service
AI solutions and agents
A business with a specific repetitive knowledge task — triage, retrieval, drafting, classification — where AI might remove real work.
What you get
- A narrow, testable use case rather than a general assistant.
- An evaluation set built from your real examples, and measured accuracy on it.
- Clear human review points where the model is not reliable enough to act alone.
- Cost per operation and a statement of the failure modes we actually observed.
How this differs from business automation
Deterministic automation should handle anything with a defined rule. AI is for the residual where judgement is needed, and it carries accuracy and cost uncertainty that a rule does not.
How we approach it
- We start from a decision that is currently made by a person and check whether it can be evaluated — if it cannot be measured, it cannot be improved.
- We keep a human in the loop wherever a wrong answer would be costly or hard to reverse.
- We define what the system must refuse to do before defining what it should do.
What is out of scope
- AI where a deterministic rule would be more reliable and easier to audit. We will say so.
- Claims about accuracy we have not measured on your data.
How it is priced
Cost is driven by how much labelled data exists to evaluate the system against; without it, the work becomes an experiment with no finish line.
See the pricing page for how that works in practice.
How to start
The first conversation is free and usually takes half an hour. Bring the problem as you experience it — what is not working and what you have already tried.
Tell us what you need
The more specific you are about the problem, the more useful the first reply will be. We read every enquiry ourselves.