Custom AI tools
Internal AI applications built for one job in your business and evaluated against it.
- Engagement
- Fixed-scope build
- Typical timeline
- 5–6 weeks
- Starts with
- Half a day watching the person who currently does the task, with their last twenty cases open.
A senior person in your team spends two days a week on one task that is mostly reading, comparing and deciding, pricing a tender, checking a claim, reviewing a supplier contract. No product on the market does that particular job, because the job is shaped entirely by rules that exist only in your company.
Hours of expert work compressed into a single interface.
What we build
Concrete artefacts, handed over and documented.
- 01
A single-purpose interface for that one task, not a chat box bolted onto your intranet
- 02
Your rules written down as a specification, reviewed line by line with the person who does the work today
- 03
A test set of 100–200 real past cases with the correct answer attached, so accuracy is a number rather than an impression
- 04
Defined behaviour when the tool is unsure: what it does, who it passes the case to, and how quickly
- 05
Running cost per case, measured at your real volume before anything goes live
What changes
The task moves from days to minutes, with a person reviewing rather than producing
The rules stop living in one expert’s head and become something the business owns
You know the cost per case and the accuracy rate before a customer is affected by either
How it runs
01Shadow the work
We sit with whoever does the task and record how the decision is actually made, including the parts not in the procedure.
02Build against the test set
Every change is scored on your past cases, so progress is visible and regressions are caught the same day.
03Pilot
The team uses it on live work with us watching, and the disagreements become the final round of fixes.
Chosen per project. Named here so you can see the shape of it.
- OpenAI
- Anthropic Claude
- Next.js
- Postgres
- Langfuse
Questions we get asked
Tell us the outcome, not the tooling.
Send us the situation you are in. We will tell you which discipline it belongs to, what we would do first and what it costs, including when the answer is to wait.