ROI Case File No.530: We Couldn't Even Tell Where to Begin
We Couldn't Even Tell Where to Begin
Chapter 1: Even Extracting the Problems Was a Struggle
"We want to advance DX. But we're already stuck at the stage of identifying what the problems are."
Keiichi Karasuyama, CEO of GlobalConstruct, said this while showing photos of the field. A construction, civil-engineering, and contracting company. "Our business processes aren't organized. So we can't see where to digitize from. We want to use generative AI, but we don't know how to utilize it on the field either. And we can't set priorities."
"What about the site managers' working hours?" Claude asked.
"Long working hours have become the norm," Karasuyama answered. "Legally, too, we're required to shorten working hours. But the site managers' work is personnel-dependent, and we can't even grasp how much time goes into what. Information on past troubles and complaints also stays in individuals' memories and isn't turned into knowledge."
"Have you ever organized your business processes before?" I confirmed.
"Never," Karasuyama answered. "The way of proceeding differs by site. The problems people carry differ by role. We've never put the whole thing into a single picture. That's why I judged we need an outside DX consultant. Honestly, though, we can't even see the entry point of where to begin."
"The reason the problems can't be seen is that the people who carry out the work aren't defined as figures," I responded. "Let's break it down with Persona_Analysis."
Chapter 2: Persona_Analysis Asks About the Figure Behind Each Role
"This case needs Persona_Analysis."
Claude wrote "P, A" on the whiteboard.
"Persona_Analysis—persona analysis—is a framework that depicts users or the people who carry out work as concrete figures with attributes, behaviors, and problems, and proceeds with design from their viewpoint," I explained. "It's known in marketing, but it's effective for business reform too. Because the vague lump of 'the company's problems' can't be handled, but break it down into 'what a figure called a site manager does in a day and where they get stuck,' and the problems become concretely visible. We draw a persona for each role and visualize the business processes."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. He entered the data Karasuyama had provided.
"The monthly opportunity-loss cost came out," Gemini read aloud. "The expected value of excess labor cost and correction risk from site managers' long working hours averages 2,200,000 yen a month. The inefficient labor from un-organized business processes averages 360 hours a month; at an hourly rate of 4,500 yen, that's 1,620,000 yen a month. The countermeasure cost of past trouble information going unused and recurring averages 1,100,000 yen a month. The management labor spinning its wheels because DX priorities won't settle averages 700,000 yen a month. The expected value of the skill-succession risk from knowledge personnel dependency averages 600,000 yen a month. The total is 6,220,000 yen a month. Annualized, that's about 74.64 million yen."
Karasuyama stared at the figures. "I was conscious of the cost of long working hours. But when you put a price on the inefficiency of un-organized processes and the dormancy of trouble information, I had no idea it was this much."
"Then let's design with Persona_Analysis," I continued.
[Persona One—The Site Manager: A Core Figure Pressed for Time]
"First, we draw the site manager's persona," Claude said. "'In their forties, handling construction management, juggling multiple sites, shuttling between site and office from morning to night, relying on memory and experience for past trouble response, with no time to leave records.' Follow this figure's day, and where time is robbed—travel time, report creation, trouble-response judgment—becomes concretely visible."
[Persona Two—The Young Engineer: A Layer Searching for Someone to Ask]
"Next, the young engineer's persona," Gemini continued. "'In their twenties with little experience, wanting to ask a veteran when uncertain in judgment but hard to catch one on site, wanting to reference past cases but not knowing how to search for them.' This figure's troubles lie in the lack of a means to access knowledge. It becomes visible that they have a different problem structure from the site manager."
[Persona Three—Management: A Layer That Wants to See the Whole but Can't]
"The management persona," I continued. "'Wanting to raise company-wide productivity but unable to see in numbers what's happening on site, bearing the duty to achieve working-hour reductions, with no material to judge where to invest.' When you draw a persona for each role, it becomes clear that, even with the same DX, the scenery they see is completely different."
[DX Measures by Persona—Pinning Down Where Generative AI Lands]
"We apply concrete DX measures to each persona's problems," Claude continued. "For the site manager, real-time work support using generative AI and risk prediction from past knowledge. For the young engineer, AI support that draws out past cases through dialogue. For management, a dashboard that visualizes field data. Because we pin down where it lands per persona, generative AI changes from 'a tool we don't know what to use for' into 'a tool that solves this person's this problem.'"
[Estimating the Investment Recovery]
"Let's estimate with ROI Proposal Generator," Gemini proposed.
