Turning veteran technicians’ tacit knowledge into shared wisdom for the entire workplace.
A workplace transformation that reduced 850 veteran-dependent monthly inquiries by 62% with AI
Industry
Automotive parts manufacturing
Implementation Area
Manufacturing line operations
Including production engineering, quality assurance, and maintenance
Implementation Period
6-week PoC → 3-month pilot rollout at one site → 6-month rollout to all sites
BACKGROUND
Knowledge transfer had become a top management priority.
At Company A, approximately 35% of veteran technicians in their late 50s were expected to retire within the next five years, making knowledge transfer a top management priority. At the same time, on-site managers had serious concerns that if an incorrect AI response caused a major incident, they would be held responsible. As a result, full-scale adoption of generative AI had been postponed repeatedly. CLAVI Mining’s patented hallucination-prevention technology and manufacturing-focused design broke through this stalemate.
Challenges Before Implementation
At Company A’s manufacturing site, three issues were intertwined.
Troubleshooting depended on veteran workers
Whenever equipment abnormalities occurred, inquiries concentrated on a small number of skilled veteran workers. After those members had left work or during holidays, responses were delayed, resulting in an average of 28 emergency night calls per month. It had become normal for veteran workers who should have been resting to be called back.
Past records were scattered
Troubleshooting records were spread across paper reports, individual Excel files, and notes on employees’ personal PCs, requiring more than 30 minutes on average to search. As a result, the same root-cause issues repeatedly occurred several times a year, creating a recurring problem loop.
Questions from younger staff were concentrated
It took an average of 18 months for new employees to become fully independent. During that time, line leaders received more than 850 basic questions per month. Their time for management and improvement activities was chronically reduced, while improvement themes continued to accumulate.
The company had been considering generative AI for two years, but when it tested a general-purpose chat service, there were cases where the service confused procedures for similar equipment. A site supervisor strongly resisted adoption, saying, “I would be too afraid to use this for work instructions.”
Reasons for Selection
After comparing multiple candidates, Company A selected CLAVI Mining for the following four reasons.
First, patented hallucination-prevention technology (Patent No. 7691787).
Through multi-layer feedback control and a fact-checking policy engine, the system is designed to fundamentally prevent AI from “inventing” information that does not exist in internal knowledge. This was the deciding factor because it quantitatively reduced the risk that site managers feared most: operational accidents caused by AI falsehoods.
Second, product design specialized for manufacturing sites.
Developer Ryowa has 30 years of experience supporting manufacturing sites. Its ability to let AI learn directly from manufacturing-specific information assets, such as equipment manuals, troubleshooting records, and scanned paper PDFs, differentiated it from other services.
Third, transparency logs and audit support features.
For every response, the system visualizes which internal document and which passage served as the basis. This supports accountability for quality management standards such as IATF 16949 and internal audits.
Fourth, support for on-premises deployment.
Because the system would include design information, processing conditions, and supplier information, a cloud-based service could not pass the management meeting. CLAVI Mining could be deployed on-premises, enabling a short approval process with the information systems department.
Results After Implementation (After 6 Months)
Six months after the pilot rollout at one site, the following numerical improvements were confirmed.
Inquiry reduction
The number of questions to line leaders decreased from 850 to 320 per month, a 62% reduction. A leader commented, “I used to spend more than half of each day answering Q&A, but now I can secure two to three hours for improvement activities.”
Trouble response time
The average time from equipment abnormality to recovery was reduced from 45 minutes to 12 minutes, a 73% reduction. By instantly presenting similar past cases, the AI significantly reduced trial and error in identifying causes.
New employee training period
The time required for new employees to become independent was reduced from an average of 18 months to 9 months, a 50% reduction. A major factor was that new employees could ask AI basic questions they might hesitate to ask senior colleagues.
Night-shift emergency calls
Emergency calls to veteran workers during night shifts decreased from 28 to 6 per month, a 79% reduction. As a secondary effect, veteran workers could rest with peace of mind, and two veterans who had been considering leaving changed their minds.
Sense of knowledge transfer
In an anonymous survey, the share of veteran workers who answered, “I feel my know-how will remain in the organization,” rose from 23% before implementation to 89% after implementation. More employees scheduled to retire began saying they wanted to leave their know-how behind during their final year.
Comment from the Site Manager (Deputy Factory Manager)
Initial concerns turned into trust on the shop floor.
“We were afraid of incorrect AI responses and could not move forward with adoption, but the accuracy enabled by patented technology and the transparency of seeing the supporting internal documents on the spot gave us great reassurance.
Now, younger employees on the shop floor naturally say,‘Ask CLAVI and you’ll know.’
Next, we are considering horizontal rollout to other sites and use for supplier quality audit support.”
Site Manager (Deputy Factory Manager)
Factory Manager