Most AI projects fail because nobody asked "should we?"
We start from your business, not from a model. If AI is the right answer we will build it end to end. If a database query and a better process would do the same job for a tenth of the cost, we'll tell you that instead.
Book an AI assessmentHow an engagement runs
Four stages, and you can stop after any of them. Most clients start at assessment and decide from there — no long lock-in before there's evidence.
- 011–2 weeks
Assessment
We map your workflows, data and cost centres, then rank the places AI could realistically help. You get a written opportunity list with effort, risk and expected payback against each item — including the ones we recommend you skip.
- 022–4 weeks
Prototype
We build the single highest-value idea as a working prototype on your real data, not a demo dataset. You can put it in front of your team and judge it on output quality before committing to a full build.
- 034–12 weeks
Production
The prototype becomes a real system: evaluation harness, guardrails, monitoring, cost controls and a human-in-the-loop path for anything high stakes. It integrates with the tools your team already uses.
- 04Retained
Ongoing
Models change monthly. We track quality, swap in better or cheaper models as they land, and keep your costs from quietly drifting upward.
What we build — and what we talk clients out of
Worth building
- Custom agentic agents scoped to your business specification — one workflow or a multi-agent system that plans and acts inside it
- Document and email triage that removes hours of manual reading each week
- Retrieval assistants grounded in your own policies, catalogues or case history
- Customer-facing assistants with strict scope and a clean handoff to a human
- Classification and extraction pipelines running quietly behind existing software
- Drafting tools that give staff a strong first version to edit, never a final answer
Usually a bad idea
- A chatbot bolted onto a site with no defined job to do
- Fine-tuning a model when a well-structured prompt and good data would do
- Automating a process nobody has documented or agreed on yet
- AI as a replacement for a decision your team should be making
- Anything where a wrong answer creates legal or safety exposure with no human check
Indicative pricing
Assessment from £1,200
Prototypes typically £3k–£8k. Production systems are quoted per scope. If we go on to build, the assessment fee comes off the build.