Contents
Every AI programme eventually meets the same question from the boardroom: what is the return? At a panel in Tokyo, two CIOs running large Japanese operations of global manufacturers gave an answer that was less about spreadsheets and more about sequence. Get the people ready, find where the organisation genuinely hurts, prove value in a pilot, and the business case tends to write itself.
The discussion was moderated by Durjoy Patranabish, VP and Head of Global Business at Tiger Analytics, who leads the firm's work outside North America. He was joined by Moran Gelber, CIO of Volkswagen Group Japan, which provides IT for Volkswagen, Audi, Bentley and Lamborghini in the country, and Manzur Mahtab, CIO and Executive Officer at Carrier HVAC Japan, the former Toshiba Carrier, where he led the post-acquisition integration and now drives digital transformation.

Durjoy Patranabish
VP, Head of Global Business, Tiger Analytics
Moran Gelber
Chief Information Officer, Volkswagen Group Japan
Manzur Mahtab
Chief Information Officer - Executive Officer, Carrier HVACPatranabish framed the problem bluntly. AI journeys often begin with executive noise: "We have to have AI." The real work starts after that, when someone has to explain why, what the benefit is, and whether the data is even ready. That early design stage, he argued, decides how much return an organisation will ever see.
Two kinds of ROI, and a licence to drive AI
Gelber opened with a caution: "ROI" means different things depending on where AI is being applied. Gelber split it into two pillars. The first is embedding AI into systems, described as the easier of the two. The second is embedding AI into the daily work of every employee, and that is a much longer journey. When a CEO asks for AI and ROI in the same breath, it is often unclear which pillar they mean.
Partner with The CIO Room
Reach the leaders this piece is about
Four or five partners per room, alongside 50–100 CIOs and enterprise AI leaders. Tell us who you want to meet.
The second pillar has a people problem at its core. There remains a large gap between what employees know and what today's tools can do. Volkswagen Group Japan's response borrows from its own industry: an AI "driving licence." Every employee follows a training plan and then sits a test. Only those who pass get access to the tools.
The point is not gatekeeping for its own sake. The programme ensures people understand how to use AI, what can go wrong when it is misused, and the boundaries of what the company provides. Around a third of employees have been onboarded so far, with the rest to follow.
Convincing the CFO: Carrier's review board and payback test
Mahtab agreed that everyone now wants AI, and that tools like Copilot have settled the question of whether it is here to stay. The hard part is showing tangible value. "Convincing the CFOs," he said, "is the toughest."
Carrier HVAC Japan handles this through an internal AI review board. As employees put forward ideas, the board asks a consistent set of questions. What business problem does this solve? Is it an individual problem or an enterprise one? What are its impact, priority and scalability?
The filter that matters most is payback. Proposals are sorted by whether they improve productivity, improve efficiency or create new business value. They then go through the internal digital team and a panel of business leaders before reaching the executive committee for sponsorship. If a project shows payback within 24 months, soft or hard, it can draw on an innovation fund the company created for AI and data.
Mahtab expects this scrutiny to fade. Nobody asks for the ROI on electricity or utilities, he noted, and he hopes AI will one day be budgeted the same way. For now it sits in capex, and capex invites questions. Patranabish agreed: in 30 years, asking whether a company needs AI may sound like one of the dumbest questions anyone could ask.
Follow the pain, and the ROI follows
With limited capacity for proofs of concept, how do you choose which ones to run? The two CIOs approached it from different directions.
For Mahtab, maturity comes first. Carrier began by cleaning up its data and checking its integrity, because otherwise AI adds no value. His selection criteria also reflect local conditions: Japan has an ageing population, factories need productivity, and hiring is hard.
Gelber went further and set ROI aside as the starting point. Companies already have plenty of internal problems to solve before chasing new business. Volkswagen Group Japan still runs on a lot of Excel, manual work and copy-paste. "If a CFO would come to me and say where is the biggest ROI," Gelber said, "I would say where is the biggest pain point." Solve that in a pilot, and convincing the rest of the board becomes much easier.
Patranabish summed it up: focus on the biggest hurdles in processes, and the return should follow soon after.
When AI goes into the product
Patranabish asked whether pilots were mostly about internal processes or about the products themselves. The answers depended on what each company does in Japan.
Volkswagen Group Japan does not manufacture vehicles locally, so Gelber's focus is on processes, including the many involved in running an import business.
