ROI Case File No.516: Employees Who Used It, and Employees Who Didn't
Employees who used it, and employees who didn't
Chapter 1: Licenses Were Distributed, But Not Used
"A year has passed since we distributed Copilot licenses to all employees. Employees who use it and employees who don't have completely split."
Ryohei Kasai, Corporate Planning Manager at Adventis, said this as he showed usage log reports. A full-service advertising agency. 220 employees, 40 departments. Copilot Chat bundled with Microsoft 365 E3 had been distributed across the company. "Heavy users 20%, light users 40%, non-users 40%. The three-way split is set."
"What are the reasons for non-use?" Claude asked.
"They're compound," Kasai replied. "'Don't know what to use it for,' 'don't know how to write prompts,' 'don't feel a need in my work,' 'can't trust the quality.' Not a single reason—each employee is stuck for a different one. We ran broad DX training, but the effect was limited."
"What's the current promotion structure?" I asked.
"We're setting up a working group," Kasai answered. "A plan to build a Copilot utilization community internally. But we don't know where to start. The comprehensive strategy proposals from external firms feel off from our situation. We want hands-on field support."
"A pattern of trying to move the whole thing at once and stalling," I responded. "Let's break it into components with BOM."
Chapter 2: BOM Asks for Decomposition into Measurable Units
"This case calls for BOM."
Claude wrote "B·O·M" on the whiteboard.
"BOM, Bill of Materials, originally refers to the parts list of a product in manufacturing," I explained. "Apply this thinking to DX promotion and the design becomes: decompose the promotion target into component units and set numerical targets for each. Convert the vague 'company-wide utilization promotion' into measurable components. That's the starting point of hands-on field support."
"Let's measure current costs first," Gemini said, opening ROI Polygraph. Kasai's data went in.
"Monthly opportunity-loss costs are out," Gemini read. "Opportunity loss from non-user license fees: 1.44 million yen/month—40% of license fees that aren't contributing to operations. Opportunity loss from unutilized AI-shortened work time: 4.8 million yen/month—the labor that could have been saved on document creation, research, summarization, and the like. Lost deals from proposal-quality variance: 1.2 million yen/month—the gap between heavy-user teams and non-user teams. Insufficient internal knowledge-sharing cost: 900,000 yen/month. Resource waste from inadequate effect measurement of training and promotion activities: 500,000 yen/month. Total: 8.84 million yen/month. Annualized: roughly 106.1 million yen."
Kasai looked at the numbers. "I'd thought it was just non-user license fees. The work-time opportunity loss is overwhelmingly larger."
"Now let's design with BOM," I continued.
[BOM decomposition — Splitting promotion into components]
"First, decompose AI utilization promotion into components," Claude said. "Component 1: Utilization rate (who is using it). Component 2: Utilization depth (how much they use it). Component 3: Utilization domain (what work it's used for). Component 4: Quality (the operational effectiveness of outputs). Component 5: Knowledge sharing (the in-house circulation of success cases). Each of the five components becomes an independently measurable indicator."
[Current value measurement — Quantify each component]
"Next, measure current values by component," Gemini continued. "Utilization rate is 60% weekly active. Utilization depth is 12 uses per month on average. Utilization domain is skewed toward document creation. Quality is 40%—the ratio of outputs usable in work as-is. Knowledge sharing is zero baseline. Lined up as five numbers, where to start becomes clear."
[Target setting — Draw the one-year destination in components]
"Next, one-year targets," I continued. "Utilization rate from 60% to 85% weekly active. Utilization depth from 12 to 30 uses per month. Utilization domain expanded from document-creation focus to five domains. Quality from 40% to 70% operational use. Knowledge sharing: an in-house case library of 100+ entries. Each component gets an independent target."
[Component-mapped initiatives]
"Map initiatives to components," Claude continued. "Department-by-department kickoff sessions for utilization rate. Use-case prompt template distribution for depth. Domain-specific pilot projects for domain expansion. Output review mechanisms for quality. An in-house case posting platform for knowledge sharing. Which component each initiative affects is explicit, so effect measurement can also proceed per component."
[Estimating investment recovery]
"Let's run ROI Proposal Generator," Gemini proposed.
