// Resources
Who Should Own AI in a Company Without a CTO?
Why the job belongs to whoever already owns the process and the budget, what the role actually decides, how much of their week it takes, whether a Head of AI is worth hiring, and the four decisions no supplier should be making for you.
By Victor Xu, Founder · LinkedIn
Last updated · 2026-10-27
The short version
One named person owns AI, and in a business without a CTO it should be whoever already owns the process and controls the budget. That is usually the owner, the general manager or the head of operations. Not IT, because the work is mostly process change rather than technology, and not whoever is most interested, because interest does not come with the authority to change how a week runs. The evidence points the same way from two directions: BCG's 10-20-70 approach puts 70% of the effort into people and processes, and McKinsey's survey of 1,491 participants, published in March 2025, found that of 25 attributes tested, redesigning workflows had the biggest effect on whether a company saw EBIT impact from generative AI. Redesigning a workflow is not a technical act. It is a management decision about who does what, which is why the owner has to be someone who can make that decision on a Tuesday without asking anyone.
Who should own AI in a company without a CTO?
The person who should own AI is the one who already owns the process and holds the budget, which in most businesses under about 200 people is the founder, the general manager or the operations lead. Give them the title in writing, even if the title is just "accountable for AI", and give them two things with it: the authority to change how work is done, and the money to pay for it.
One owner, not a committee. Shared responsibility is the normal arrangement here and the one that stalls, because every decision in the list below has a natural objector and a committee hands each of them a veto. Name one person, and give everybody else a route to argue with them rather than a seat.
The owner's job is not to be the expert. It is to decide. Which problems get worked on, which tools people may use, what data a system may see, when a workflow changes, and when something gets switched off. A good owner here is someone who knows where the business loses time and can call a meeting that changes it.
Two practical notes. Pick someone who will still be in the business in two years, because the second year is when the decisions get harder and the institutional memory matters. And do not give it to the person who is already the bottleneck in every other process, which is a common instinct and guarantees the work sits.
Our strategy engagement is built around that person: we interview the people who do the work, map how a week actually runs, and hand the owner a prioritised roadmap they can act on, then stay on as their advisor while it happens. What an AI roadmap is describes the artifact itself.
Why not put IT or the most enthusiastic person in charge?
Neither the IT function nor the most enthusiastic person in the building can change how work gets done, which is why putting either one in charge fails. The job is roughly nine parts management to one part technology, and handing it to the technical function or the volunteer puts the authority in the wrong place.
Take the proportions seriously. BCG, in Scaling AI Requires New Processes, Not Just New Tools of 20 January 2026, sets out the 10-20-70 split it draws from many AI transformations across sectors: devote 10% of the effort to algorithms, 20% to technology and data, and the remaining 70% to people and processes, which is what makes the changes stick. An IT lead, or the managed service provider that fills that role in most smaller firms, is equipped for the 30% and has no mandate for the 70%. They can connect a system. They cannot tell the service manager to stop printing the schedule.
The enthusiast fails differently. Enthusiasm produces pilots, and pilots are exactly what the evidence says gets cancelled. Gartner predicted in a press release of 25 June 2025 that more than 40% of agentic AI projects would be cancelled by the end of 2027, and the causes it named were escalating costs, unclear business value and inadequate risk controls. All three are the absence of an owner with budget authority asking what this is for.
There is a version that does work. The enthusiast becomes the builder and the operator becomes the owner. The enthusiast finds what is possible, the owner decides what is worth doing and clears the path, and each of them is doing the thing they are actually equipped for.
What does the AI owner actually do?
The AI owner makes seven recurring decisions, and the role is better described by that list than by any job title. Most of them take minutes once the first pass is done.
| Decision | What it means | How often |
|---|---|---|
| What gets worked on next | Picks the next problem off the roadmap, by what it is worth and what it would take | Monthly |
| Which tools are approved | Keeps a short written list of what people may use and for what | Monthly, and on request |
| What data a system may see | Decides which records and systems each build is allowed to touch | Per project |
| When a workflow changes | Tells people the new way is the way, and backs it when it is awkward | Per project |
| Who runs it day to day | Names the person who watches the output and fixes the edge cases | Per project |
| What gets measured | Sets the before figure, reads the after figure, and publishes both | Monthly |
| When something stops | Kills a build that is not paying, and removes the tool | Quarterly |
The last one is the test of whether the role is real. A business that has never switched anything off does not have an owner, it has a sponsor. Killing a system that somebody liked is the decision that cannot be delegated, and the willingness to make it is what stops the subscription pile growing.
The one deliverable we would insist on is a single page listing the approved tools, the live systems, the measure for each, and the person who watches it. Keep it where everyone can see it. That page is the governance, at this size.
Do we need a Chief AI Officer or a Head of AI?
A Chief AI Officer is the wrong hire below a few hundred people, and almost everything written in favour of one is written for organisations many times that size. The decisions that need making do not amount to a full job, and splitting them off from operations is what creates the diffusion the role is meant to fix.
