How ready is your AI, really? Take the 12-minute assessment →

AI transformation inside regulated businesses.

The advice, the systems, and the product leadership to keep them running. Over twenty-five years where “it works in the demo” was never the end of the conversation.

Barclays Global Investors · Absa · Skrill · Simply Business · TransferGo
Start the assessment Or talk to me
The receipts

Every number here tells you what it isn’t.

These are the strongest figures I have. Each one carries the limit of my own claim to it, because you were going to ask anyway — and because a number without its attribution is just a number.

142M+
Automated FAQ interactions served by Absa Chat Banking.
What I did not do I commissioned and negotiated the NLP architecture and set the usage-based commercial structure. I did not build the model, and I had left before this was measured.
proof/absa-142m-faq-deflection · confidence: medium-high
Feb 2022
A Camvas deck roadmapping intelligent summarisation, semantic search and automated transcription — before the ecosystem commoditised all three.
How to read it Don’t take my word for the foresight. The deck is dated and I’ll send it. Judge the dates.
proof/camvas-pre-llm-ai-foresight · confidence: high
224%
YoY growth reported on Clickatell’s AI-enabled Chat Commerce platform.
Not my result I piloted the AI capability the growth is attributed to. I did not deliver the growth and I wasn’t there when it was measured. Downstream evidence, not a claim.
proof/clickatell-224pct-yoy-growth · public press release
C+
The architecture grade I gave my own AI coding assistant, in writing, in its own repository.
Why it’s here Because a portfolio of nine builds with nine glowing self-assessments would tell you nothing. This one tells you how I audit.
projects/superdev · self-audit, published

This is the whole method, in miniature. My evidence base is a single source of truth, and every claim in it carries an attribution field written before anyone asked. The assistant below reads those fields and leads with them. It is not being modest — it is reading the data.

Engagements

One subject. Three ways in.

Ordered by depth of commitment, lightest first. Every engagement has a fixed scope and a fixed duration before it has a price.

01
Consult / Hardening AI Sprint
You have shipped AI and it hasn’t landed. Two to six weeks, six named artefacts, no build commitment.
2–6 WEEKS
FIXED SCOPE
↗
02
Collaborate / Venture Build
Custom agentic builds, multi-tenant platforms, zero-to-one. Founder-operator, shared risk.
3–12 MONTHS
PARTNERSHIP
↗
03
Hire / Executive Role
Fractional, interim, or permanent. CPO, Head of Product, Head of AI Product. The deepest way to work with me.
1 DAY/WK → FULL TIME
OPEN
↗

Most people arrive at 01 and never need the rest. That is a good outcome.

AI readiness assessment

It’s a conversation, not a form.

Twelve minutes. No radio buttons. It adapts to what you say, and it will tell you when your answer changes the diagnosis. You get the score, the read, and the plan — before anything is asked of you.

Readiness analystTurn 4 of ~14
You’ve got two production models and a governance process that was written for the data warehouse. That’s the most common shape I see, and it’s not a tooling problem.

What breaks first when a regulator asks how a decision was made — the logging, or the person who has to explain it?
Honestly? The person. We can pull the logs.
That’s the more fixable of the two, and it moves your Accountability score more than any model change would.

One thing I should say plainly: I’m scoring you against the NIST AI RMF GOVERN-1.2 — the levels are mine, not NIST’s. NIST deliberately doesn’t define maturity levels. You can check the subcategory yourself.
Score, diagnosis and your Monday morning plan are free. The roadmap and the PDF ask for an email.
How this is scored

Every question maps to a named subcategory of the NIST AI Risk Management Framework 1.0 — four functions, nineteen categories, seventy-two subcategories. It is free, it is public, and you can read the exact line I scored you against.

The 1–5 levels are mine. NIST doesn’t define maturity levels — it is explicitly “outcome-focused and non-prescriptive”, and it says its actions “do not constitute a checklist”. Anyone selling you a “Level 3 against NIST” invented that ladder too. The difference is that mine is published: 0 / 1 / 2 per subcategory, mean × 2.5.

On the governance dimension I also name the relevant ISO/IEC 42001 Annex A control areas. I don’t reproduce the standard — it’s paywalled at roughly CHF 225, which is precisely why I won’t pretend you can verify a score against it.

I am not going to tell you how you compare to your peers. No public dataset supports that percentile, I don’t have a respondent population yet, and a number like “+2.2 above average” implies a measurement nobody has made. When I have the data, I’ll publish the n.

If that answer is unsatisfying, it is the same answer you will get from me on a Tuesday afternoon about your own numbers.

Evidence

Nine systems, currently running.

Not a portfolio grid. Ask the assistant anything about any of them — what shipped, what didn’t, which technology decision I reversed and why.

Knowledge systems under regulatory load
Where the graph has to be defensible to someone who audits it.
AIssure — 50k nodes · attribution pending InfoTouch — pgvector over Neo4j, reasoning published Guided — MCP server, ~90-doc controlled wiki
Multi-tenant platforms, zero to one
Shipped to real users, with the migration count to prove it.
ILALI — 32 migrations, 47 endpoints, pilot Mar 2026 GenAI / PDVP — schema-per-tenant, WORM audit Reptile Basics — WooCommerce + PayFast
Agent infrastructure and research
Including the one I graded C+.
SuperDev — self-audited architecture JobEasy — repository / strategy / command RoboS — ROS2 · Three.js · WebSocket
Not a fit

Three times you should not engage me.

Each one comes with what to do instead, because naming a problem and walking away from it isn’t honesty — it’s just a shrug.

You need AI added to a product that isn’t working yet.
AI will make a bad product fail faster and more expensively. Instead: fix activation first. That’s a product problem, and a good PM costs less than I do.
Your leadership hasn’t agreed that this matters.
Transformation work without executive buy-in becomes an expensive internal argument. Instead: get one sponsor with budget authority to say it out loud, then call me.
You want someone permanent, in the room, five days a week.
Fractional is the wrong shape and I’ll tell you so in the first call. Instead: hire — and if the role is right, that conversation is engagement 03.