Trust is not a feature added at the end. Our trust framework designs each automation so that a sceptical person can ask these questions and get a specific answer: decide whether to automate, design it well, demonstrate that it works, document what happened.
01 · Decide
Should this be automated, and are we allowed to?
The business case, the compliance position and what “good enough” means, agreed before anything is built.
We write down the purpose, the lawful basis, the accountable owner, the risks accepted and the standard it must meet. That opens the warrant: the evidence that the automation is lawful, tested and defensible, which every later step adds to. Sometimes the right answer is not to automate.
02 · Design
How do we automate it well?
Automate the effort. Constrain the AI. Escalate the interpretation.
Routine, rule-based steps are fully automated. Where AI judgement is used it is tightly constrained: closed sets of outputs, rules enforced in code that a prompt cannot talk past, independent runs that vote, adversarial testing, and a defined way to say “this does not fit”. Decisions about meaning go to your people, by design.
03 · Demonstrate
How do we know it is good enough?
Judged against expert golden sets, with every figure carrying a confidence interval.
Anything involving judgement is tested against work your experts produced by hand, using a proper statistical framework rather than raw percentage agreement, and against the threshold set in the warrant. A result can be pass, fail or inconclusive, and we say which. We publish what it has and has not been tested on.
04 · Document
What happened in this case?
An audit trail for everything.
Every output comes with a full record from raw data to result: each transformation, each automated judgement and each human decision. Any figure or quote can be traced back to the source behind it.
Governance runs through all four
Before work starts we agree and write down where data is processed, who can see it, how long it is kept, whether any AI service may retain or learn from it, and which decisions need a named person's approval.
Not a replacement for your peopleThey design, interpret and stand behind the work.
Not autonomous end to endEvery process has points where a person decides.
Not a black boxIf we cannot show how an output was made, we do not deliver it.
Not a platform to buyA method that travels with you. The tests and records are yours.