According to the 2024 White Paper on Monozukuri (Manufacturing), jointly issued by the Ministry of Economy, Trade and Industry, the Ministry of Health, Labour and Welfare, and the Ministry of Education, Culture, Sports, Science and Technology, the number of young workers aged 34 and under in manufacturing fell from 3.84 million in 2002 to 2.59 million in 2023. Meanwhile, the baby-boomer generation has already passed the age of 75, and the mass retirement of veterans has become a reality.

The problem of skills transfer is not unique to manufacturing. Across every industry—construction, healthcare, the professions, sales—work that has been sustained by the experience and intuition of veterans is becoming difficult to maintain due to a shortage of successors.

Against this backdrop, skills transfer using AI is drawing attention. Efforts are beginning to spread to digitize the tacit knowledge of veterans—knowledge that conventional OJT and manuals could not fully convey—using generative AI and RAG, and to accumulate it as an organizational asset.

This article first organizes the background to why skills transfer has become difficult, then explains, across industries, four patterns for transferring knowledge using AI.

Why Skills Transfer Has Become More Difficult

The Structural Problem of "No Room to Teach"

The single biggest reason skills transfer does not progress is a simple structural problem: "the veterans who should be teaching are too busy." Because of labor shortages, veterans themselves keep standing on the front lines of operations, so they cannot secure time to mentor their successors.

In the Ministry of Health, Labour and Welfare's "FY2024 Basic Survey on Human Resources Development," among establishments that reported some kind of problem with human resource development (79.9% of all establishments), 59.5% cited "a shortage of personnel to provide guidance" and 47.4% cited "no time to carry out human resource development."

The Wall of Tacit Knowledge

Another reason skills transfer is difficult is that much of the knowledge to be passed on exists as tacit knowledge. Tacit knowledge refers to knowledge based on experience and intuition that is difficult even for the person themselves to put into words.

"You can tell the quality of this material by touch," "watch out when you hear this sound," "with this customer, set the first proposal a little high"—such knowledge is hard to write into a manual, and the reality is that it cannot be conveyed without watching closely alongside the work through OJT.

Tacit knowledge comes in three types—procedural, judgment-based, and values-based—each calling for a different transfer method (for details, see our article on How to Pass On Tacit Knowledge with AI).

The Limits of the "Watch and Learn" Culture

Japanese skills transfer has traditionally centered on a "watch and learn" approach over long periods of OJT. This approach was effective in environments with ample time and personnel, but it has stopped functioning amid labor shortages and rising turnover.

According to the 2024 White Paper on Monozukuri, the share of manufacturing companies using digital technology rose from just under 50% in 2019 to over 80% in 2023. While the digitalization of the shop floor is advancing, the digitalization of skills transfer still lags behind.

Skills Transfer Challenges by Industry

The challenges of skills transfer differ in nature by industry. You need to grasp where your own challenges lie before choosing the right countermeasures.

Manufacturing

The knowledge that is hardest to pass on centers on equipment-adjustment know-how, the knack for quality judgments, and rules of thumb for anomaly detection. Many judgments are based on the five senses (sound, vibration, color, touch), making this the most difficult area to convert into text. In recent years, digitalization through sensor data and motion capture has advanced; the 2024 White Paper on Monozukuri introduces a case in which Konno Seisakusho, a company with 39 employees, adopted motion capture in welding training to visualize the movements of skilled technicians.

Construction and Civil Engineering

Because conditions differ at every site, this is an industry with much situation-based tacit knowledge, such as "for this ground, respond this way" or "in this weather, switch to this construction method." Knowledge is scattered across site photos and daily work reports, and cases where it has not been systematically organized stand out.

The Professions and Specialized Services

In areas such as legal affairs, tax, and consulting, the core is judgment-based tacit knowledge: "how to apply this provision to this case." Records of how past cases were handled are often buried in individuals' files and email, and there is a lack of mechanisms to accumulate and leverage them as an organization.

