Roles

Who should own AI in your company?

Daniel Kerambrum Bicourt · Founder, CO-ORDI · September 2026 · 5 min read

Ask five people in a 30 person business who owns AI, and you will get five different answers: the founder, IT, marketing, whoever bought the ChatGPT licence, or a shrug. MIT's 2025 study of enterprise AI found that 95 percent of generative AI pilots delivered no measurable financial return. The tools were rarely the failure. The missing single owner was.

The default answer is nobody, and nobody is expensive

Without a named owner, AI still gets adopted. It just gets adopted badly. Marketing picks a writing tool. Sales picks a different one for outreach. Operations tries a third for reporting. Each purchase makes sense on its own and none of them are told to talk to each other, because nobody has the mandate to insist on it. Six months later the business is paying for a stack of clever, isolated tools that never compound into anything, and every renewal invoice arrives with no one able to say what it actually earned.

The four people who get asked, and why each stalls

Most businesses try one of four owners before admitting the gap. Each has a real strength and a fatal limit.

• The CEO. Right person to set the ambition, wrong person to run it week to week. Founders who try to own AI personally end up as the bottleneck: nothing ships without them, and they have a business to run.
• IT or the CTO. Good instincts on security and infrastructure, often thin on commercial judgement. AI ownership means prioritising which business process to fix first, not just which platform to buy, and that is a different skill to keeping the network up.
• A committee. Built to spread accountability, which is another way of saying built to avoid a decision. Committees meet monthly. AI adoption needs someone making calls between the meetings, not during them.
• An outside agency. Ships what it is briefed to build, then leaves. An agency can execute a defined scope well, but it was never going to stay and own what happens to the system after delivery day, because that was never the job it was hired for.

What ownership actually requires

Whoever holds this, the job has the same four requirements, and most candidates above only clear one or two of them.

• Authority across departments. The right to say sales and operations should use the same underlying system, not four different subscriptions.
• Enough technical fluency to say no. Most AI pitches are worse than what a business already has. Someone needs to be able to tell the difference.
• Time inside the business, every week. Ownership exercised once a quarter is not ownership. It is a review.
• Accountability for a written outcome. A number the business checks monthly, not a strategy deck that ages in a drive.

The real answer: a named person, not a function

Ownership is a person, not a department. That is what the AI Architect role exists to be: one accountable owner who sits across sales, operations and delivery, with the authority to prioritise, the technical judgement to reject bad pitches, and enough time inside the business to actually build. For a 5 to 100 person company this usually arrives embedded or fractional, from around £3,500 to £10,000 a month, which is less than one senior hire and far less than the cost of six months of uncoordinated tool spend. See how that work is scoped and priced before you commit to anything.

If the right owner is already in the building

Sometimes it is not a hiring problem. An operations lead or a technical co-founder may already have the judgement and the standing, and simply needs the mandate, the budget and a written scoreboard to act on it. If that person exists and has spare capacity, give them the job before you look outside. The failure mode to avoid is not "we hired the wrong person". It is "we never named anyone at all".

Whichever route you take, the test is the same in ninety days: one accountable name, one written plan, and a number the business checks every month. Everything else, the tools, the training, the agency projects, is downstream of that decision.