ROI Case File No.542: The Too-Small Chores No One Looked at One by One
The Too-Small Chores No One Looked at One by One
Chapter 1: Looked at Individually, Every Chore Was Small
"We want to make our operations more efficient with AI. But every service we've heard of covers only a part, and the cost-effectiveness never added up."
Hiroaki Kaji, CEO of TechnoCraft, said this as his expression clouded over. A manufacturer of about fifty employees. "Labor shortages are serious. Orders arrive handwritten or by fax, and we process them by hand. The order forms don't carry all the necessary information, so we pull from other documents to complete the order record. Converting drawings into NC programs is manual too."
"Is each piece of work large?" Claude asked.
"That's the thing—looked at individually, they're small," Kaji answered. "There are mountains of fine chores. One by one, you think 'this much can be done by hand.' But bundle them, and it becomes a considerable volume of work. Because each is small, we couldn't make the decision to introduce anything, and it's been left alone."
"How do you measure cost-effectiveness?" I confirmed.
"I want to compare it against labor cost," Kaji answered. "Whether AI exceeds the ¥200,000-to-¥500,000-per-month labor cost. And I'm uneasy about data being used for learning, and about security. We can't use it unless safety is guaranteed."
"Then we need to bundle the chores that look small individually, measure them, and narrow the solution down twice," I responded. "Let's break this down with the double diamond."
Chapter 2: What DOUBLE_DIAMOND Asks—Passing Through Divergence and Convergence Twice
"This case needs DOUBLE_DIAMOND."
Claude drew two diamonds side by side on the whiteboard.
"DOUBLE_DIAMOND is a framework that 'widens then narrows' both the problem and the solution, across four stages: discover, define, develop, and deliver," I explained. "In the first diamond you broadly surface the problems and define them into one; in the second you broadly generate solutions and narrow to one. Try to crush fine chores individually and the cost-effectiveness never adds up. It's a tool that first diverges everything and bundles it, then converges on the single point that works."
"First, let's measure the current cost," Gemini said, opening the ROI Polygraph. He entered the data Kaji had provided.
"The monthly cost is in," Gemini read out. "Manual processing labor for handwritten and fax orders: 190 hours per month on average, at ¥3,800 per hour, ¥722,000 per month. Labor to supplement missing order-form information from other documents: ¥480,000 per month on average. Manual conversion labor from drawings to NC programs: ¥550,000 per month on average. Buried cost from the bundle of fine, individually small tasks: ¥330,000 per month on average. Opportunity loss from stalled adoption due to data-learning and security concerns: ¥400,000 per month on average. A total of ¥2,482,000 per month. Annualized, about ¥29,780,000."
Kaji stared at the figures. "Chores I thought were small one by one come to over ¥29 million when bundled. Was it because I looked at them individually that I left them alone all this time?"
"Now, let's design with DOUBLE_DIAMOND," I continued.
[Discover—Widen Out Every Fine Chore]
"First, we map the entire workflow and diverge the chores," Claude said. "Order-taking and placing, information supplementation, NC conversion, document processing—we line up the individually small chores without missing a single one. Only when we widen them out does the weight of the bundle become visible."
[Define—Narrow to the Point That Works]
"Next, we converge the widened chores into a single problem," Gemini continued. "The heaviest in labor and the easiest for AI to replace are order processing and NC conversion. We define these as 'the problem to solve first.' Because we narrow after widening, we don't miss."
[Develop—Widen the Solutions and Prototype]
"For the defined problem, we diverge solutions and prototype," I continued. "AI scanning and data extraction of order documents, automatic NC-program generation from existing data, closed data management. We build multiple prototypes and verify the security requirements at the same time."
[Deliver—Converge on One Solution and Implement]
"Finally, we narrow the prototypes to one and implement," Claude continued. "We measure effect and safety in a field test, and keep only the solution whose cost-effectiveness exceeds labor cost. It's a structure that fully narrows the second diamond and hands it to implementation."
[Calculating the Investment Recovery]
"Let's run the estimate with the ROI Proposal Generator," Gemini proposed.
