ROI Case File No.529: The People Who Could Operate CAD Were Vanishing, Year by Year
The People Who Could Operate CAD Were Vanishing, Year by Year
Chapter 1: The People You Can Ask About Operation Are Disappearing
"The talent that can use CAD well is decreasing year by year. The people you can ask are disappearing."
Ryo Udagawa, design department manager at TechSphere, said this while showing a 3D model screen. "We use two CADs, NX and CATIA. The main one is NX, but there are still many CATIA users too. The operability and UI differ, so confusion arises on the floor. On top of that, the people who can do CAD modeling are decreasing, and licenses are short. The personnel dependency is serious."
"Concretely, what kind of support is needed?" Claude asked.
"Two big things," Udagawa answered. "Operation support for those with no CATIA experience. And support for CATIA-experienced people transitioning to NX. Many people stumble over the operational differences. Asking a veteran is fast, but those veterans are decreasing. And CAD beginners find it hard even to put into words what they don't understand."
"How are you thinking of proceeding?" I confirmed.
"We want an AI chatbot specialized for the CAD software," Udagawa answered. "One that teaches operation through dialogue and guides you from screen captures. But building it large-scale right away is frightening. We want to start from a PoC, see the effect, and expand. But across many wide-ranging problems, we can't judge which to tackle first for the highest return on investment."
"If the problems are wide-ranging, you need to judge priorities by return on investment," I responded. "Let's break it down with the ROI model."
Chapter 2: The ROI Model Asks: Judge Priorities by Return on Investment
"This case needs the ROI model."
Claude wrote "Return On Investment" on the whiteboard.
"The ROI model is a framework that, to maximize return on investment, proceeds through a sequence of steps: problem identification and prioritization, solution design, implementation and verification, and rollout and evaluation," I explained. "When problems are wide-ranging, tackling them all at once disperses the investment and dilutes the effect. We judge which problem, if solved, yields the highest return on investment, verify with a PoC, then expand. It's a technique for deciding the investment order by effect."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. He entered the data Udagawa had provided.
"The monthly CAD-operations cost came out," Gemini read aloud. "The labor of veterans taken up responding to CAD operation questions averages 200 hours a month; at an hourly rate of 5,600 yen, that's 1,120,000 yen a month. The productivity loss from the learning delay of CATIA-inexperienced and NX-transitioning people averages 900,000 yen a month. The opportunity loss from work backlog due to a shortage of CAD modeling talent averages 800,000 yen a month. The waiting-time cost from license shortages averages 400,000 yen a month. The expected value of the personnel-dependency risk in CAD operation know-how averages 600,000 yen a month. The cost of delayed resolution from beginners' difficulty verbalizing questions averages 300,000 yen a month. The total is 4,120,000 yen a month. Annualized, that's about 49.44 million yen."
Udagawa stared at the figures. "Even just the labor of veterans taken up answering questions—I had no idea it was this much. Add the learning delay and the talent-shortage opportunity loss, and the scale is different."
"Then let's judge priorities with the ROI model," I continued.
[Problem Identification and Prioritization]
"First, we lay out the problems and prioritize," Claude said. "CAD talent shortage, demand to shorten development periods, CATIA license shortage, stumbles in NX transition, beginners' difficulty verbalizing questions, operation support across UI environments—the problems are wide-ranging. Among these, where an AI chatbot produces the most effect is 'dialogue support for operation questions.' It directly cuts veterans' labor and speeds up learning. We concentrate investment here."
[Solution Design]
"Next, we design the solution," Gemini continued. "An AI chatbot that supports the basic operation and modeling methods of CATIA and NX through dialogue. Further, we implement a function that reads screen captures and guides operation. Even if a beginner can't put into words what they don't understand, showing the screen returns guidance. A design that crosses the wall of question verbalization."
[Implementation and Testing—Verifying Return on Investment With a PoC]
"The PoC phase," I continued. "Instead of rolling out company-wide right away, we verify the effect with small-scale deployment. We measure whether veterans' labor actually decreases and whether beginners' learning speeds up. We gather feedback and improve. We confirm the return on investment with real data before making the judgment for full rollout. The PoC is the investment's safety device."
[Rollout and Evaluation]
"Last is rollout and evaluation," Claude continued. "Once the PoC confirms the effect, we formulate a rollout plan for the whole department. Even after rollout, we continuously evaluate and report return on investment. We expand from the areas that produced results and rethink the design for areas that didn't. A structure that keeps making rollout judgments by numbers, with ROI as the axis."
[Estimating the Investment Recovery]
"Let's estimate with ROI Proposal Generator," Gemini proposed.
