ROI Case File No.311 | 'NeoFab's Invisible Judgment'

Chapter 1: The Fortress of Expertise—A Factory Where Judgment Concentrates in One Person

The week after AutoForge's Double Diamond case was resolved, a consultation arrived from Ibaraki regarding AI implementation at a precision machinery parts manufacturer. Opening Volume 26, "The Pursuit of Reproducibility," Case File 311 tells the story of transforming veteran judgment into a system anyone can replicate.

"Detective, our manufacturing lines depend on a single veteran. Without him, responses to defects are delayed. Young workers cannot make judgments and must stop the line to call the veteran. During that time, production halts. We've reached our limit."

Koji Tamura, Production Manager of NeoFab Manufacturing Co., born in Tsukuba, visited 221B Baker Street with exhaustion written across his face. In his hands were the veteran employee's work schedule and, in stark contrast, production stoppage records marked "judgment delay."

"We manufacture precision machinery parts in Ibaraki. Parts for automobiles, aircraft, and industrial machinery. We have 3 production lines and 45 workers. However, responses to defects rely entirely on one veteran, Kimura."

NeoFab's On-Site Judgment Stagnation: - Established: 1992 (precision machinery parts manufacturing) - Annual Revenue: 2.8 billion yen - Employees: 52 (45 in manufacturing) - Production Lines: 3 lines (24-hour operation) - Veteran Decision-Maker: Kimura (58 years old, 32 years of service) - Initial Response Time for Defects: Average 18 minutes until Kimura arrives - Monthly Production Stoppage Time: Cumulative 42 hours (waiting for judgment) - Problem: Young workers cannot make judgments when Kimura is absent, causing production stoppages

Tamura's voice carried deep urgency.

"Kimura retires next year. We want to pass his judgment ability to young workers. But his judgment is 'experience and intuition.' It cannot be verbalized. We tried to create a manual, but all he could say was 'it depends on the case.'"

Typical Response to Defect Occurrence:

2:00 AM, Line A: Sensor alarm: "Product dimensions out of specification"

Young Worker (24 years old): "This is bad, a defect occurred. But I don't know how to respond... I need to call Kimura-san."

Phone call: "Kimura-san, there's a dimensional anomaly on Line A. Please come."

Kimura (20 minutes by car from home): "Got it. I'll be right there. Stop the line until then."

20 minutes later, Kimura arrives: Kimura picks up the product and visually inspects it. He also checks the mold temperature, material lot number, and previous process records.

"Ah, this is mold wear. C-pattern wear. Replacement time is approaching. For now, if we lower the mold temperature by 3 degrees, it should last another 2 hours. Prepare for replacement during that time."

Setting adjustment complete, line restart: Stoppage time: 25 minutes

The young worker was stunned.

"How did Kimura-san make that judgment?"

"Experience. There are 3 types of mold wear patterns. You can tell from the sound, product surface, and dimensional output. After 30 years, you just know by looking."

"Can't we put it in a manual?"

"It's difficult. There are too many cases."

Chapter 2: The Prison of Experience—Knowledge That Cannot Be Reproduced Is Not an Asset

"Tamura-san, what approaches have you taken to reproduce Kimura-san's judgment?"

To my question, Tamura answered.

"We attempted to create manuals. However, Kimura's judgment is too complex to document. When we try to write 'In this case do A, in that case do B,' there are hundreds of 'cases.' In the end, it becomes 'Ask Kimura.'"

Current Approach (Person-Dependent Type): - Response: Depends on Kimura's judgment - Succession: Attempted manualization but failed - Problem: Young workers cannot make judgments when Kimura is absent

I explained the importance of systematizing immediate response judgment.

"Kimura-san's experience is valuable. However, as long as it remains closed within one person, it does not become an organizational asset. OODA—Observe, Orient, Decide, Act. If we support this loop with AI, anyone can reproduce Kimura-san's judgment."

⬜️ ChatGPT | Catalyst of Conception

"Don't lock up experience. Transform veteran judgment into everyone's weapon with OODA."

🟧 Claude | Alchemist of Narrative

"AI doesn't replace people. It transforms 30 years of experience into a form usable in 3 minutes."

🟦 Gemini | Compass of Reason

"OODA is the technology of rapid response. Accelerate Observe-Orient-Decide-Act and return decision-making authority to the field."

The three members began their analysis. Gemini unfolded the "OODA × AI Framework" on the whiteboard.

OODA Loop AI Support Design: 1. Observe: Real-time data collection via sensors and cameras 2. Orient: AI instantly classifies anomaly patterns 3. Decide: On-site workers make judgments based on AI suggestions 4. Act: Execute responses and reflect results in learning

"Tamura-san, let's rebuild NeoFab's on-site judgment with OODA and AI."

Chapter 3: The Eye of Observation—Converting 30 Years of Experience into Data

Phase 1: Observe - Construction of Data Collection Infrastructure (6 weeks)

First, we thoroughly observed and digitized Kimura's judgment process.