- Initial cost: AI scanning infrastructure build, NC-program auto-generation algorithm development, closed data-management system, field testing, and training—¥5,800,000 total
- Monthly cost: System operation and model-update ongoing fees combined, ¥240,000 per month
- Monthly reduction effect: Order-processing labor reduction = ¥580,000 per month (assuming 80% reduction), information-supplementation labor reduction = ¥380,000 per month, NC-conversion automation = ¥440,000 per month, efficiency from bundling fine tasks = ¥250,000 per month, totaling ¥1,650,000 per month
- Net monthly reduction: ¥1,650,000 − ¥240,000 = ¥1,410,000 per month
- Payback period: ¥5,800,000 ÷ ¥1,410,000 = approximately 4.1 months
"Recovery in just over four months," Gemini summarized. "What works is implementing from the heavy single point made by bundling the fine chores. Automate them one at a time and the cost-effectiveness never adds up. Because we diverge and bundle, then converge on the solution that exceeds labor cost, the investment doesn't miss."
Kaji confirmed the figures. "I was looking at them one by one, thinking 'small, so by hand is fine.' Widen and bundle them, and where to start gets decided."
"DOUBLE_DIAMOND is a tool that bundles small chores and narrows to the point that works," I responded.
Chapter 3: A Deployment Plan That Implements From the Point That Works
"Let me organize the approach," I said, standing at the whiteboard.
"Month one—mapping the entire workflow, diverging the fine chores. Month two—defining the problem, converging on order-taking and NC conversion. Months three and four—prototyping AI scanning and NC auto-generation, verifying security requirements. Month five—trial operation on the floor, measuring cost-effectiveness. Month six—convergence of the solution and full implementation. Month seven onward—expanding the automation scope to fine tasks, establishing operation in a closed environment."
"Is security really going to be all right?" Kaji confirmed.
"We design it in a closed environment," Claude responded. "If data-learning use worries you, we run it in an environment where data doesn't leave your premises. In the develop stage we prototype multiple solutions and converge only on those that satisfy safety. It's a structure that fully narrows both cost-effectiveness and safety in the second diamond."
Kaji said, taking notes, "Widen then narrow, passed through twice. I can see the sequence now."
Chapter 4: The Day the Bundled Chores Grew Light
Nine months later, a report arrived from Kaji.
Handwritten and fax order processing was reduced 80% from before after AI scanning was introduced. "Order forms we used to re-key by hand become data through scanning. The work of pulling information from other documents has become almost automatic too," Kaji wrote.
Manual NC-program conversion also dropped sharply. A mechanism that auto-generates from existing data erased the manufacturing floor's chore. "Work where we assembled one piece at a time while looking at a drawing became just auto-generation and confirmation. The labor for revisions also dropped," the report read.
The most surprising change appeared in how cost-effectiveness looked. An investment we'd given up on as 'not adding up' held up once bundled. "Looking at the fine tasks one by one, we could never have introduced anything. Bring it in from the heavy bundled point, and it clearly exceeded labor cost," Kaji wrote.
The security concern was resolved too. With the move to a closed environment, the worry of data leakage vanished. "Because it's designed so data doesn't leave our premises, we can use it with peace of mind. We stopped halting things on the grounds of safety," the report read.
As a secondary effect, the standard for adoption decisions changed. The mindset of measuring bundled rather than individual took root in the company. "We stopped 'this is small, so leave it for later.' We came to judge by whether it's heavy when bundled," Kaji wrote.
At the end of Kaji's report it said: "Fine chores stay neglected as long as you look at them one by one. The moment we widened everything out with the double diamond, bundled it, and narrowed to the point that works, the cost-effectiveness held up. Small chores too become objects of investment once they are bundled."
The day a company where no one had bundled and looked at the too-small chores became a company that could bundle and measure them, operational efficiency had changed from individual resignation into a design that passes through divergence and convergence twice, the report noted.
"'AI covers only a part and the cost-effectiveness doesn't add up'—it's a phrase heard again and again in manufacturing consultations. But the real problem lies in looking at fine chores one by one. Individually small, bundled heavy. What DOUBLE_DIAMOND asks is the flow of passing through divergence and convergence twice. Widen out all the chores and bundle them into one problem, then widen the solutions and narrow to one point. The day a company where no one bundled the too-small chores could bundle and measure them, what changed was not the AI tool but the very perspective that bundles small chores and narrows to the point that works."
Related Files
Tools Used
- ROI Polygraph — Visualizing order-processing labor, NC-conversion labor, and the bundled cost of fine tasks
- ROI Proposal Generator — Investment-recovery simulation for AI operational efficiency rooted in divergence and convergence
