ROI Case File No.505 'Why the AI Training Heat Never Reached the Work'
Why the AI Training Heat Never Reached the Work
Chapter One: Training Ended, but the Work Didn't Change
"It gets exciting during the training. Once it ends, nothing in the work has changed."
Narumi Kuwabara, Director of Administration at EduTech Innovations, showed us the training survey. Ninety percent of attendees answered they "want to apply this to their work." "But two months after the training, fewer than 20% have actually begun using AI in their work. The remaining 80% lose heat without knowing where to start."
"How large is the faculty and staff?" Claude asked.
"120 people," Kuwabara answered. "Including office staff and teachers, 100 were training participants. There's wide variation in follow-through afterward. The gap is widening between staff who know AI well and staff who can't use it at all."
"There are school-specific constraints, too?" Gemini asked.
"Personal information," Kuwabara answered. "Student data, parent data, grade data—we can't feed these into external AI services. We need an in-house environment, and that isn't set up either. Staff are experimenting with personal ChatGPT subscriptions, and governance isn't in place."
"There's also a heavy administrative burden, I understand," I confirmed.
"Teacher time with students is being eaten by administrative work," Kuwabara answered. "If AI could streamline that work, that time goes back to their core function. That was the purpose of the training. But training alone doesn't reach there."
"There's a structural gap between training and work," I replied. "Let's redesign with LOGIC."
Chapter Two: LOGIC Asks—The Five-Stage Build
"This case calls for LOGIC."
Claude wrote five letters on the whiteboard. L, O, G, I, C.
"LOGIC stands for Learn, Organize, Generate, Implement, Control—a five-stage framework for advancing an initiative," I explained. "Training is only the Learn stage. Skipping Organize, Generate, Implement, and Control means learned knowledge never reaches the work. Most AI adoption stalls come from missing post-Learn stage design."
"First, let's measure the current cost," Gemini said, opening ROI Polygraph. He input the data Kuwabara provided.
"The monthly administrative cost has come out," Gemini read. "Administrative time across 50 teachers averages 750 hours per month at 4,000 yen per hour, or 3 million yen monthly—time that should be going to student interaction. Routine work by 50 administrative staff averages 600 hours per month at 3,000 yen per hour, or 1.8 million yen monthly. Workflow delay and variability from AI skill gaps cost 300,000 yen monthly. Governance risk from personal AI usage has an expected cost of 200,000 yen monthly. Training investment isn't being recovered—an opportunity cost of 500,000 yen monthly from training fees that don't translate into improvements. Total: 5.8 million yen monthly. Annualized: approximately 69.6 million yen."
Kuwabara looked at the figures. "Including the lost training value, the room for improvement is significant."
"Now let's design with LOGIC," I continued.
[L—Learn: Redesign Learning at the Point of Contact With Work]
"Training has already been done, but we reinforce it," Claude said. "Generic AI training tends to end at feature explanations. We rerun training with explicit ties to specific work. For administrative staff: 'document drafting, inquiry responses, data aggregation.' For teachers: 'lesson material creation, evaluation feedback, parent communication.' Concrete examples show how AI connects to each job."
[O—Organize: Organize Work Processes]
"Next, we list and prioritize the work that AI can streamline," Gemini continued. "Trying to AI-ify every task fails. Solicit ideas from staff and map them on a two-axis chart: effect and feasibility. Start with high-effect, high-feasibility tasks. Building organizational agreement on the priorities is what drives later adoption."
[G—Generate: From Ideas, Generate Real Projects]
"At Generate, we launch small projects for each priority task," I continued. "Not company-wide rollout but pilots of about five staff each. A lesson-material team, an automation team, an inquiry-response team—each moves small and shares results. Small starts produce success stories."
[I—Implement: An In-House AI Environment]
"At Implement, we build an in-house AI environment," Claude continued. "We deploy AI infrastructure that doesn't send data externally, allowing student information and grade data to be safely fed to AI. This widens the range of work staff can do with AI without governance concerns."
[C—Control: Effect Monitoring and Continuous Improvement]
"Finally, control and continuous improvement," Gemini continued. "Monthly monitoring of post-deployment time savings, staff AI usage rates, and changes in work quality. Individualized follow-up for staff with skill gaps. Skill gaps won't disappear in the short run, but the range of practical use can be leveled."