Sales and Customer Service

Rules of thumb such as "contact this customer at this timing" or "for this industry, this proposal angle resonates" tend to exist only in the heads of top salespeople. Separate from the quantitative data recorded in CRM, the transfer of qualitative sales know-how is a challenge.

Human Resources

The HR function, too, holds a great deal of veteran tacit knowledge—sizing up candidates in hiring interviews, how to run employee training and onboarding, and how to engage with subordinates in evaluation meetings and management. Know-how such as "for a candidate with this background, probe this point in the interview" or "communicate it this way to an employee at this stage" shapes the quality of recruiting, employee training, and management, yet it tends to remain the rule of thumb of an individual. The challenge is to document and turn into data the hiring criteria, interview records, and training know-how so they can be passed on as organizational assets.

Healthcare and Nursing Care

This is an area where experience-based knowledge—responding to patients, reading symptoms, judgment during sudden changes—is directly tied to quality. While healthcare has an evidence-based body of knowledge, the transfer of "clinical intuition not found in textbooks" still depends on individuals.

Four Patterns of Skills Transfer Using AI

Skills transfer with AI divides into four patterns according to the type of tacit knowledge and the characteristics of the work.

Pattern 1: Document-Search AI (for Procedural Tacit Knowledge)

This is a method of feeding existing manuals, procedure documents, work records, and response histories into a RAG system so that employees can search them in natural language. When asked "What is the procedure for changing the XX setting?" or "How did we handle it the last time XX trouble occurred?", the AI searches the relevant documents and answers with sources.

It has the lowest barrier to adoption, and for organizations whose existing documents are reasonably well maintained, results can be obtained in a relatively short period.

Pattern 2: Veteran-Interview AI (for Judgment-Based Tacit Knowledge)

This is a method of verbalizing decision criteria through interviews with veterans and building them into training data for the AI. The difference from Pattern 1 is that it targets "knowledge not yet documented" rather than existing documents.

The tacit-knowledge transfer initiative that Lion and NTT Data launched in June 2024 is a representative example of this pattern. In the manufacturing process for powdered laundry detergent, they extracted the undocumented tacit knowledge of experts, documented it as a "collection of key points," and incorporated it into a generative-AI-powered "knowledge transfer AI system" (Source: NTT Data press release, June 3, 2024).

Specifically, veterans are asked questions such as "How do you make a decision in this situation?" and "Why do you choose that method?", and the content is converted into text and built into RAG. The concrete techniques for knowledge-extraction interviews are explained in detail in our article on How to Pass On Tacit Knowledge with AI.

This pattern takes effort to extract knowledge, but it is a high-impact approach in that it lets you preserve, as an organizational asset, the "decision criteria that are not written in any manual yet are indispensable to the work."

Pattern 3: Motion-Analysis AI (for Physical Skills)

This pattern is effective in industries where skills based on physical movement are important, such as manufacturing and construction. Cameras and sensors record the movements of skilled workers, and the AI analyzes and quantifies the movement patterns.

In the Konno Seisakusho case introduced in the 2024 White Paper on Monozukuri, motion capture is used to visualize the welding movements of skilled technicians, helping to strengthen the skills of younger workers and transfer expertise. By quantifying the angle, speed, and pressure of movements, it becomes possible to objectively show the subtle differences that "watch and learn" could not convey.

That said, this pattern requires installing sensors and cameras, so the initial investment is larger than for the other patterns. Specialized services that combine data analysis with consulting to support the conversion of tacit knowledge into explicit knowledge—such as Mitsubishi Research Institute's "Takumi AI"—also exist, but adoption cases are limited at this point.

Pattern 4: Personal AI (for Transferring Values and a Philosophy of Judgment)

This is a method of training AI on the statements, principles of conduct, and the values behind the decisions of a veteran or executive, so they can be referenced through dialogue. When you ask "How would my senior think about this project?", you get an answer based on that person's thought patterns.

