Plan
AI operating model
How AI actually gets decided, funded, built and owned in your organisation: decision rights, a use-case intake that runs without us, the roles and the funding model, and named accountability for every system once it is live.
Duration
4–6 weeks
Built on
ISO/IEC 42001 · ISO/IEC 38500 · ISO/IEC 27001:2022
Indicative price
€8,500–19,000 per engagement
Who this is for
COO
Owns the way the company works and is being asked to absorb a new kind of project.
CIO or CTO
Has requests arriving from every function and no way to arbitrate them.
Chief transformation or digital officer
Ran the pilots and now has to make this routine.
CFO
Needs a funding model that is not a series of exceptions.
Pilots are managed personally. Production has to be managed structurally.
The first two AI projects succeed because a small number of determined people carried them. That is not repeatable and it does not scale. The symptom is familiar: the third and fourth stall, every new request lands on the same two people, and nobody can say who decides whether a use case proceeds.
An operating model answers four questions in a way that survives the departure of any individual. Who decides — what gets built, on what evidence, and who is allowed to say no. How it is funded — because an annual capital cycle fits badly with a portfolio in which most candidates should be killed in week three. Who builds — the balance of internal team, partner and vendor, and the standards that apply whoever builds. And who owns it afterwards — the question that quietly determines whether a working system is still working a year later.
That last one is where most models are thin. A deployed AI system is not finished. It drifts, the data behind it changes, the process around it changes, and somebody has to notice. If the answer to 'who owns this in production' is the project team, the answer is nobody — the project team is on the next project.
The intake process matters more than the org chart. A good intake is cheap to enter, quick to reject, and produces a recorded reason, so the same idea does not come back every quarter with a new sponsor. Most of the value of an operating model is in the rejections it makes routine and defensible.
We design this to fit the organisation you have rather than a reference diagram. A company of eighty people does not need a centre of excellence; it needs three named people, a one-page intake and a monthly decision. Scaling the artefacts to the organisation is most of the craft.
How we do it
- 01
Current decision map
1 week
How AI decisions are actually made today, including the informal routes. Drawn from interviews rather than from the org chart, because the two rarely match.
- 02
Design principles
3 days
Centralised, federated or hub-and-spoke, and why. Agreed with the executive before any detail, because reversing this later costs months.
- 03
Decision rights and roles
1 week
Who decides, who is consulted and who is accountable at each stage, from intake to retirement. Mapped onto the roles you have, with gaps named as gaps rather than filled with job titles nobody intends to hire.
- 04
Intake and funding
1 week
The intake form, the qualification bar, the decision cadence, and a funding model that allows small bets and fast kills.
- 05
Build and run standards
3–5 days
What applies to anything that ships, whoever builds it: documentation, evaluation, human oversight, monitoring, and the handover that makes ownership real rather than nominal.
- 06
Pilot and handover
1 week
We run the first two or three intake decisions with you, then hand the process over and step back. A model we have to operate is a model that failed.
Named artefacts
What you receive
- Operating model design, sized to your organisation
- Decision rights matrix from intake to retirement
- Use-case intake form and qualification bar
- Decision forum: membership, cadence and terms of reference
- Funding model, including how small bets are approved and killed
- Role descriptions for the roles the model actually requires
- Build and run standards applying to internal teams, partners and vendors alike
- Production ownership and handover definition, with the monitoring expectation
- Escalation path for systems that misbehave
- Transition plan from where you are today
What we need from you
- An executive sponsor who will chair the decision forum rather than delegate it.
- The real decision history, including the decisions made informally — that is where the current operating model is visible.
- Finance in the room for the funding design. A funding model designed without finance is a proposal.
- Willingness to name owners. Every unnamed owner in the model is a gap you will meet again later.
What changes
- 01Use-case decisions are made on a cadence, on evidence, by named people.
- 02Rejections are recorded with reasons and stop recirculating.
- 03Every production system has an owner and a monitoring expectation.
- 04Funding allows a candidate to be killed cheaply, which is the point of it.
- 05The process runs without us — and without any single individual.
What it costs
€8,500–19,000 per engagement
All prices exclude VAT.
Questions
Do we need a centre of excellence?
Most organisations we work with do not. A CoE is a reasonable answer above a certain scale and a bottleneck below it. We design to the organisation you have, and if the answer is three named people and a monthly meeting, that is the answer.
How is this different from the AI value assessment?
The value assessment ranks the portfolio you have today, once. The operating model builds the machine that does it every quarter without us. Clients often buy them together, and the intake bar is calibrated from what the assessment learned.
We already have a project governance process. Can this sit inside it?
Usually yes, and it should. What has to be added is the qualification bar, the evaluation and oversight standard, and the production ownership definition. Building a parallel governance structure for AI is a common and expensive mistake.
Does this cover what ISO/IEC 42001 expects?
It builds much of what the standard expects around leadership, roles, decision-making and operational control, and we design it to be evidenced. It is not a management system on its own — that is the ISO 42001 implementation engagement — but it is most of the hard part.
What does it cost?
€8,500–19,000 per engagement. The range reflects the size of the organisation and how many functions will run AI work.
Leave with your top three risks documented
Thirty minutes with a senior practitioner. No slideware, no sales engineer.