Who should own AI in a company, and do you need a chief AI officer?

One person, named in writing, with a budget line they control and a standing slot in an existing meeting. That is the answer. The title on their business card matters far less than whether anyone in the building can say the name without checking.

Most businesses between 50 and 10,000 people do not need a Chief AI Officer. They need to write down who gets called when an AI tool produces a bad output that reaches a customer, and who signs off on the next AI spend. If you cannot answer both of those with the same name, you do not have an ownership problem you can solve with a job advert. You have a decision nobody has made yet.

Why the title is the least important part

Grant Thornton’s 2026 AI Impact Survey found that 78% of business executives lack strong confidence they could pass an independent AI governance audit within 90 days. Not that they would fail. That they do not know. Those are different states, and the second one is worse, because it means nobody has looked.

IBM’s 2026 survey puts a similar figure on the gap from the other direction: 77% of executives acknowledge AI adoption is outpacing their current governance capabilities. The tools went in faster than the rules. That happens in every technology cycle and it is not a scandal on its own. What makes it dangerous is that in most mid-market firms, there is nobody whose job it is to notice.

A Chief AI Officer does not fix that by existing. Plenty of large organisations have one and still cannot answer basic questions about which tools hold customer data. The title creates a focal point for board conversations, which is useful at scale. Below roughly 1,000 people it mostly creates a post with no authority, because the budget still sits with the CFO, the systems still sit with IT, and the person with the new title spends their first year asking colleagues for information they have no right to demand.

Ask a different question. Not “what should we call them”. Ask: who can stop a project?

A committee is not an owner

The most common thing a mid-market business does with AI ownership is form a working group. Someone from IT, someone from operations, someone from finance, a representative from the commercial side, monthly meeting, shared document.

This feels responsible. It produces almost nothing, for a structural reason rather than a people reason. Committees are good at surfacing concerns and bad at absorbing consequence. When six people share accountability, the honest reading is that none of them has it. Ask any member of an AI steering group what happens to them personally if a tool leaks client data next quarter and watch the answer dissolve into process.

There is a corroborating number here. 75% of executives admit their company’s AI strategy is more for show than actual internal guidance. A document written to satisfy a board rather than direct a business is exactly what a committee produces, because committee output is consensus text, and consensus text contains no decisions anyone can be held to.

Committees have a real use. They are good oversight bodies. They are terrible owners. The distinction is that an oversight body reviews what the owner did, and an owner is a single person who can be wrong.

The three models that actually work

For a business of 50 to 10,000 people, there are three realistic places to put this. Each works. Each breaks in a specific, predictable way, and knowing the failure mode in advance is most of the value.

Model one: the operations owner

The COO, operations director, or whoever genuinely owns how work gets done. They hold the process map in their head. They know which team still rekeys invoices. They can tell you where the bottleneck is without opening a dashboard.

This is the strongest default for most mid-market firms, because AI value in this size of business comes almost entirely from process, not from models. The operations owner is already measured on throughput and cost, so an AI project that does nothing shows up in their numbers.

Failure mode: technical naivety. The operations owner approves a tool that cannot integrate with anything, or signs a contract without asking where the data goes, or green-lights a pilot on a process so poorly documented that the AI has nothing stable to attach to. If three people in your business complete the same task three different ways, an AI tool will simply learn all three, inconsistently. That is one of the process questions the audit asks directly, and it is where this model trips.

The counterweight: a named technical reviewer with a veto on integration and data handling. Not a committee. One person who reads the contract.

Model two: the technology owner

The CIO, CTO, or IT director. Obvious choice, and in businesses where AI touches regulated data or core systems, often the right one.

They can actually evaluate what a vendor is claiming. They know what is already in the stack, what the licence renewals look like, and whether the integration the salesperson described exists or is on a roadmap. They will ask where the data is processed before anyone else thinks to.

