LISTEN

Listen to this perspective

Written by Oleg Cohen. Companion narration uses the Brian voice.

Here is a sequence most leadership teams will recognize.

The board or the CEO says the company needs an AI strategy. The request goes to the CIO, the CTO, data science or an innovation team. They evaluate tools, and a vendor sells a product. Pilots launch, and most of them meet their targets. Individuals get faster. A year later, leadership asks where the enterprise value is.

Nobody in that sequence did a bad job. The problem was set at the first step.

There is a second sequence, in which nobody decides anything and AI arrives anyway. Part II, Busy With AI, covers that one. This essay is about the handoff.

A category error

The first step classified AI as a technology to implement. Once that happens, the rest follows. Technology goes to the technology organization, and the technology organization translates the request into the terms it owns. Those terms are models, copilots, agents, integrations, data platforms, security and vendors. The questions become which model, which framework, whether to build or buy, and which use cases to pursue.

Those are legitimate questions, and someone has to answer them well. They are also second-order questions. The first-order question never got asked. What could this company become capable of doing, now that intelligence is becoming abundant?

I think this is a category error. AI is a technology, but for an operating company it matters as a new source of capability. Software can now do some of the interpreting and reasoning that only people used to do. It can be placed before a decision, between two departments, or beside a frontline employee. Where to place it is a question about the business.

What it looks like in practice

I have seen this firsthand, in two versions. I should say where I was standing. In both cases I approached the company from the outside, as a solution provider, so this is a vendor’s view of what happened. I’ve kept to what I observed.

At a medical devices company, commercial leadership delegated AI adoption to the IT department. IT’s first act was a policy that the company would use AI tools only from established industry vendors, with a minimum number of years in business. For an IT organization that is a reasonable rule. It limits security exposure and vendor risk, and those are the things IT is measured on. It also shut out every newer company, mine included. We were set aside on that criterion, not on what we offered. In a field this new, a years-in-business requirement selects for vendors that predate the technology. Vendor risk is real, and there are other ways to manage it, such as a bounded pilot, data portability and clear exit terms. Commercial leadership had a question about how to sell and serve customers better. What it got back was a vendor policy.

At a large pharmaceutical company I approached, AI innovation had been assigned to the data science team. The team opened an RFP process that ran for months, and its position was that AI wouldn’t happen until the data was sorted out. That is a reasonable instinct for data scientists, who are trained to distrust messy data. But which data needs sorting depends on which capability the company is trying to build. Without that answer, sorting out the data has no finish line.

In both companies, capable people did what their function is built to do. IT reduced risk, and data science demanded rigor. Neither group was asked what the business should become capable of, and neither could have decided it.

The questions nobody owns

Business leaders own outcomes. Technology leaders own systems. Between them sits a set of questions that belongs to no one.

  • What should now become possible for this company?
  • Which capability should change, and what should it become?
  • How should people, AI and systems work together in it?
  • What should be uniquely ours, and what should we simply rent?
  • How will the value compound, and who will see whether it does?

In most companies those questions fall between the two groups, so they go unasked. That is the missing owner.

So is IT or data science the right group to lead? The two companies above suggest the question is framed around the wrong thing. Both functions are necessary, and in many mid-market companies the CIO is the only executive who understands the technology at all. But neither was designed to decide what the company sells, how it prices, or what it promises customers. The department matters less than two other things: the question the owner starts from, and the authority the owner has.

A test: five decisions

Take the services company from the earlier essays. Management wants to offer a preventive service with a guaranteed response time. AI makes it feasible. Launching it requires at least five decisions.

  1. What to charge, and how the pricing model changes.
  2. What the guarantee promises, and what it excludes.
  3. How technicians are staffed, scheduled and trained.
  4. Which parts are stocked, and how much capital that ties up.
  5. How sales, service, procurement and finance change their processes to match.

Now ask which of those five your AI leader can decide. If the answer is none, that person runs a technology program. It may be a good one. It isn’t the company’s AI strategy.

What the owner needs

I won’t propose a new title, because the first essay argued that a title can’t settle these arrangements. The owner of this work needs four things.

  • Ownership of a business outcome, with accountability for it.
  • Authority over pricing, staffing, process and customer promises, or a direct line to the people who hold it.
  • Enough technical understanding to see what has become possible, and a strong technology partner for the rest.
  • A reporting line to the CEO or the board, which reviews evidence of capability and not adoption charts.

A CIO can be this person, given that authority and that starting question. So can a COO or a business-unit president with a strong technical partner. The principle is that AI capability design is a business responsibility with technology participation. It is not a technology project with business sponsorship.

Owning it is not delivering it

There is one more requirement, and ignoring it causes most of the wrecks I see. Owning a business outcome is one thing. Designing a differentiated capability is another. Delivering the technology that makes it work is a third, and it takes real engineering expertise and a track record of putting complex systems into production. AI has made prototypes cheap. It has not made delivery cheap.

MIT’s 2025 report, The GenAI Divide, is best known for its claim that 95 percent of organizations saw no measurable P&L impact from generative AI. That number has been widely repeated and widely criticized, and the first essay in this series argued the picture is more mixed. I put more weight on a quieter finding in the same report. Initiatives built with external partners succeeded about twice as often as internal builds. I’m a vendor, so that finding suits me. The reason behind it holds either way. A team that has delivered this kind of system before has already made the integration and adoption mistakes that a first-time internal team is about to make.

So the business needs delivery capability as well, in-house or through a partner. When judging either one, look at what the people have actually delivered. That track record belongs to people and teams more than to a company’s age, which is why a bounded pilot tells you more than a years-in-business rule.

The private equity version

Sponsors face the same choice at fund level. A portfolio-wide Head of AI who selects tools for every company can reproduce the category error across the whole portfolio. The role can still be valuable, as the first essay said, when its mandate is written in business terms.

AI belongs in the value-creation plan. The useful question for an operating partner is which value-creation capabilities can now be redesigned. Pricing, sales effectiveness, service, working capital and acquisition integration are the usual candidates. The next questions are who at each company owns that redesign, and what that person is allowed to decide.

Where I sit: I advise through Kainora and work as a fractional Chief AI Officer, so I have an interest in this gap being recognized. I’m also building technology shaped by these views.

Two questions for you. Who in your company can say what it could now become capable of doing? And can that person change anything?

Part II, Busy With AI, looks at the companies where AI never had an owner at all.

Sources and further reading

Share your perspectiveConnect with Oleg on LinkedIn (opens in a new tab)