Who should own AI in a mid-size company?

One person should own AI in a mid-size company. Not a function, not a committee, not a workstream. One named executive with real budget authority over both business process and data, who reports directly to the CEO, whose performance review contains an AI outcome, and who can shut a project down without asking anyone’s permission. In a 200-person business that is often the COO or the Managing Director. In a 3,000-person business it is more likely the COO or a CDO with genuine P&L reach. It is almost never the CIO, and it is never a steering group.

That sounds obvious until you look at who actually holds the job today. Deloitte’s 2026 Global CSO Survey found that 65% of chief strategy officers do not own the top strategic decisions in their own function. Those are people whose job title is literally strategy. If the strategy chief does not own strategic decisions, then whoever you have quietly designated as “the AI person” almost certainly does not own AI decisions either. They convene, they coordinate, they produce a slide. Someone else decides.

Why ownership is the variable that predicts everything else

The failure numbers for AI are now well documented enough to be boring. S&P Global Market Intelligence found the share of firms scrapping most of their AI initiatives rose to 42% in 2025 from 17% a year earlier, with the average organisation scrapping 46% of its proofs of concept before production. PwC’s 2026 CEO Survey found 56% of CEOs report neither increased revenue nor decreased costs from AI in the last 12 months, with only 12% reporting both.

Everyone reads those numbers as a technology problem. They are not. A project with no owner has no one to make the unpopular calls: which of the six half-built pilots gets killed, whose team has to change how it works, which system gets rewired, who gets told their data is not good enough to build on. Those decisions are all political. A committee cannot make them, because committees exist precisely to avoid the moment where one person carries the blame.

Here is the tell that ownership is missing. Ask who is accountable for AI outcomes and you get a list of names. Ask who gets fired if nothing works and you get silence, or a shrug, or the CEO. “The CEO” is the same answer as “nobody”, because the CEO is accountable for everything, which means they are operationally accountable for nothing specific.

The three arrangements that do not work

It sits with IT

This is the default in maybe two-thirds of the mid-size businesses I walk into. It is the most defensible-looking wrong answer, because AI arrives as software, and software is IT’s job.

The problem is scope. Your CIO or IT Director can procure tools, integrate systems, and handle security. What they cannot do is walk into the finance team and say the month-end close is going to change. They cannot tell the sales director that his pipeline hygiene is the reason the forecasting model produces nonsense. They have no authority over the two things that determine whether AI pays: how work is done, and what state the data is in. So they buy the tool, deploy the tool, and then spend eighteen months explaining that adoption is a business issue.

That is not a criticism of IT. It is a criticism of giving someone accountability without giving them the levers. If you want to test this, look at where your data quality actually sits. In most mid-size businesses nobody owns it at all, which is why dirty data is the cheapest AI mistake to fix and the one most companies skip.

It sits with an AI steering committee

The committee is what you form when you know the decision matters and you do not want to make it alone. It has the CIO, the COO, someone from finance, someone from HR, and a keen middle manager who has been reading about agents.

Committees are excellent at three things: surfacing information, spreading buy-in, and diffusing responsibility. The third one kills you. The committee will approve pilots because approving a pilot costs nothing and refusing one makes you look like a Luddite. It will not kill pilots, because killing a pilot means telling a colleague across the table that their idea was not good enough. So the portfolio grows, spend creeps, and eighteen months in someone asks what any of it delivered.

Writer and Workplace Intelligence found that 75% of executives admit their company’s AI strategy is “more for show” than actual internal guidance, with 48% calling AI adoption a massive disappointment, up from 34% the year before. A strategy that is for show is what a committee produces when nobody at the table has the standing to say no.

Nobody owns it, formally

The third arrangement is the most common and the least admitted. There is no owner. There are enthusiasts. Marketing has a copy tool, finance has someone doing something clever with a spreadsheet and a chatbot, and an engineer built an internal assistant on a weekend.

