No discovery phase that never ends.
Six weeks from first conversation to something working, with a decision point after each stage. You are never more than a few hundred pounds committed before you have seen enough to judge whether this is worth doing.
- Week 0 — 25 minutes, free
A short call
You describe what eats the day. I ask what happens when it goes wrong. By the end we both know whether there is anything here worth doing. If there is not, I say so — that call costs you nothing and I would rather not sell you something that will not pay for itself.
What I need from you: Half an hour and honesty about what actually happens.
- Week 1 — Efficiency Review, £450
Watch the work, count the hours
Two sessions with the people doing the job. I time a sample of real work, count volumes, and map the process as it exists rather than as it was designed. I look at what happens on a bad day, because the exceptions are what decide whether automation is realistic.
What I need from you: About two hours of your team’s time, and access to a sample of real (anonymised) work.
- Week 2 — findings
A ranked list and a number
You get a written map of every repetitive task with hours and loaded cost attached, ranked by value against difficulty. It includes the things I would not automate and why. Then a fixed-price proposal for the top item, with acceptance criteria written before anything is built.
What I need from you: A decision. The review is yours to keep and act on however you like.
- Weeks 3–5 — build
Built against your own cases
I build with a sample of your genuine historic work. Deterministic rules do the deterministic parts. AI gets used only where reading messy human language genuinely helps. You see something running by the end of the second week, not at the end.
What I need from you: One person as a point of contact, and answers when a rule is ambiguous.
- Week 5 — acceptance
You test it before it counts
We run the acceptance tests agreed at the start, including the edge cases and the failure behaviour. You confirm the outputs are right on your own examples. Nothing goes live until that passes and your team knows the manual fallback.
What I need from you: A sign-off, once you are satisfied.
- Week 6 onwards
Live, measured, and yours
Thirty days of included fixes. We compare the after figure to the before figure we agreed in week one. From there you either run it yourself with the documentation, or take a Care Plan if you would rather I kept an eye on it.
What I need from you: Nothing, unless you want changes.
Five rules I do not bend.
Understand the work
I sit with the people doing the job and watch what actually happens, including the workarounds nobody documents. Owners describe the process they designed; teams run the process that exists.
Measure and rank
Every repetitive task gets a frequency, a time and a loaded cost. Then it gets ranked by value against difficulty, so we start where the return is clearest — and I tell you which things are not worth automating.
Build against your real cases
I build with a sample of your genuine historic work, not invented examples. Deterministic rules do the deterministic parts. AI is used only where judgement about messy language genuinely helps.
Test, then hand over
You get acceptance tests, edge cases and a documented manual fallback. Nothing goes live until it passes on your examples and your team knows what to do when it does not.
Measure the result
We agreed the before figure at the start. We measure the after figure against it. If the number is not there, I tell you and we fix it or stop.
Deterministic first, AI second
If a task can be done with a rule, it gets a rule. Rules are cheaper, faster, testable, and they behave the same way every time. A language model earns its place only where the input is genuinely messy human language and judgement helps — and even then a person checks anything that leaves the building.
Am I the right fit for you?
I would rather turn work down than take on something that will not work. Roughly, this is a good fit if:
- You employ between 3 and 50 people.
- At least one person spends five or more hours a week on repetitive work.
- You can name the process and the person who owns it.
- The person on the call can approve a few thousand pounds of spend.
- You want to remove work, not add an AI feature to your website.
And a poor fit if
- You want a chatbot on your website. That is a different job and other people do it well.
- The process changes every week and nobody can describe it. Automate a moving target and you get an expensive mess.
- The volume is low. Twenty minutes a week is not worth £1,800 — I will tell you that on the first call.
- You need something to make a decision with no human involved, where being wrong is expensive. I do not build that, at least not as a first project.
Start with the 25-minute call
No charge, no pitch deck. If there is nothing worth automating you will know by the end of it, and so will I.