Services
A. Governance assurance covers the work your own teams produce with AI.
B. Release assurance covers the AI you put in front of customers.
Both are measured against the same scorecard. To preserve our independence, we assess the controls and outcomes but do not build or implement the fixes ourselves.
A / Governance assurance
Who it's for
Organisations where teams use AI assistants and agents in work that carries financial, regulatory, legal or operational consequence.
Why it pays for itself
What limits AI-assisted work is the capacity to check what was generated. Organisations respond by reviewing everything, which makes senior people the bottleneck and gets sloppier as volume rises. Knowing which checks actually work lets you automate the ones a machine can run, and give your reviewers back the time to do the rest properly.
What we assess
Decision rights and boundaries, specification before generation, codified procedure and agent tool authorisation, automated verification, human oversight capacity, data boundaries, and evidence and traceability.
01 / One week, one team
£7,500 fixed
The scorecard applied to a single area of the business, to establish where you actually sit.
You get a scored profile across all seven domains, the evidence gaps behind each score, and a written remediation specification you can hand to whoever implements it.
Most organisations use this to find out whether the problem is as large as they suspect before committing further. The report includes a scoped price for a full assessment.
02 / Two to three weeks
Scoped and quoted from the baseline
Across the organisation, evidence-based rather than interview-based: we inspect controls, test whether they block what they claim to block, and sample actual work against the specifications that authorised it.
A scored profile per team or workflow class, with evidence for every score
The required minimum level per domain, based on what a defective output would cost you
A prioritised remediation specification, scoped and costed
A report suitable for a board, risk committee or regulator
03 / Ongoing
Programme fee or monthly
For organisations implementing a governance programme and wanting it done to a standard they can defend.
We define the standard, your team or an implementation partner builds to it, and we assess their work against the scorecard as it is delivered. You get independent verification of your own programme, and a written opinion at each stage.
We do not implement any part of what we assess.
B / Release assurance
Who it's for
Organisations with a customer-facing AI feature, product or agent scheduled for release.
Why it pays for itself
AI launches get delayed by uncertainty far more often than by defects. Nobody senior wants to approve something they cannot evaluate, so dates slip while people ask for more testing without being able to say what would satisfy them. An independent assessment settles it: either the release clears, or you get a specific list of what has to change.
What we assess
Output integrity and failure modes, data boundaries and privacy, verification coverage, operational resilience, economic control, auditability and defensibility, and the delivery health of the underlying build.
01 / Five days
£7,500 fixed
A focused assessment of the three domains that account for most release incidents: output failure modes, data handling, and verification coverage.
You get findings ranked by severity and a list of what to resolve before release. If your release is a few weeks away, this is the assessment that fits.
02 / Two to three weeks
Scoped and quoted after the review
All seven domains, including the quality and maintainability of the underlying implementation.
A verdict: proceed, proceed with conditions, or do not release yet
A scored profile against the minimum level appropriate to the consequence of failure
Every finding ranked, with what would resolve it and how it would be verified
Where the verdict is negative, exactly what would change it
A single domain below the required minimum is enough for a negative verdict. Scores are never averaged, because strength in one area leaves the gap in another exactly where it was.
03 / Ongoing
Monthly retainer, scoped to the product
AI products change without your code changing. Models are updated, prompts are revised, behaviour moves.
We re-assess on change, provide release sign-off against the scorecard, and issue a written opinion each quarter that you can put in front of a board or regulator.
We define the standard and test against it. We do not implement the fixes we recommend.
Findings come with a written remediation specification: what needs to change, to what standard, and how it will be verified. Your own team can implement it, or we can introduce implementation partners we work with regularly. We take no fee from that work.