Failure mode: the pilot that nobody wanted. Technology owners are judged on delivery, and AI delivery is easy to demonstrate and hard to value. You end up with a well-architected, properly governed, securely hosted tool that solves a problem no operational team had. RAND’s research found 80.3% of enterprise AI projects fail to deliver promised business value, with 28.4% reaching production but failing to deliver expected value. That middle category is this failure mode. The thing works. It just does not matter.

The counterweight: every project gets a named operational sponsor who has agreed in advance what metric moves. Not a benefit statement. A number that was measured before.

Model three: the finance owner

The CFO or finance director. Less common, increasingly sensible, particularly where the board has started asking hard questions about spend.

Finance owners impose the discipline the category has been missing. BCG found only 45% of finance executives could quantify the return on their AI investments, with median reported ROI at 10% against a 20% target. A CFO who owns AI will not approve a second tool until the first one has shown something, which sounds slow and is the fastest route to a defensible position. If your board is already pushing, the honest conversation about what to tell them on AI ROI is one this owner is already equipped to have.

Failure mode: everything gets measured, nothing gets tried. Finance ownership tends toward a standard of proof that no early-stage AI project can meet, because the benefits in year one are usually time saved rather than cost removed, and time saved is real but awkward to book. Add a shadow AI problem on top, where staff use personal accounts because the official process takes six weeks, and you have the worst version: strict governance on the sanctioned tools and no visibility at all on the rest.

The counterweight: a ring-fenced experiment budget with a lower proof bar and a hard stop date. Small, fast, written off by default.

What the named owner actually has to hold

Whichever model you pick, the role is the same four things. If the person you have in mind does not have all four, you have given them a title, not a job.

A budget line they control. Not a request route into someone else’s budget. An actual number they can spend and will be asked about. Without this, the owner is a coordinator, and coordinators get ignored by anyone senior to them.

Authority to stop something. The power to kill a pilot, block a purchase, or pull a tool out of production without needing to assemble a quorum. This is the one most firms skip, and it is the one that determines whether the role is real.

A review cadence in an existing meeting. Monthly, fifteen minutes, in the operations or exec meeting you already hold. Never a separate AI meeting, because separate AI meetings get cancelled when something urgent happens, and then quietly stop. The owner reports what is live, what it cost, what moved, and what they stopped.

A register of what is running. Every AI tool in use, what data it touches, who sponsored it, what it was meant to do. One spreadsheet is fine. Most businesses discover during this exercise that the list is longer than anyone thought, which is itself the finding.

When you genuinely do need the title

There are cases. If you are above roughly 5,000 people, operating across jurisdictions with divergent AI regulation, or in a sector where a regulator will eventually want a named accountable executive, a formal Chief AI Officer or equivalent is reasonable. The title then does real work, because it signals externally and gives the person standing in rooms they would otherwise have to negotiate into.

Below that, be honest about what the title is for. If the appeal is that it would look good in the annual report, you are buying the thing that 75% of executives already admit they have: a strategy for show. The difference between AI governance and compliance matters here, because a CAIO hired for compliance optics ends up writing policy nobody reads, while the operational decisions carry on being made by whoever bought the tool.

And be honest about the board. Deloitte found 66% of boards have limited-to-no knowledge of AI, while 61% of CEOs say their boards are rushing AI transformation. A board pushing hard on something it does not understand will accept a job title as evidence of progress. That is not a reason to create one.

Start by answering the question in writing

The audit we run asks who is responsible for the quality and accuracy of your data, and who is responsible for making sure core processes are followed and improved. Those sit in two different pillars and they are the questions people stall on most. Not because the answer is embarrassing. Because the answer requires naming someone, and naming someone makes it real.

If you want the whole picture rather than two questions, the free AI readiness audit is 30 questions across six pillars, takes 7 minutes, and gives you a score out of 120 with a band for each pillar including governance. It will not tell you who should own AI in your company. It will tell you, pillar by pillar, how much there currently is to own.

Then write the name down. One name, one budget line, one standing slot. You can argue about the title later.