ROI Case File No.564: 'They Wanted to Use It, but Couldn't Let It Out'
They Wanted to Use It, but Couldn't Let It Out
Chapter 1: I Want to Use It, but I Can't Let the Information Out
"I want to use AI. But I can't send our know-how and secrets outside."
Mamoru Katayama, manufacturing director at TechDynamics, described the dilemma. His was a small-to-mid manufacturer with a lot of analog work. "We have a lot of analog tasks and I want to streamline them with AI. But we have no one skilled in AI or systems. Above all, we can't let our know-how and confidential information go outside. I'm trying ChatGPT, but I'm scared of how much I can put in, so I only type in harmless things."
"You want to use it, but you aren't using it fully," Claude asked.
"That's it," Katayama answered. "I'm scared of a leak, so I can't put in the very data that matters. So the streamlining stays half-done. I want to automate meeting minutes, go paperless, and do equipment troubleshooting, but I can't step in."
"Have you thought about a form that satisfies both the wish to use it and the need to keep it in?" I asked, to confirm.
"...I haven't," Katayama answered. "I thought it was a choice—use it, or protect it. I never thought both were possible."
"The dilemma of wanting to use it and not being able to let it out resolves if you judge by separated conditions," I replied. "Let's break this down with 5F."
Chapter 2: 5F Asks—Judge by Five Conditions
"This case calls for 5F."
Claude wrote on the whiteboard: "Fiability, Flexibility, Functionality, Feasibility, Finance."
"5F—Fiability, Flexibility, Functionality, Feasibility, Finance; reliability, flexibility, functionality, feasibility, and finance—judges a measure by five conditions to derive a form that holds," I explained. "The key is not to make wanting-to-use and wanting-to-protect a binary. Reliability keeps the secrets, functionality delivers what you want to do. Find the one point that satisfies all five at once, and the dilemma turns into coexistence."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. The data Katayama provided was entered.
"Here is the monthly cost," Gemini read out. "Inefficient labor from manual analog work: 170 hours per month on average, at ¥3,600 per hour, ¥612,000 per month. Unmet effect from half-done AI use out of fear of a leak: ¥380,000 per month. Manual labor for minutes and document processing: ¥340,000 per month. Person-dependence and response delay in equipment troubleshooting: ¥360,000 per month. Stalled adoption from a shortage of AI literacy: ¥300,000 per month. A total of ¥1,992,000 per month. Roughly ¥23.90 million per year."
Katayama stared at the figures. "I was only looking at preventing leaks. Add the cost of never using it fully, and it comes to this much."
"Then let's design it with 5F," I continued.
[Reliability—Keep the secrets in a closed environment]
"First, secure reliability," Claude said. "Run the AI in a closed environment, designed so internal data never leaves. Only with a foundation that keeps secrets can you put business data in with confidence. We resolve one side of the dilemma here."
[Flexibility—Fit the requirements of manufacturing]
"Next, build in flexibility," Gemini continued. "Make an AI that can be customized to manufacturing-specific requirements. Add speech recognition that handles dialects, so the floor's own words work as-is. Left generic, it won't take on the floor."
[Functionality—Deliver what you want to do]
"After flexibility, build functionality," I continued. "AI-driven automation of minutes, and accumulation and visualization of a troubleshooting knowledge base for equipment. Deliver what you wanted to do, inside the protection. Because there's reliability, you can step in."
[Feasibility, Finance—Judge by a form that runs and by cost]
"Finally, judge by feasibility and finance," Claude continued. "Raise in-house AI literacy through training so it runs on the floor. Choose a solution that holds initial investment down while securing long-term ROI. With all five conditions in place, the dilemma turns into coexistence."
[Estimating the payback]
"Let's run the numbers on ROI Proposal Generator," Gemini proposed.