What a smaller business needs is the ownership named and the technical judgement bought in. That pairing shows up repeatedly in the practitioner advice on this question: one accountable business owner inside the firm, plus senior technical judgement on demand, whether that is a fractional technology lead, an outside engineering firm or a firm like ours. The ownership stays inside. The expertise is rented.
Three signs you have outgrown that arrangement. A full-time person is justified when AI systems touch enough of the operation that someone is managing them most days, when regulated data is involved and somebody has to answer to a regulator, or when you are building something your customers use rather than something your staff use. None of those is reached by a thirty person firm running an answering agent and an invoice pipeline.
The in-house versus outside question for the building rather than the ownership is a different trade, and automation consultant versus in-house runs the numbers on it.
How much of the owner's time does this take?
The owner's time comes to two to four hours a week for the first quarter, then one to two hours a week after that. The first quarter is heavier because the decisions have not been made yet, not because the volume of work is high.
Where that time goes: an hour a fortnight on the roadmap and what comes next, half an hour a week reading the measures, and the rest in the conversations that change how the work is done. The last category is the one people underestimate, and it is the only one that cannot be moved. A workflow change announced by email does not land, and McKinsey's March 2025 survey of 1,491 participants found only 21% of respondents reporting generative AI use said their organisations had fundamentally redesigned any workflows, which is a measure of how rarely anyone does the conversation properly. That sample skews to large organisations, and the finding is about the difficulty of the act rather than the size of the company.
What it is not is a second job. If the owner is spending a day a week, something has been loaded onto them that belongs to a builder or an operator: writing prompts, testing output, chasing a vendor. Those are real tasks and none of them is the owner's.
One honest caution about the first month. Expect it to feel like inventory rather than progress, because most of it is finding out what is already running, what it costs and who touched it last. Finding out which AI tools your team is already using is the fastest version of that, and it usually takes an afternoon rather than a project.
What decisions should we never hand to a vendor?
Four decisions stay with you whichever supplier you use, and that includes us. What problem is worth solving, what data leaves your systems, what the system is allowed to do without a person, and when something gets switched off.
The reason is not distrust, it is incentives. A vendor's roadmap is built from what they can build. Your roadmap has to be built from what the business needs next, which routinely includes doing nothing, buying something off the shelf, or fixing a process before any software touches it. When we run a strategy engagement the roadmap says all three where they apply, and the recommendation includes tools you already pay for.
What a good vendor should own is delivery and the running of it. In our case the engineering team builds what the roadmap calls for, the system lives on our infrastructure, and we keep it running and adjust it as the business moves. The Universidad Maimónides specimen database has run that way since launch, which is the arrangement we would describe as working: they own the decisions, we own the outcome of the build. The case study sets out what was built and what it returned, which was about 15 hours a week, roughly $39,000 a year of skilled time at $50 an hour.
The question to ask any vendor, ours included, is what happens to the four decisions above during the engagement. If the answer is that the vendor makes them, you have bought a dependency rather than a capability. Questions to ask an AI automation company has the rest of that list.
Frequently asked questions
How much does it cost to have someone own AI internally?
The cost is the owner's time rather than a line item, which is why it is usually underpriced rather than overpriced. Two to four hours a week of a general manager's attention in the first quarter is the realistic figure, and the thing it replaces is the slower cost of nobody deciding: tools bought and unused, pilots that stall, and the same conversation about the same bottleneck every quarter. If the role needs rented expertise alongside it, what moves that cost is how many systems are running and how much of the work is new rather than configured.
How do we set this up in the first 30 days?
Name the person in writing, then have them do three things. Produce the inventory of what AI is already in use and what it costs. Write the one page list of approved tools, live systems, measures and the person watching each. And pick one problem with a number attached and decide what happens to it this quarter. Thirty days of that beats a strategy document, and it also tells you quickly whether the person you named has the authority they need.
Is an internal owner better than hiring an agency or a fractional CTO?
They answer different questions, so it is not a choice. The internal owner holds the decisions and cannot be rented. An agency or a fractional technology lead supplies judgement, build capacity and the ability to say what is possible, which an operations lead does not have and should not pretend to. The arrangement that fails is the one where the outside firm is also the owner, because then the roadmap follows what the firm sells. The arrangement that works is one accountable name inside, expertise bought in around it.
Can our operations manager own AI if they are not technical?
Yes, and not being technical is closer to a qualification than a problem. The decisions in the table above are about process, money and priority. What the role does need is authority to change how work is done, enough standing that a workflow change sticks, and the time to do it, which usually means taking something else off their desk. What it does not need is any ability to read code or evaluate a model.
What happens if the AI owner leaves?
The one page document is what survives them, which is the main argument for keeping it. Approved tools, live systems, the measure for each and who watches it, in a place the next person can find. Without it the knowledge leaves with the person and the next owner starts with an inventory exercise. Name a deputy early too, usually whoever runs the systems day to day, because the second worst outcome after no owner is a three month gap with nobody deciding.
Give the decisions somewhere to sit
We map how your business actually runs, put a number on the opportunities, and hand your owner a prioritised roadmap they can act on. Then our engineering team builds what the roadmap calls for and we run it for you from there. We own the outcome, not the deliverable.