AI consulting fromthe people who build
Most AI consulting produces a deck. Ours produces a working system: scoped with real numbers, measured against ground truth, and in production inside three to six months. We're the AI expert. You're the domain expert.
You've seen the demo. A model answers a question about a document, the room nods, and then the pilot quietly dies before it touches real work. The gap between a demo and a production system is engineering, and most of the market is selling the demo. Meanwhile the work your team repeats every day, reading documents, re-keying data, drafting the same letters, answering the same questions, was impossible to automate five years ago. Now it's merely very difficult. The winners of this wave will be the businesses that already do the work, not the startups circling them.
Measured first.
Then built.
Scope it like an engineering project
Before anything is built, you get an estimate you can make a decision from: what's feasible, what it costs, how long it takes, and what it should return. Just as important is what we rule out. If AI is the wrong tool for the problem, we say so in the scoping conversation, not in month four.
Build the measurement before the cleverness
The first thing we construct is a ground-truth test set, built with your team, because they know what a correct answer looks like better than anyone. An evaluation harness then scores every change against it. Without that, week twelve of an AI project is guesswork with confidence.
Engineer for production, not for the demo
Structured output between every stage, so each step hands the next something checkable. Confidence scores route uncertain cases to a person instead of letting the system guess, and a human sits at every decision that matters. This is the unglamorous work that separates a system from a demo, and it's most of the job.
Ship inside three to six months, then hand over properly
Something must be in production creating measurable value within three to six months, whatever the longer roadmap says. Otherwise an AI project becomes a permanent initiative that never shows a return. When it ships, the documentation, the code and the data are yours.
The shape of a problem
AI can actually solve
"We've never been able to figure this out. Could AI?" Usually not. If nobody in the business has ever done the task, there's nothing to learn from and no way to check the output. We'll tell you that in the first conversation, before it costs you anything.
People doing the same task over and over and banging their heads against it. Repetitive, rule-bound, and checkable against reality. Computers can now read and make simple decisions across almost any domain, and that unlocks precisely this work. The aim is the same team doing several times the work, not a smaller team.
Built, measured,
and running.
Our research engine answers professional questions from a corpus of more than forty thousand specialist documents. Every answer carries citations back to the source passage. The corpus tracks and re-indexes itself on a schedule, and an evaluation harness scores accuracy against known-good test sets before any change ships.
We come at this as operators, not observers. We've built our own ventures and taken them to market, and we've shipped for property, planning, retail, events and sport. Generalists in the business sense, which is what this work needs, because every automation lives inside a business process.
The method carries across unchanged: every claim cited back to source, and nothing ships without a score against ground truth.
What you walk away with
A system in production
Working software wired into your operations, with a person at every decision that matters, live inside three to six months. Pilots don't count.
Numbers you can defend
A ground-truth test set and an evaluation harness, handed over with the system. When we say it's accurate, we show the score. When a model upgrade lands, we can prove it broke nothing.
Ownership without lock-in
The code, the data and the documentation are yours. Built on Claude today and tied to no vendor: the evals make a future model swap provable rather than a leap of faith.
AI is a tool, not the strategy
Document Processing
A natural first step. Extract and check what your business receives, and answer questions from what it already holds, with citations.
Learn moreWorkflow Automation
Automation that takes a task from arrival to done: a task completer rather than another copilot, with exceptions routed to a person.
Learn moreTechnology Strategy & Consulting
AI is one tool in a wider technology picture. The fractional CTO engagement looks at the whole of it: systems, data, process and product, then decides where AI belongs.
Learn moreManaged Technology
An AI system needs somewhere to live and someone watching it. We host what we build on our own infrastructure, monitor it, and keep the evals running after handover.
Learn moreGot work your team
shouldn't still be doing by hand?
Tell us what they repeat every day. If AI is the right tool, we'll scope it with real numbers. If it's the wrong tool, we'll say so in the first conversation.