Information Used in Kimura's Judgment: - Product dimensions (micrometer measurement) - Mold temperature (temperature sensor) - Material lot number (records) - Previous process machining time (records) - Product surface condition (visual inspection) - Machine sound (auditory)

Automatic Collection via Sensors and Cameras: - Product dimensions: Continuous measurement with non-contact sensors - Mold temperature: IoT temperature sensors (every 10 seconds) - Surface condition: High-resolution cameras (photographing each item) - Machine sound: Acoustic sensors (frequency analysis) - Investment: 12 million yen for sensor and camera installation

After 6 weeks: Installation of 48 sensors and cameras across 3 lines completed

Phase 2: Orient - AI Model Construction (3 months)

Next, we trained the system on Kimura's past judgment data.

Learning Data Collection (Past 2 years): - Defect occurrence records: 820 cases - Data linked to each record: - Sensor values (dimensions, temperature, sound) - Images (product surface) - Kimura's judgment content (response methods) - Post-response results (success/failure)

AI Model Design: - Classify defect patterns (15 patterns) - Mold wear (C pattern) - Material defects (hardness anomaly) - Temperature anomaly (cooling insufficiency) - ... etc. - Present recommended response methods for each pattern - Display confidence scores ("80% probability of C pattern wear")

After 3 months: AI model completed. In testing, accurately classified 90% of past data

Phase 3: Decide - Authority Transfer to On-Site Workers (2 months)

We introduced the AI model to the field and built a system where young workers could make judgments.

New Defect Response Flow:

2:00 AM, Line A: Sensor alarm: "Product dimensions out of specification"

Young Worker: Opens tablet. AI immediately displays analysis results.

【AI Judgment Result】
Defect Pattern: Mold wear (C pattern)
Confidence: 85%
Recommended Response:
1. Lower mold temperature by 3 degrees
2. Prepare for mold replacement in 2 hours
3. Return temperature setting to normal after replacement
Past Success Rate: 92%

Young Worker: "I see, I should lower the mold temperature. Let me try it."

Setting adjustment complete, line restart: Stoppage time: 5 minutes (80% reduction from previous 25 minutes)

No need to contact Kimura

Phase 4: Act - Construction of Learning Loop (Ongoing)

We had the AI learn response results to continuously improve judgment accuracy.

Learning Mechanism: - Workers follow AI suggestions and respond - Record response results (success/failure) - AI learns results and reflects them in next judgment - Update model weekly

After 3 months: - AI judgment accuracy: 90% → 94% - Kimura's judgment accuracy: 96% (approached within just 2% difference)

Chapter 4: The Outcome of Liberation—From Teacher to Supervisor

Results After 6 Months:

Dramatic Reduction in Production Stoppage Time: - Monthly production stoppage time: 42 hours → 8 hours (81% reduction) - Response time per case: Average 18 minutes → Average 4 minutes (78% reduction)

Improvement in Young Workers' Judgment Ability: - With AI support, young workers can make independent judgments in 85% of cases - Dependence on Kimura: 100% → 15% (complex cases only)

Change in Kimura's Role:

Before: "Without me, the factory doesn't run. I get called in the middle of the night. I'm exhausted."

After: "Young workers can now make their own judgments using AI. I only need to support complex cases. And while watching AI's judgments, I sometimes think 'I see, there was that pattern too.' AI has become my teacher."

Organizational Change: - Judgment know-how remains in the organization even after Kimura's retirement - Accelerated growth speed of young workers (can gain experience with AI support) - Stable 24-hour operation system (anyone can respond)

Business Results After 12 Months:

Productivity Improvement: - Operation rate: 82% → 93% (effect of reduced production stoppage time) - Annual production volume: +11%

Financial Impact: - Investment: 12 million yen (sensors, AI development) - Revenue increase from production increase: 300 million yen annually - Investment recovery period: 4.8 months

Customer Evaluation: "NeoFab's on-time delivery rate has improved. Previously there were delays due to defect responses, but now there are almost none."

Chapter 5: The Detective's Diagnosis—Opening of Volume 26 "The Pursuit of Reproducibility"

That night, I contemplated the essence of OODA and AI.

NeoFab depended on a single veteran. Kimura's 30 years of experience were valuable, but as long as they remained closed within one person, they were not an organizational asset.

However, by supporting the OODA loop with AI, we transformed Kimura's judgment into a form anyone could use. Sensors perform observation, AI presents judgment, and young workers make decisions.

"AI doesn't replace people. It democratizes experience. It transforms veteran expertise into a weapon everyone can wield."

Volume 26, "The Pursuit of Reproducibility," opens here.

The next case will also depict the moment when OODA generates field responsiveness.

"Don't lock experience within one person. With OODA and AI, transform veteran judgment into everyone's weapon. Responsiveness means everyone can make decisions"—From the Detective's Notes

Related Files