This is not benign. PagerDuty’s 2026 Shadow AI Survey of 1,250 office professionals at companies with $500m or more in annual revenue found two-thirds had used unauthorised AI tools at work, while only 18% of organisations report having formal AI security policies in place. I have written elsewhere on whether shadow AI is a real risk at 500 people, and the short version is that unowned AI does not mean no AI. It means AI you cannot see, cannot audit, and cannot switch off.

Why the owner must hold process and data together

The reason IT ownership fails, and the reason a committee fails, comes down to the same structural point. AI value is produced at the intersection of two things that usually sit under different bosses.

The first is process. If three people in your business complete the same task three different ways, there is nothing stable for a model to sit on top of. You are not automating a process, you are automating one person’s version of it. The second is data. If nobody is accountable for whether your customer records are accurate, then every output built on them carries an error nobody can quantify.

In most mid-size companies, process sits with operations and data sits with IT or finance or nowhere. Splitting them means every meaningful AI decision requires two executives to agree, and executive agreement is slow and reversible. Put both under one person and the decisions collapse into one conversation.

This is also why the answer to “do we need a chief AI officer” is usually no. A new C-level title with no operating authority is a committee with a business card. If your COO already controls process and can be given data accountability, appoint the COO. If nobody controls both, that is your actual finding, and it matters more than the org chart.

Why the CEO cannot just keep it

CEOs often hold AI themselves for the first year, and there is a case for it. Nobody else has cross-functional authority.

The problem shows up in the finance relationship. Gartner data cited in coverage of the ROI gap found 65% of CEOs report misalignment with their CFO on AI’s long-term value. Think about what that means mechanically. The person authorising the spend and the person who has to sign off the return are working from different assumptions about what the return is. When the CEO owns AI personally, there is no one between those two positions to force a reconciliation. There is no operator whose job it is to arrive with a baseline number, a target, and a date.

An owner one level down changes the conversation. The CFO now has a counterparty who has to bring evidence rather than conviction. That is the whole mechanism by which AI outcomes become something you can prove to a board instead of something you assert.

The ownership test, run it in one meeting

This takes about twenty minutes with your leadership team in the room. Do not send it round in advance. The value is in the pause before people answer.

1. Name the person. Ask everyone to write down, privately, the name of the single person accountable for AI outcomes. Collect the answers. If you get more than one name, you do not have an owner. If you get “the CEO” from anyone other than the CEO, you do not have an owner.

2. Test the kill authority. Ask the named person: can you cancel any current AI project this week, without seeking approval? If the honest answer is no, they are a coordinator.

3. Test the budget. Ask whose cost centre AI spend lands in, and whether that person set the number. Accountability without budget is decoration.

4. Test the data lever. Ask: if the owner determines that a core dataset is not fit to build on, who has to agree before remediation starts? Every additional name is a month of delay.

5. Test the process lever. Ask: can the owner mandate a change to how a department does its work? If they can only recommend it, expect adoption to stall exactly where you thought it would.

6. Find the review line. Ask whether an AI outcome appears in the owner’s written objectives for this year. Not a project milestone. An outcome, with a number.

7. Ask the CFO. Directly, in the room: do you agree with the value this is supposed to deliver? Watch the pause. That pause is the misalignment figure showing up in your own business.

A business that passes all seven is rare. Most fail at three, four and six, which is a consistent pattern: you have named someone, but you have not armed them.

What to do with the answer

If the test goes badly, do not start by hiring. Start by fixing the mandate for someone already on the payroll. Write down who owns AI outcomes, what budget they control, what they can cancel unilaterally, and what number they are judged on. One page. Circulate it. The circulation is the point, because ownership that has not been announced does not exist.

Then work out what the owner is actually inheriting. Ownership of a mess is still ownership of a mess, and the patterns that kill AI projects rarely start with the model. Our free AI readiness audit takes 7 minutes, asks 30 questions across six pillars including Data, Process and Governance, and gives you a score out of 120 with a band for each pillar. Two of the questions ask, in plain terms, who in your business is responsible for data quality and who is responsible for making sure core processes are followed. Most leadership teams cannot answer either one without a discussion, and that discussion is worth more than the score.