Most organisations cannot say who owns AI. A 300-person organisation needs five named seats (Sponsor, Steward, Use Owners, Champions, Assurer), filled starting with the Steward, to replace the default of nobody.
Ask the leadership team of almost any mission-driven organisation who owns AI, and you will get three answers in ten seconds. The Executive Director thinks it sits with operations. Operations thinks it sits with IT, or with the data protection lead. IT points at the departments, because they are the ones actually using it.
Everyone is partly right, and that is the problem. When ownership is shared by default, it is held by nobody in particular, and the decisions that matter (which tools are approved, what data may enter them, who checks the output) get made by whoever happens to be typing. An AI operating model is simply the answer to the question "who decides what?", written down before an incident asks it for you.
AI arrived in most organisations sideways. It was not procured, it was discovered: a staff member tried a free assistant, saved an afternoon, and told a colleague. The 2026 survey of 346 nonprofits cited on our website found that 92% now use AI, but 81% do so individually, with no shared workflows, and 47% have no governance policy at all.
That pattern is what a missing operating model looks like from the outside. Usage is high, coordination is low, and every person is running a private pilot with the organisation's data. Large firms respond by creating a Chief AI Officer and a steering committee. An organisation of 300 people cannot afford either, and does not need to. It needs clear seats, not a new department.
Our view is that a proportionate operating model for an organisation of 100 to 500 staff has five seats. One person can hold more than one seat, but every seat must have a name against it.
The model borrows from how organisations already handle finance. There is a person who sets the budget, a controller who keeps the rules, budget holders who answer for their spend, and an audit function that checks. AI needs the same separation, at a lighter weight.
It is not a committee. Committees diffuse the very accountability the model exists to concentrate. A standing forum can be useful, but a forum with no named seat-holders is the old problem in a new room.
It is not a technology decision either. Nothing in the Five Seats depends on which assistant you use. That matters because tools change every quarter, whereas the answer to "who is accountable for this department's use of AI?" should survive several tool changes.
And it is not a compliance artefact. A slide showing five boxes does not constitute an operating model. The test is whether a new staff member, on their first Monday, can find out in under a minute who to ask about a specific document and a specific tool.
An operating model is not what the organisation says about AI. It is what happens when a member of staff is unsure at four in the afternoon and has to decide alone.
Sequencing matters more than completeness. We recommend filling the seats in a deliberate order, and the second step is the one most easily skipped.
Organisations often want to begin with the Sponsor, on the reasonable ground that leadership must lead. We would begin with the Steward, because leadership cannot decide well without an accurate picture of current use.
AI ownership does not need a new department, but it does need five named seats: Sponsor, Steward, Use Owners, Champions and Assurer, filled starting with the Steward. Without them, every member of staff is making the organisation's AI decisions privately, and the organisation learns about them from an incident rather than an inventory.
Prepared by M.A.I. Consulting.
If this is the question on your desk, a thirty-minute call tells you whether the service fits, or that you do not need us yet.