A large consulting firm
A deep bench and a solid framework, delivered by a team that moves on before the framework meets a real release. The partner you met is rarely the person on your file.
AI assurance & architecture · Regulated organizations
The independent read on AI you will have to defend.
Manseur reviews, designs and governs the AI systems that answer to regulators, auditors and boards. One senior practitioner, independent of the build, in English and French.
The problem
Your policy and your systems are written bypeople who do not read each other.
The AI policy is drafted by risk and legal. The retrieval platform is built by engineering. Both documents are competent on their own terms. Neither answers the question an auditor actually asks: on what evidence was this system approved for production, and who has the authority to stop it?
That gap is not a documentation problem. It is an architecture problem: decision rights, data contracts, evidence gates. That is why it survives every new policy draft, and why most AI programs stall somewhere after the pilot.
We work in that gap. The same person reads the directive and the reference architecture.
How organizations solve this today.
A law firm
An accurate reading of the obligation. It stops precisely where system design begins.
Your own team
The people who know your systems best also own delivery. That is precisely why their review of it is not independent.
A governance platform
A register and a workflow. Useful once you know what to record. It does not tell you whether to proceed.
Each is a reasonable choice. None of them puts one accountable person across both the obligation and the architecture.
Whoever you hire, require these five things.
This list is useful even if you never call us. Bring it to whichever firm you are evaluating.
- 01
One person who reads both
The regulation and the reference architecture. Ask them to explain your retrieval permissions and your privacy-assessment obligation in the same sentence.
- 02
The person who diagnoses is the person in the room
Ask who delivers, by name, and what else is on their plate this quarter.
- 03
Independence from the build
If they also implement, their recommendation has a commercial interest in the answer.
- 04
A decision, not a maturity score
“Level 2 of 5” is not something your board can act on. Ask what the deliverable concludes.
- 05
French, if you are a Quebec public body
Written in French, not translated from an English deliverable. Ask to see one.
Three modes, one practitioner.
01
See the services in detail →Review
A focused independent review of one consequential initiative or architecture. Get a defendable decision, prioritized findings, and clear conditions before you commit.
02Design
Shape the target architecture, data contracts, controls, and sequencing your teams can build against. Leave with decisions translated into implementable patterns.
03Govern
Turn policy into an operating model across intake, procurement, release, and monitoring. Establish ownership, evidence gates, and a repeatable steering rhythm.
Proof
Seventeen years inside systems that answer to more than their users.
Payment platforms answerable to card schemes and auditors. A national payment switch replacement. Public institutions answerable to access-to-information law, privacy regulators and their boards.
AI platforms now answer to all of the above at once. That convergence is the whole practice.
17years in regulated industries
1practitioner, from diagnosis to delivery
2working languages, EN and FR
0implementation contracts, by design
The thinking, published
All insights →The Governed Stack
Monthly analysis for leaders and practitioners working where AI ambition meets architecture, risk and accountability.