[Estimating the Payback]
"Let's run it through ROI Proposal Generator," Gemini proposed.
- Initial cost: 8.4 million yen (in-house AI platform build, retraining program, pilot team coaching, work manuals, continuous monitoring design)
- Monthly cost: 280,000 yen (AI platform fees and ongoing coaching combined)
- Monthly savings: teacher administrative time reduction = 1.2 million yen (40% reduction assumed); administrative routine work reduction = 720,000 yen; resolving skill-gap delays = 200,000 yen; governance risk reduction = 150,000 yen; recovery of training investment = 300,000 yen. Total: 2.57 million yen monthly
- Net monthly savings: 2.57 million − 280,000 = 2.29 million yen
- Payback period: 8.4 million yen ÷ 2.29 million yen ≈ 3.7 months
"Under four months for payback," Gemini summarized. "The educational value of teacher time returning to students is also large in ways the numbers don't capture. The essential purpose is to build a structure that recovers the training investment."
Kuwabara looked at the numbers. "I can see structurally why training alone wasn't enough."
"Learning is a beginning, not an end," I replied.
Chapter Three: A Five-Stage Rollout Plan
"Here's the implementation plan," I said, standing at the whiteboard.
"Month 1: inventory work processes, identify priority tasks, solicit AI ideas. Month 2: select and prepare the in-house AI platform, design the retraining program. Month 3: deliver retraining, form three pilot teams. Months 4–5: run pilots, accumulate success stories, fully launch the AI platform. Months 6–7: replicate success stories, staged rollout to all staff. Month 8 onward: monthly monitoring, continuous improvement."
"What's the size of the pilot team?" Kuwabara confirmed.
"Fifteen people across three teams," Claude replied. "Not everyone at once—first produce results with fifteen. Once the results are visible, voluntary participation by other staff increases. Eighty percent stopped at training because participation motivation was weak. The pilot's success becomes the best training material."
Kuwabara took notes. "Learn, Organize, Generate, Implement, Control—it explains beautifully why training alone stalled."
Chapter Four: The Day Teachers Got Their Time With Students Back
Ten months later, a report arrived from Kuwabara.
Four months after the AI platform launched, teacher administrative time fell 38% versus baseline. Lesson material creation, parent-facing documents, and attendance reporting saw the largest impact. "Teachers who used to stay late are now leaving on time, with student interaction during the day," Kuwabara wrote.
Administrative routine work also dropped roughly 40%. AI-drafted inquiry responses, AI-generated document templates, and data aggregation automation came online sequentially. "Administrative staff time began flowing into strategic work and teacher support," the report said.
The most surprising change appeared in AI skill gaps. The gaps that had been wide right after training narrowed through shared success stories from the pilot teams. "Veteran teachers had said it 'didn't apply to me.' Seeing pilot teachers produce results, they started learning on their own," Kuwabara wrote.
The in-house AI platform launched on schedule. The range of work that could use AI on student data expanded, and governance violation risk structurally dropped. "Personal subscriptions nearly disappeared. Just preparing the platform changed how staff use AI," the report said.
As a side effect, student satisfaction scores rose. Comments like "the teacher has more time to listen to me" and "feedback comes back faster" stood out. "Reducing administrative time directly translated into educational quality," Kuwabara wrote.
The framing of training fees also shifted. "Training fees are no longer treated as a one-time expense but as part of an investment in operational improvement. Future training will be planned with the latter four stages built in," Kuwabara wrote.
The final line of the report read: "The reason training heat faded was not staff motivation but a missing structure. The moment we designed the four stages after Learn, the heat reached the work. AI adoption succeeds or fails not on AI but on the structure connecting AI to the work."
On the morning teachers got their time with students back, AI had become invisible, the report said.
"Training is a beginning. Not an end. To carry the heat of Learn into the work, you need Organize, Generate, Implement, and Control. LOGIC asks for stage design that embeds learning into the organization. AI adoption looks like a technology story but is a structure story. Prepare an in-house environment, pick priority tasks, start small, replicate success—skip this order and training fees float. The day teachers got their time with students back, what changed wasn't technology but the design that connects technology to work."
Related Files
Tools Used
- ROI Polygraph — Visualizing teacher administrative time, skill gaps, and governance risk
- ROI Proposal Generator — Payback simulation for in-house AI platform deployment
