is measured against it, which is what makes the thing safe to modify after we leave.
The system itself
Retrieval, extraction, classification, routing, agents — whatever the problem actually needs. Model-agnostic by construction, so you can change providers later on evidence rather than on faith.
Integration and human-in-the-loop design
It has to land inside the tools your team already opens. Where a decision carries risk, we design the review step and the audit trail alongside the automation, not after it.
Handover that holds
Runbook, architecture notes, a pairing week with your engineers, and an on-call window from us while your team settles in.
Timeline
What happens, and when. Dates are agreed at scoping and do not move without you agreeing to it in writing.
- 01Before we start
Scope and baseline
A fixed written scope, a fixed price and a fixed end date, plus the measurement of the current process we will be judged against.
- 02Week 1
Evaluation set and architecture
We assemble the graded cases and settle the architecture. Assumptions get corrected here, while changing them is still cheap.
- 03Weeks 2–6
Build in the open
A weekly demo of the working system and a shared channel. You see it as it takes shape rather than at a reveal.
- 04Final week
Integrate and hand over
Production cutover, documentation, runbook and a working session with the people who own it next.
Related work
What this looked like in practice.
FAQ
Questions about AI Automation. The site-wide questions are answered on the FAQ page.
What if we already have something half-built?
That is a common starting point. We audit what exists, keep what works, and are direct about what should be replaced. Rewrites are a last resort, not a default.
Who owns the code?
You do, from the first commit. Everything runs in your cloud accounts, your repositories and your CI. There is nothing to migrate when the engagement ends because nothing was ever hosted with us.
What happens when a model provider changes something?
The evaluation harness catches it. That is the main reason we build it first — you find out from a failing score rather than from a customer complaint, and you can swap models on evidence.
Other services
Not quite the right shape? The other two engagements, and when each one is the better call.
Get started
Let’s find out what AI is actually worth to you.
A 30-minute call, no deck. Bring the workflow that frustrates you most and we will tell you whether it is worth automating.
Typically replies within one business day.