- Initial cost: a closed-environment AI foundation, customization/speech recognition, minutes automation, troubleshooting visualization, and AI-literacy training—¥5.1 million total
- Monthly cost: system operation and maintenance combined, ¥200,000
- Monthly savings: automation of analog work = ¥430,000 (assuming a 70% reduction); achieving the effect by stepping into AI use = ¥340,000; automation of minutes and document processing = ¥340,000; more efficient troubleshooting = ¥360,000; totaling ¥1,470,000 per month
- Net monthly savings: ¥1,470,000 − ¥200,000 = ¥1,270,000 per month
- Payback period: ¥5.1 million ÷ ¥1.27 million = about 4.0 months
"Four months to recoup," Gemini summarized. "What works is not making it a binary of use-or-protect, but finding the coexistence point with five conditions. Don't use it out of fear and the effect is zero; drop the protection and use it and the risk remains. Protect with reliability and use with functionality, and the dilemma resolves. The investment doesn't miss."
Katayama checked the figures. "I was stuck on use-or-protect. Judge by separated conditions, and there's a form where both hold."
"5F is a tool for judging a dilemma by five conditions and making both hold," I replied.
Chapter 3: An Implementation Plan That Judges by Five Conditions
"Let me lay out the approach," I said, standing at the whiteboard.
"Month one—define the requirements of the closed environment and design for reliability. Month two—customize for manufacturing requirements and design speech-recognition flexibility. Months three and four—build minutes automation and troubleshooting visualization. Month five—AI-literacy training and floor application. Month six—trial operation and effect verification. Month seven onward—expand the scope of use and continuously check the five conditions."
"Can we really use it while keeping secrets?" Katayama asked.
"You can," Claude replied. "You can't because you're making protect-and-use a binary. Fix reliability first with 5F and build a closed environment, and you can put business data in without it leaving. Build functionality on that foundation, and you can step in and use it. Not 'don't use it in order to protect it,' but 'use it having protected it.' The dilemma turns into coexistence."
Katayama took notes. "Before agonizing over use-or-protect, judge by five conditions. Now I see the order."
Chapter 4: The Day Coexistence Became Possible
Ten months later, a report arrived from Katayama.
After automation, analog work dropped sharply. "Processing we ran by hand was automated with AI. What ended half-done in streamlining became something we could step into," Katayama wrote.
Minutes and document processing grew lighter, too. Automation handled the drafts, and people moved to checking. "Transcribing by hand after every meeting became something that came together automatically," the report read.
The biggest change showed in stepping into use. From unable to use it fully out of fear of a leak, to using it fully within the protection. "I couldn't put business data in out of fear. Once we built the closed environment first, I could step in with confidence," Katayama wrote.
Equipment troubleshooting sped up as well. Person-bound response was shared through visualized knowledge. "The 'can't fix it unless the expert's here' decreased," the report read.
As a side effect, the criterion for judgment changed. Not a binary of use-or-protect, but judging by conditions, took root. "We stopped halting at 'using it is scary.' We started asking how to protect it so we can use it," Katayama wrote.
At the end of Katayama's report was this: "I thought the AI-adoption struggle was fear of a leak. But the real problem was making use-or-protect a binary and giving up on coexistence. The moment we split it into five conditions with 5F, a form that uses it while protecting it came into view. Before stepping in, separating the conditions came first."
The day a company that wanted to use AI but couldn't let it out became a company that could use it while protecting it, AI adoption had shifted from shrinking back at leak risk to a design that judges by five conditions and makes both hold.
"AI-adoption requests usually arrive as 'I want to use it, but I can't let the secrets out.' But before giving up, there's a question to ask: are use and protect really a binary? What 5F asks is the five—reliability, flexibility, functionality, feasibility, and finance. Protect with reliability and use with functionality, and the dilemma turns into coexistence. The day a company that couldn't let it out could use it while protecting it, what changed was not the AI's performance but the very perspective that judges a dilemma by five conditions and makes both hold."
Related Files
Tools Used
- ROI Polygraph — Visualizing analog-work labor, unmet effect from half-done use, and person-dependence risk
- ROI Proposal Generator — Payback simulation for closed-environment AI adoption, starting from a coexistence design across five conditions