Rather than the transfer of the technology itself, this pattern is suited to conveying values such as "why that technology is valued" and "what stance to take toward quality." When used to pass on an executive's philosophy, it is sometimes built as a "CEO bot" (for details, see our article on What Is a CEO Bot?).

Five Steps to Adopting Skills-Transfer AI

Step 1: Take Stock of the Skills to Be Transferred

First, identify which skills and know-how within the organization have become dependent on specific individuals. Simply posing the simple question "Which task would cause the most trouble if this person left?" to managers in each department already brings priorities into view. Mapping along two axes—the level of risk (how irreplaceable it is) and frequency (how often it is used)—clarifies the areas to tackle first. The thinking behind this kind of stocktaking is also explained in detail in our article on How to Eliminate Key-Person Dependency.

Step 2: Classify the Type of Tacit Knowledge

For each skill you have taken stock of, classify whether it falls under the procedural, judgment-based, or values-based type. If you are unsure how to classify it, ask "Can this be written into a document?" Can be written → procedural; can be written with conditions → judgment-based; cannot be written → values-based; that is the rough guideline.

Step 3: Choose the Right AI Pattern

According to the type of tacit knowledge and the organization's current situation (the volume of existing documents, IT environment, budget), choose the best of the four patterns above. In many cases, starting with Pattern 1 (document-search AI) is realistic. When there are few existing documents, start with Pattern 2 (veteran-interview AI), advancing knowledge extraction and documentation at the same time.

Step 4: Start Small

Rather than aiming for a company-wide rollout from the start, run a PoC limited to a specific department or task. The key is to choose an area that is "high in key-person dependency risk" and where "veteran cooperation is easy to obtain." Because AI adoption cannot advance without the cooperation of the veterans themselves, it is important to carefully share the purpose: "We want to make your knowledge an asset of the company."

Step 5: Operation and Continuous Improvement

It does not end with building the AI system; you need an operational structure that continuously adds and updates data. When there are new technologies or process changes, update the AI's training data accordingly. Ideally, set up a regular review every six months to a year and build in a mechanism to check the AI's answer accuracy and usage.

Cases AI Alone Cannot Solve

Not all skills-transfer challenges can be solved with AI. In the following cases, there are things you should do before adopting AI.

When the OJT structure is not yet in place. If a training mechanism (assigning people in charge, an education plan, evaluation criteria) does not exist in the first place, you won't be able to make good use of AI even if you adopt it. You should first establish a basic OJT framework and then position AI as a means of making it more efficient.

When the skills to be transferred are not clear. At the stage of "not knowing what should be transferred," adopting AI is premature. The priority is first to spend time on the stocktaking in Step 1 and clarify priorities.

When the veterans themselves are uncooperative. Skills transfer with AI is premised on interviews with veterans and the provision of data. If the person feels "reluctant to share my know-how" or "worried that AI will take my job," forcing it forward results in an effort that is only superficial. You need to prepare a mechanism in which preserving the veteran's knowledge for the organization leads to their own evaluation, and then ask for their cooperation.

Summary

The problem of skills transfer is a management challenge common to every industry amid advancing labor shortages and an aging population. In an era that conventional OJT and manuals can no longer fully address, AI becomes a powerful means of converting tacit knowledge into explicit knowledge and accumulating it as an organizational asset.

The pattern of AI use differs by the type of tacit knowledge. Document-search AI for procedural knowledge, veteran-interview AI for judgment-based knowledge, motion-analysis AI for physical skills, and personal AI for values—using the right one for the challenge is important.

Start by identifying "which skill would cause the most trouble if this person left," and classifying the type of tacit knowledge. In many cases, Pattern 1, which makes existing documents searchable, is the most accessible first step.

At Teraverse, we provide generative AI solution development tailored to skills-transfer challenges, from search AI for internal knowledge to chatbots that pass on veterans' decision criteria. If you are interested in using AI for skills transfer, please feel free to contact us.