How we work
Research first, people second, code third
Most AI consulting produces a recommendation and stops. We treat the recommendation as the cheap part. What follows is the work: building the systems, and getting an organisation to actually use them.
The analysis gets deeper, in stages you control
We do not run one big discovery exercise and hand over a document. The work runs in passes, and each pass is scoped by what the previous one found: read the industry, form hypotheses about your business, test those hypotheses with your people, then cost what survives.
Every pass ends with a decision that is yours, not ours: go deeper, change direction, or stop. Nothing escalates into paid scope on its own.
Every assertion is sourced
If we tell you something about your industry, your competitors or your own operation, it carries a source -- research we can show you, something one of your people said, or a public reference. Assertions that cannot be sourced are marked as speculation or they do not ship.
Numbers carry their assumptions with them. An ROI figure without a stated assumption set is decoration, not analysis.
We plan for people not telling us the truth
When an organisation brings in AI consultants, staff often and correctly read it as a question about their jobs. Asking those same people to be candid with a vendor is asking a lot, and pretending otherwise produces a report full of agreeable nonsense.
So we say what actually changes for people rather than promising nobody is affected, we offer an anonymous route for the things nobody wants attributed, and we only report anonymised material in aggregate. Where candour is low, we treat that as a finding about adoption risk, not as an absence of data.
Your particulars stay yours; we keep patterns
Interview transcripts, process maps and org charts stay in storage bound to your engagement. They are not pooled, and they are not training material.
What we carry forward is the shape of a problem, not the instance of it: the kinds of friction that recur in an industry, ranges rather than figures, patterns rather than particulars. Anything moving from an engagement into that shared body of knowledge is reviewed by a person first, and the default is that it does not move.
We know what our own work costs to produce
Our delivery runs on the same kind of agentic systems we build for clients, and we measure what each piece of it costs. That is the practical meaning of treating tokens as the unit of intelligence: cost per deliverable is a number we can show you.
It changes the commercial conversation. Pricing anchors to the value of the outcome with a floor at measured cost, rather than to how many people we can keep busy.
The cheapest way to find out if this fits
The readiness assessment is free, takes about five minutes of your time, and commits you to nothing.