Nobody Signs for AI

Is AI a CIO initiative, a CDO initiative, or a business initiative?

Alfee Lee has watched a C-suite argue that question to a standstill. He is the vice president and head of international business for Singapore and Thailand at Fujitsu, 20 years into the company, and he raises it the way a doctor raises a symptom.

Most boards never settle it. They fund it anyway, and nobody signs for the result.

Three budgets wearing one name

Lee splits enterprise AI into three layers, and the useful thing to note is that each layer has a different owner, a different sponsor, and a different way of failing.

The first layer is infrastructure AI — the machine intelligence now baked into every network, endpoint, and observability tool an enterprise already owns. That belongs to the chief information officer, cleanly, and Lee says it is where most of them are already spending.

The second is internal adoption: AI applied to the company’s own processes across HR, finance, legal, procurement. This is where it gets messy. In Lee’s telling, CIOs keep getting handed the chair of an internal committee, told to gather cross-functional teams, hunt for use cases and drive the agenda. “That’s a bit tricky,” he says — an understatement that will land with anyone who has been volunteered for that job.

The third is business AI, where the sponsor is the business itself, and IT is the enabler. Lee has seen this movie before, in the smart factory and smart retail waves. “Now we are just complementing, or an enabler, for that business case.”

Three layers, three owners, one budget line, and one two-letter acronym. When a board asks “what’s our AI strategy,” it is usually asking about all three at once and getting an answer about one.

Why the org chart is the bottleneck

The failure data backs him up, and it points at governance rather than at the models.

MIT’s NANDA initiative found that 95% of enterprise generative AI pilots delivered no measurable P&L impact, attributing the gap to integration and organizational learning rather than model performance. Gartner separately predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.

Both deserve a caveat. The MIT figure has drawn methodological pushback, and “no measurable ROI” often means nobody set up the measurement. But the direction holds across studies, and it matches what Lee describes from the field. The constraint is not the technology. It is who is accountable for it.

Selling your own homework

Fujitsu is in an unusual position to make this argument, because it ran the experiment on itself first.

In October 2020, the company launched Fujitra — Fujitsu Transformation — a company-wide overhaul backed by more than JPY100 billion, owned by chief executive officer Takahito Tokita in his second hat as chief digital transformation officer, run day to day by Yuzuru Fukuda, the CIO who had joined from SAP Japan months earlier, with DX officers appointed across 15 corporate and business units and five overseas regions.

“Customer says, ‘Alfee, I need to put all the data into one platform.’ I will say no.” — Alfee Lee @ Fujitsu

Lee is one of the downstream consequences. He was named CDO for ASEAN, and he is honest about the shape of the assignment: “Even I was appointed as CDO for ASEAN as my additional task.”

A company that spent nearly a billion dollars on its own transformation still runs its regional CDO function as somebody's side job. That is not a knock on Fujitsu but a realistic picture of the work — and more useful than any maturity model.

It has also become a sales asset. Customers now ask Fujitsu to present its internal journey to their own CEOs. “It’s not only about implementing the AI,” Lee tells them. “It’s also how you want to structure your organization to empower this team to transform.”

Breaking the deadlock

Lee’s method for resolving the three-way standoff is deliberately low-tech, and not borrowed from a consulting deck. Design thinking and agile were the two frameworks Fujitsu named at Fujitra's launch as the means of breaking its own “vertical division between different units.” Lee now runs the same instrument on customers.

Get the business heads and the CEO in a room. Ask what the ideal looks like. Test what is viable. Then assign the thing that actually matters. “From there we drive the accountability of what they need,” he says, “and we emphasize that technologies are the enabler.”

Lee applies the same logic to data, pushing back on the reflex to centralize. “Customer says, ‘Alfee, I need to put all the data into one platform.’ I will say no.” The goal is the ability to pull data in, process it and push it back into decisions — not a single physical repository. Fujitsu sells that as Data Intelligence PaaS, a cloud platform for consolidating scattered data, which in turn sits on infrastructure including Palantir Foundry and Azure. Lee says customers almost always want a proof of concept first.

Buy the outcome, not the parts

The same gap shows up in how enterprises buy. Which brings Lee to procurement, where he is most direct.

Enterprises routinely split the purchase. One tender goes out for contract staff to run a function; a separate tender goes out for the software those staff will use. Two contracts, two winners, two scopes and nobody accountable for the join.

Lee has spent years on the receiving end of that split. “Today I’m providing a service based on what you purchase,” he says. The vendor can deliver only what it was sold, never the thing that spans both. “You will suffer, because I cannot join together.”

His alternative is to contract for the result. Take the service desk as an example: the conventional buy is licenses plus headcount. The outcome buy is employee experience, with the vendor free to reach for auto-remediation engines that close tickets without a human touching them — and, more importantly, carrying the risk if they don’t. “I will manage this for you,” as he puts it. “I can build an outcome for you, other than a solution for you.”

That reframing also disciplines the partner conversation. Lee tells vendors there are two ways in. “One is your product play. I respect that; I will go with you whenever it’s needed by the customer. But integrating a solution play is always based on outcome.” No single company, he adds, has the budget to build everything itself.

The timing is not incidental. Lee sees customers coming out of restructuring with fewer people managing more surface area, and outcome contracts move integration risk onto the vendor exactly when enterprises can least absorb it.

The three-year clock

Here is where Lee breaks from the prevailing mood.

Singapore became a super-aged society this year, and Prime Minister Lawrence Wong has framed AI explicitly as the answer to structural labor constraints. The pressure to move is real, and the vendor incentive is to amplify it.

Lee does the opposite. “If you start now, probably three to five years, you will get the result of the basic foundation of what you want. You might not see the urge or the needs now.”

He is arguing that compounding starts at the foundation layer, and that boards mistaking a fast pilot for a fast capability are watching the wrong clock. It is also the hardest version of the accountability problem. A three-to-five-year foundation bet needs someone willing to own a result that arrives after their own tenure ends. That is a signing problem long before it is a technology one.

Fujitsu shortened its own research cycle for the same reason. It used to hold new IP internally until a late-stage viability review; Kozuchi now pushes unfinished technology out to employees and customers early. That’s because a five-to-10-year path from lab to market no longer survives a market where, as Lee puts it, “every three months you’ll see a new technology coming up.”

Ask where the discipline comes from, and he goes back to being 30, leaving the army, walking into IT with no formal background. “I’m quite clear what I want to change,” he says. “I’m not clear where I want to go.”

For most boards, that is the honest version of an enterprise AI strategy. The trick is deciding who signs.

Image credit: iStockphoto/patpitchaya