- Initial cost: CAD-specialized AI chatbot development, CATIA/NX operation data learning, screen-capture guidance function implementation, PoC environment construction, and field training—7,800,000 yen total
- Monthly cost: Chatbot operation plus model-update ongoing cost—280,000 yen a month combined
- Monthly reduction effect: Veteran question-response labor reduction = 780,000 yen a month (assuming 70% reduction), productivity recovery from learning delay = 630,000 yen a month, elimination of work backlog = 500,000 yen a month, license waiting-time reduction = 250,000 yen a month—2,160,000 yen a month total
- Monthly net reduction: 2,160,000 yen − 280,000 yen = 1,880,000 yen a month
- Payback period: 7,800,000 yen ÷ 1,880,000 yen = about 4.1 months
"Recovery in a little over four months," Gemini summarized. "What matters is that, among wide-ranging problems, we concentrate investment in operation support, where the effect is largest. We don't tackle all problems at once but start from the high-ROI area. Because we confirm the effect with a PoC before expanding, the risk of investment failure is low."
Udagawa confirmed the figures and said, "There were so many problems I couldn't decide which to tackle first. When you line them up by return on investment, operation support turns out to work best."
"The ROI model is a tool for lining up too many problems in order of effect," I responded.
Chapter 3: A Deployment Plan That Expands From a PoC
"Let me organize how we'll proceed," I said, standing before the whiteboard.
"Months one and two—finalizing the problem priorities, collecting CATIA/NX operation data and FAQs, and designing the chatbot specifications. Months three and four—developing the AI chatbot and implementing the screen-capture guidance function. Month five—starting the PoC (small-scale deployment), trial operation with some teams in the design department. Month six—gathering feedback and improving, and verifying return on investment. Month seven—after confirming the effect, formulating a rollout plan for the whole design department. Month eight onward—departmental rollout, continued ROI evaluation, and expanding the scope of application."
"If the PoC doesn't produce an effect, what do we do?" Udagawa confirmed.
"That's the judgment we do the PoC for," Claude responded. "If we try it at small scale and it doesn't produce an effect, we rethink the design or redirect the investment to a different problem. The loss is far smaller than building large-scale and then failing. The ROI model is a structure that makes the full investment only after gaining proof that an effect will appear. A PoC isn't waste—it's a process that creates the basis for the investment judgment."
Taking notes, Udagawa said, "I was trying to solve all the problems at once. With the order of lining them up by effect and starting from a PoC, it finally looks like I can move."
Chapter 4: The Day Someone to Ask Is Always There
Nine months later, a report arrived from Udagawa.
Veterans' question-response labor was cut 70% versus before, after full rollout following the AI chatbot's PoC. "Operation questions are answered first by the chatbot. What comes to veterans is only the truly difficult cases. The veterans' time returned to design work," Udagawa wrote.
Beginners' learning also sped up. The function that guides operation from screen captures supported beginners who couldn't put their questions into words. "A beginner in the state of 'not knowing what they don't know' gets guidance just by showing the screen. Resolving stumbles got faster, and the period to becoming independent shrank," the report said.
The biggest change appeared in the coexistence problem of CATIA and NX. The floor, which had been confused by the operational differences of the two CADs, was bridged by the chatbot's guidance. "When a CATIA-experienced person moves to NX, the chatbot teaches the operational correspondences. The stumbles of transition decreased significantly," Udagawa wrote.
The PoC judgment also functioned. Because we confirmed the effect with small-scale deployment before expanding, the investment wasn't wasted. "We could confirm by numbers in the PoC that 'this works.' So the budget for departmental rollout passed smoothly. Had we built large-scale right away, it wouldn't have gone like this," the report said.
As a secondary effect, accumulation of know-how advanced. The logs of questions to the chatbot visualized the floor's stumbling points. "Where everyone stumbles is visible in data. The focus of training became clear," Udagawa wrote.
Resilience to talent shortage also rose. With CAD operation support coverable by AI, the dependency on veterans eased. "Even as the people who can operate CAD decrease, it's becoming a structure where the floor runs. The sense of crisis about personnel dependency has surely eased," the report said.
At the end of Udagawa's report, he had written: "There were so many problems I couldn't decide which to tackle first. With the ROI model, I lined them up in order of return on investment, verified with a PoC, then expanded. Too many problems, judged by effect, settle into an order."
It was recorded that the day a design department that had been dreading the prospect of CAD operators vanishing year by year gained a day where someone to ask is always there, CAD support changed from a craft that relied on people into a mechanism that always responds.
"The more wide-ranging a consultation's problems, the more people freeze up. They want to solve this and that, the investment disperses, and everything ends up half-finished. What the ROI model asks is which problem, if solved, yields the highest return on investment. Line them up by effect, concentrate investment in the area that works most, and expand after gaining proof with a PoC. When a design department where CAD operators were vanishing year by year gained a day where someone to ask is always there, what changed was not the chatbot's features but the very judgment of cutting through too many problems by effect."
Related Files
Tools Used
- ROI Polygraph — Visualizing veteran question-response labor, learning-delay loss, and CAD personnel-dependency risk
- ROI Proposal Generator — Investment-recovery simulation for PoC-rooted CAD-support AI deployment