- Initial cost: Hands-on promotion program, department-by-department kickoffs, prompt template development, knowledge-sharing platform, effect measurement foundation, and community operation design: 8.2 million yen total
- Monthly cost: Ongoing promotion program operation: 400,000 yen/month
- Monthly savings: Conversion of unused license fees to operational contribution = 1.2 million yen/month, expanded work-time utilization = 3.2 million yen/month, revenue contribution from proposal quality improvement = 800,000 yen/month, operational efficiency from knowledge sharing = 600,000 yen/month. Total: 5.8 million yen/month
- Net monthly savings: 5.8 million yen − 400,000 yen = 5.4 million yen/month
- Payback period: 8.2 million yen ÷ 5.4 million yen = approximately 1.5 months
"Under two months," Gemini summarized. "What matters is the structure that lets us track progress per component during promotion. Components that aren't producing effect can have their initiatives revised early. Comprehensive strategy proposals can't deliver effect measurement at this granularity."
Kasai checked the numbers. "We'd stalled trying to do company-wide rollout in one go. Decomposed into components, initiatives and effects pair up one-to-one."
"BOM is a tool that carves complex promotion into measurable units," I responded.
Chapter 3: A Rollout Plan That Moves Per Component
"Let's lay out the path," I said at the whiteboard.
"Month 1: Of 40 departments, select 5 pilot departments; precise measurement of current values per component. Month 2: Develop prompt templates; conduct department kickoffs. Month 3: Launch the knowledge-sharing platform; accumulate initial cases. Months 4–5: Measure effects in the 5 departments; check progress per component. Month 6: Extract success factors; design rollout to the remaining 35 departments. Months 7–12: Phased company-wide rollout; track against per-component targets."
"You're not deploying company-wide from the start?" Kasai asked.
"Company-wide rollout in one shot is exactly why past DX training had limited effect," Claude responded. "Establish per-component improvement patterns in 5 pilot departments first, then expand laterally. Rolling out with quantifiable initiative effectiveness gets more buy-in from the field."
Kasai took notes. "Hearing BOM applied outside manufacturing for the first time. I hadn't imagined it applied to promotion projects."
Chapter 4: The Day the Components Started Moving
Eleven months later, Kasai's report arrived.
Weekly active utilization, at the six-month mark of promotion, rose from 60% to 78%. Monthly average use rose from 12 to 26. "We reached 90% of target. Tracking per component, exactly where it's growing and where it's stalling is visible," Kasai wrote.
Particularly strong results came from the knowledge-sharing platform. In-house case postings totaled over 120, popular cases attracted likes and comments, and a community formed on its own. "We used to be in a 'no idea who's using it' state. Now heavy users are identified, and those people naturally became internal influencers," the report noted.
Variance in proposal quality also narrowed. With prompt templates in place, even beginners could produce outputs near veteran quality. "Situations where new staff submitted veteran-class proposal drafts started appearing. The learning curve shortened," Kasai wrote.
Utilization domain also diversified. AI utilization, initially focused on document creation, expanded to five domains: research, summarization, planning brainstorms, copywriting, and data analysis. "Even with the same Copilot, usage differs by domain. The effect of placing domain-specific pilots was large," the report noted.
The most surprising change appeared in the non-user segment. Initially 40%, non-users had dropped to 15% by month eleven. "For non-users stuck on 'don't know what to use it for,' internal cases worked best. Seeing how the colleague next to them used it got them to try," Kasai wrote.
The working group's operation also changed. The promotion staff had expected solo effort, but heavy users voluntarily joined operations, and it became a field-led community. "The promotion person's role shifted from leader to facilitator," the report said.
A side effect: the relationship with the external DX firm was also rearranged. The contract with the firm offering comprehensive strategy proposals was passed up, and the form shifted to individual contracts with multiple vendors specializing per component. "When we know what to measure, the support we need can be ordered specifically," Kasai wrote.
At the end of the report, Kasai wrote: "We'd stalled on the vague target 'promote company-wide.' The moment BOM decomposed it into components, initiative, effect, and target paired one-to-one. The resolution of ambiguity was a problem of structural resolution."
The three-way split that had set into the organization started moving per component, and on the desks of the promotion staff, monthly reports had become something to look forward to, he wrote.
"Few words are harder to handle than 'company-wide promotion.' As long as what gets measured when by whom and how is undefined, initiatives become discussions of grasping at clouds. BOM asks for decomposition into component units. Just as a manufacturing parts list converts complex products into measurable units, decomposing the promotion target into components pairs targets, initiatives, and effect measurement one-to-one. In an organization where the three-way split had set in, on the day numbers per component lined up, what the promotion staff were moving wasn't employees—it was the very measurable structure."
Related Files
Tools Used
- ROI Polygraph — Visualizing unused license fees, work-time opportunity loss, and quality variance
- ROI Proposal Generator — Investment recovery simulation for component-based DX promotion programs
