Consulting Desk
Seven services,
in the order a client
actually meets them.
Each service names the engagement it runs under, the method behind it, the gate it has to pass, and the artefact you hold at the end. All of them can end in “not this year”. What gets built once a decision is made is on the Factory Floor; how any of it is contracted is under how we work.
01 — Decide
Before anything gets built
Four services for the period when the question is still what to do, not how to do it. All four end in a written position you can take to a board, and all four can end in “not this year”.
AI Opportunity Assessment
A scored portfolio of candidate use cases across the business, each mapped to a decision or process it would actually change, and scored on business value, technical feasibility and the state of the data it would need. The bottom of the list is argued as carefully as the top — knowing what not to attempt this year is the cheaper half of the output.
- EngagementAdvisory Sprint
- MethodValue × feasibility × data readiness
- GateEvery use case maps to a named decision
You hold: the scored use-case portfolio + the three we would start with, sequenced, with the reasoning for each and the ones we set aside.
AI Readiness Assessment
Whether this organisation can support the AI it wants — data reachable in the form a model needs it, tooling, skills, and the governance to approve something and to stop it. Scored against what your candidate use cases actually require rather than against a generic maturity ladder, because a business can be ready for one and nowhere near ready for the next.
- EngagementAdvisory Sprint
- MethodScored against the use case, not a ladder
- GateEvery finding names a system and an owner
You hold: readiness findings by domain + the specific gaps blocking your top use cases + a remediation sequence with named owners.
AI Strategy & Roadmap
The board-ready version: where AI creates value in this business over the next four to six quarters, in what order, at what cost, and what has to be true before each step can start. Written to be read by a board and defended in the meeting afterwards — every number in it has a source and every sequence has a reason.
- EngagementAdvisory Sprint → Production Build
- MethodSequenced by dependency, costed per phase
- GateEvery step names its precondition
You hold: the costed roadmap with dependencies + the board paper + an investment case per phase + the assumptions it would fall over on.
AI Estate Review
An inventory of the AI already running here — the sanctioned systems, the pilots nobody closed, the tools a team bought on a card, and the models quietly embedded in software you licensed. Each one classified by what it touches, who owns it, and whether anyone could explain its output if a customer or a regulator asked. The count is usually wrong in both directions.
- EngagementAdvisory Sprint
- MethodDiscovery before policy
- GateEvery system has a named owner, or is listed as unowned
You hold: the estate inventory + a risk classification per system + a keep, fix or retire call on each + the systems nobody knew were running.
Paper or live: this review ends on paper. Sentinel is the same review delivered through Citadel — choose it when you want the answer to stay live. Citadel keeps it that way.
Also on the desk: Build, Buy or Partner Review
The decision that comes before the shortlist. What is genuinely commodity and should be bought, what is your differentiator and is worth building, and what the market sells as a product that is really a services engagement in a wrapper. We test supplier claims against what they put in writing rather than against the demo they controlled, and the recommendation stands whether or not we build it. Runs as an Advisory Sprint. Read the service
02 — Govern & operate
How the organisation runs AI once it has some
Two services that deploy the operating model we run ourselves — published in full on our delivery model, not described to you in a proposal.
AI Operating Model Design
Who commissions AI work, who reviews it, what a human must decide, where escalation goes, and how the whole thing is funded. Designed around decision rights and review gates first, because an operating model that cannot say who stops a project is an org chart with new job titles on it.
- EngagementAI Workforce Deployment
- MethodDecision rights and review gates first
- GateEvery AI decision has a named human owner
You hold: the operating model + decision rights matrix + review gates + escalation routes + the written standards your teams read.
AI Governance & Assurance
The governance layer a scrutinised organisation needs: model inventory, risk classification, review gates before deployment, and an audit trail that still answers the question when someone asks about a decision a year later. Mapped to the obligations you are actually under — EU AI Act, ISO 42001, your sector regulator — rather than to a framework we happen to prefer.
- EngagementProduction Build
- MethodMapped to your obligations
- GateAn audit trail that answers a question a year later
You hold: model inventory + risk classification + pre-deployment review gates + audit trail design + the evidence pack behind it.
03 — Lead
When you need someone in the accountability line
One service, and the one most clients actually want first. A seat with decision rights, held by a founder, not a block of advisory hours.
Fractional AI Leadership
One of the two founders embedded one to two days a week to own AI direction inside your organisation — commissioning work, reviewing it, killing what is not working, and being the person accountable when the board asks. A seat with decision rights, not advisory hours, held by someone running an AI company the rest of the week.
- EngagementFractional AI Leadership
- MethodA seat with decision rights
- GateDecision rights agreed in writing before day one
You hold: a standing decision log + a quarterly written position on the AI portfolio + direct founder access between them.
Where the rest of the work lives
This desk decides. The other two build and research.
Once a decision is made, delivery moves off this page. Keeping the two separate is deliberate — advice priced as the front end of a build we have already assumed is not advice.
Build it
The Factory Floor
Four classes of system: production LLM systems, governed multi-agent workforces, decision-support tooling, and model evaluation and red-teaming infrastructure.
See what we build
Deliver it
Proof of Concept, Team Enablement, Venture Build
Three shapes that sit under how we work rather than on this desk — testing a use case, leaving the discipline with your team, or building a venture with equity on the table.
See the seven shapes
Research it
The AI Lab
Where the question has no answer yet — red-teaming under Article 9, agentic governance, ISO 42001 pathways, six dossiers documented as they run.
Explore the Lab
Not sure where you are on this list?
Tell us the decision, not the deliverable.
Most clients arrive thinking they need a build and start with an Opportunity Assessment. Tell us the decision you are trying to make and we will tell you which service answers it — including when the answer is that you do not need us yet.
Bring us the question you can’t get answered.
Thirty minutes with a founder. Tell us the decision in front of you and we will tell you which of the seven services answers it — and when none does yet, we will say that too. No sales deck, NDA available beforehand.
Price agreed before we start · Founder-led · NDA available