The recommendation is the product
No resale, no commission, no implementation arm downstream. If the honest answer is that you do not need us, that is the answer you get.
The practice
Ten years of this, most of it before it was fashionable.
Inference Institute exists because the same conversation kept happening: an organisation with a real problem, a supplier with a product to sell, and nobody in the room whose only job was to work out what should actually be built.
We architected and ran AI systems years before ChatGPT made them a board topic — statistical models at scale, the infrastructure underneath them, and the regulatory obligations that arrived with them. That order of experience is why we start with the data and the decision, not the demo.
Principles
Everything ends in writing
A design, an assessment, a decision record. Not a workshop, not a slide pack, and never a verbal steer that evaporates when the people change.
We publish no number we cannot evidence
No invented efficiency percentages, no client logos we did not earn permission to show, no outcome statistics with no method behind them.
Where the experience comes from
From people who have already done it — in estates where getting it wrong was expensive, under supervision, and with somebody obliged to put their name to the decision at the end. That is the room this practice came out of.
Systems built and run inside financial services, insurance, health and the public sector — under supervision, with auditors in the room and a regulator entitled to ask how a decision was reached.
PhD researchers who go to the literature at source, keep up with a field that moves monthly, and can tell you which result holds up outside the paper that reported it.
People who have taken models into production and then operated them — which is where an architecture turns out to be either true or expensive, and where opinions stop being cheap.
Nobody learns on your estate. The person who scopes the work is the person who does it, and the same people are still there when the difficult question arrives.
What we work out gets written down — the engagements end in a decision record, and the thinking behind them is published at Research. Named client work sits on Work.
Frameworks we work to
EU AI Act
ISO/IEC 42001
NIST AI RMF
UK regulatory expectations
US federal and state rules
Sector-specific requirements
We deliver readiness and alignment — formal interpretation of any regulation stays with your legal counsel.
Architecting intelligence, engineering success.
Start here