- Initial cost: Persona design, business-process visualization, site-manager generative-AI work support, a risk-prediction system from past knowledge, AI support for young engineers, a management-visualization platform, and field training—13,200,000 yen total
- Monthly cost: AI operation plus data-update platform ongoing cost—480,000 yen a month combined
- Monthly reduction effect: Excess-cost reduction from shortening site managers' working hours = 1,600,000 yen a month, process efficiency = 960,000 yen a month, trouble-recurrence prevention = 760,000 yen a month, elimination of wheel-spinning by clarifying DX priorities = 500,000 yen a month—3,820,000 yen a month total
- Monthly net reduction: 3,820,000 yen − 480,000 yen = 3,340,000 yen a month
- Payback period: 13,200,000 yen ÷ 3,340,000 yen = about 4.0 months
"Recovery in four months," Gemini summarized. "What matters is that, by applying measures per persona, the generative-AI investment doesn't swing and miss. It's not a vague investment in 'the company's DX' but an investment that 'solves this problem of the site manager,' so the effect can be measured. The cost reduction from shortening working hours especially works hard."
Karasuyama confirmed the figures and said, "I was thinking in the too-large lump of 'the company's problems.' When you split into personas, whose which trouble you solve becomes visible. Where to use generative AI also becomes concrete."
"Persona_Analysis is a tool for turning a vague lump into a concrete figure's problems," I responded.
Chapter 3: A DX Plan That Proceeds From Personas
"Let me organize how we'll proceed," I said, standing before the whiteboard.
"Months one and two—interviewing the major roles, designing personas for the site manager, young engineer, and management, and visualizing the business processes. Month three—prioritizing problems by persona and finalizing where generative AI lands. Months four and five—building the real-time work support and risk-prediction system for site managers. Month six—trial operation at a pilot site and effect verification. Month seven—rolling out measures for young engineers and management. Month eight onward—continuing to turn past troubles into knowledge, and designing additional personas and expanding measures."
"Which persona should we start with?" Karasuyama confirmed.
"With the site manager," Claude responded. "There's a legal requirement to shorten working hours, and the cost-reduction effect is also the largest. It's the top-priority persona. If the site manager's work gets lighter, that effect ripples out to the whole company. Rather than starting all personas at once, we solve from the figure with the largest impact. This is prioritization using persona analysis."
Taking notes, Karasuyama said, "We couldn't even extract the problems, yet just drawing one persona makes it this concrete—I didn't expect that."
Chapter 4: The Day Whose Which Trouble It Is Became Visible
Ten months later, a report arrived from Karasuyama.
Site managers' working hours were substantially shortened four months after the generative-AI work-support system went live. With report creation and trouble-response judgment supported, monthly excess labor visibly decreased. "Site managers who had been chased by record-keeping could leave the site at close to regular hours. We now have a prospect of meeting the legal requirement to shorten working hours," Karasuyama wrote.
Turning past troubles into knowledge also advanced. Trouble response that had stayed in individuals' memories began to accumulate in the generative-AI risk-prediction system. "Relying on the memory of 'there was a similar trouble at that site' was replaced by data. Recurrence prevention began to run as a mechanism," the report said.
The biggest change appeared in the entry point of DX becoming visible. From a state where even extracting problems was a struggle, it was broken down into concrete measures per persona. "The cloud-grasping talk of 'the company's DX' changed into the concrete 'support this work of the site manager.' Where to begin became visible," Karasuyama wrote.
Support for young engineers also produced results. With AI support that draws out past cases through dialogue, the time spent searching for a veteran decreased. "When a young engineer is uncertain in judgment, they can ask the AI first. The frequency of stopping a veteran's hands dropped," the report said.
Management's field of view also changed. With field data visualized, the field situation that had been invisible could now be grasped in numbers. "For the first time, what's happening on site became visible in numbers. We got the material to judge where to invest," Karasuyama wrote.
As a secondary effect, field-born improvement proposals increased. Through the persona-design interviews, the field gained the experience of verbalizing its own problems. "Once they understood that their own troubles connect to measures, the field began to speak up. It stopped being forced-on-us DX," the report said.
At the end of Karasuyama's report, he had written: "Trying to extract the problems of DX, I'd been stuck at its entry point. When I drew a figure for each role with Persona_Analysis, whose which trouble we solve became visible. The company's problems can't be solved as a lump. They start to move only once broken down into figures."
It was recorded that the day a company that couldn't even tell where to begin became a company that can discern whose which trouble it is, DX changed from a cloud-grasping directive into figure-rooted concrete improvement.
"Many companies stall at the problem-extraction stage of DX. The lump of 'the company's problems' is too large, and where to begin can't be seen. What Persona_Analysis asks is the resolution of the figures who carry out the work. The site manager, the young engineer, management—draw a figure for each role, and what each one does in a day and where they get stuck become concretely visible. Where generative AI lands is also settled from that figure's problems. The day a company that couldn't even tell where to begin discerned whose which trouble it is, what changed was not the DX tool but the very perspective of breaking problems down into figures."
Related Files
Tools Used
- ROI Polygraph — Visualizing site managers' long-working-hour cost, the inefficiency of un-organized processes, and knowledge dormancy
- ROI Proposal Generator — Investment-recovery simulation for role-based, persona-rooted DX promotion