Carrier offered the clearest product story of the session. Every chiller it sells is connected, sending sensor data to a command centre in Japan, where the company holds a 47% share of the commercial chiller market. Inspections and reports used to be done manually by field technicians. Carrier built a platform with embedded AI that now handles digital inspection and predictive maintenance.
Mahtab described the customer experience in a Tokyo summer: a notice that a part will be needed within a month, before the air conditioning fails. That service now generates what Carrier calls digitally enabled revenue through service agreements, and customers are willing to pay for it. Asked about the ROI, Mahtab said the payback came within a few months. The business no longer asks the question; it simply wants the platform expanded and scaled.
From the sterile pilot to the real world
Scaling is where pilots meet reality. As Patranabish put it, a proof of concept lives in a box. Moving it to production touches new stakeholders and raises infrastructure, people and delivery-time hurdles. A team may finish ten pilots but cannot productionise them all at once.
Gelber's rule stays the same: scale the pilot that solves the biggest pain point. That choice also exposes what the pilot did not anticipate. Pilots often run in a fairly sterile environment, and problems such as data volume only appear at scale. Choosing the target early shows which stakeholders to involve, whether people need training, and whether data must be reorganised.
What about competing claims, when finance, procurement and customer service each insist their problem is worst? Gelber's answer is to zoom out to the company view. The biggest gain might sit with a very small team, such as a PDI group in Toyohashi that is losing large amounts of time and resources. Fixing that may deliver the best return, and it becomes the argument taken to the board.
Mahtab observed that functions like finance and procurement often need business intelligence and clean data before they need AI. He then shared a project now scaling at Carrier: AI OCR combined with RPA. Fax remains stubbornly alive in Japan, and Carrier's sales and service teams still receive an estimated 50,000 to 60,000 inquiries and parts orders by fax, across roughly 1,000 template patterns. Modern OCR can now read even handwriting, and linking it to RPA frees salespeople from back-office processing.
The approach travels. Markets in the Middle East and Southeast Asia rarely use fax, but they rely on PDFs, so the same pipeline can be replicated there. Carrier also tracks adoption closely, because enthusiasm for AI tools tends to drop after a few months. Its reporting shows how many people are using each tool and how much flows through it.
Does pain always mean profit?
An audience member pushed back on the pain-point principle. It is a good rule of thumb, he said, but internal pain does not always translate into value. People complain about many things that never move the numbers. Product pain points are easier to link to return. He asked what benefits framework should come next, citing models such as RICE that weigh reach and impact.
Gelber accepted that processes are harder to measure than products. Rolling out Copilot to everyone and sprinkling AI everywhere will not help by itself. The real question is whether people use these tools as a glorified search engine or to create value. That is why the driving licence matters: trained employees can point to their own pain points more precisely.
Gelber gave a concrete case. Dealer complaints arrive through the call centre and must be analysed to find whether the cause is a system, a feature or something else. With more than 500 dealers across Japan, doing that manually takes hours. AI does it far faster, and time saved eventually becomes money, or capacity to invest elsewhere.
Patranabish added the view from consulting. Clients mostly ask for AI in two places: manual processes that agents and agentic workflows can take over, and undocumented knowledge that AI can capture. Sales and marketing have long led on data use. Finance, HR and procurement, long behind the scenes, are now getting attention because of the manual work buried in them.
The human side of the business case
If AI changes roles, should IT partner with HR to rewrite job descriptions and day-to-day work
- Just deliver tools and training
Mahtab's answer was grounded in his company's history. Toshiba is a traditional company of more than 130 years; Carrier, he joked, is a "100-plus-year-old startup." Change moves at that speed. Many employees with 30 or 35 years of service see real risk to their job security, and that can lead them to hold information back.
Rather than going through HR, Mahtab works directly with business line leaders. He tries to understand what they need and propose solutions that let people reskill and move toward work such as sales. Winning over HR has been hard. Going to the gemba, the shop floor, with the business leader, works.
The takeaway
Neither CIO treated ROI as a number to calculate up front. Both treated it as something earned in stages.
At Volkswagen Group Japan, that means licensing people before giving them tools, then aiming AI at the worst manual bottlenecks. At Carrier HVAC Japan, it means clean data first, a review board with a 24-month payback test, and pilots that become revenue lines, like the chiller platform that paid back in months.
It is missing from projects that start with the technology
- The problem, or with tools instead of people
CIO summit Tokyo · Compare this with our AI conference for CIOs.










