AI Can Preserve the Future of SMEs
Small and medium-sized enterprises often carry knowledge that is hard to see from the outside. A family-owned food producer knows when fermentation is going wrong before the lab result arrives. A precision casting shop knows how a slight change in temperature will affect the final part. A leather workshop knows how material, humidity, and timing interact. These capabilities are rarely stored in a manual. They live in people, routines, judgment, and years of accumulated trial and error.
This creates a problem that many economies now face: what happens when the next generation is not there to learn?
The usual conversation about AI and business focuses on productivity. Can AI reduce costs? Can it automate tasks? Can it make workers faster? Those questions matter, while a deeper issue for SMEs deserves equal attention. For firms built around tacit knowledge, AI may matter because it can help preserve knowledge across time and move knowledge across industries.
That is the core idea behind the paper AI as an Intermediary for Cross-Industry and Cross-Time Knowledge Transfer in SMEs. The paper argues that AI should be understood as more than a tool for automation. For some SMEs, AI can become a knowledge infrastructure.
The Succession Problem Is a Knowledge Problem
Many traditional industries face succession gaps. Younger workers move into technology, finance, or other high-wage sectors. Entry-level roles shrink as firms reorganize around software and automation. Apprenticeship pipelines weaken. When this happens, the firm loses more than labor hours. It risks losing the tacit knowledge that makes it productive in the first place.
This matters because tacit knowledge is hard to write down. A manual can describe a process. It rarely captures the full judgment behind the process. It can say what to do under normal conditions, while often failing when the environment changes, when materials behave differently, or when a small anomaly needs interpretation.
In the paper’s model, this problem is represented as a stochastic succession process. A successor may arrive quickly, or the firm may wait for a long time. When someone does arrive, that person may have high or low human capital. This distinction matters. An AI system can preserve part of expert knowledge, and the preserved knowledge still needs someone capable of using it.
This leads to the first major idea: AI can act as a knowledge baton.
AI as a Knowledge Baton
Imagine a retiring expert who cannot fully teach a successor because the successor has not yet arrived. Without a preservation mechanism, the expert’s knowledge decays with time. Some of it remains in documents, machines, and organizational routines. Much of it fades.
AI changes this situation by creating what the paper calls a codified asset. This asset is produced when a firm invests in AI tools that observe, model, and preserve parts of expert behavior and decision-making. It may take the form of decision-support systems, workflow models, prompts, process simulations, or structured knowledge bases.
The baton metaphor is useful because the asset does not replace succession. It makes succession more durable. If a high-quality successor arrives later, the firm has something meaningful to pass on. If a lower-skill successor arrives first, the firm may preserve the asset while waiting for someone better suited to absorb it.
The key result is intuitive: the worse the succession gap, the more valuable this baton becomes. If successors arrive regularly, the firm can rely on direct apprenticeship. If successors arrive late, AI preservation becomes much more important.
Another result is equally important: AI codified assets and high-quality successors are complements. AI is more valuable when there is someone capable of absorbing and applying what it preserves. This challenges a simple replacement story. The value of AI for tacit knowledge comes from preserving knowledge during gaps and helping the right workers learn faster once they arrive.
AI as an Essence Extractor
The second mechanism is cross-industry transfer.
Traditional measures of industry relatedness often rely on labor flows. If many workers move from industry A to industry B, economists infer that the two industries use related skills. This is useful, while capturing only visible mobility. Some industries may share deep similarities even though workers rarely move between them.
For example, traditional cuisine and semiconductor processing may look unrelated if we measure relatedness through labor flows. Chefs rarely become semiconductor process engineers. Yet both domains may involve precise control, sensitivity to timing, material response, process windows, and anomaly detection. The surface skills differ. The underlying judgment patterns may overlap.
AI can help reveal these hidden similarities. By analyzing text, behavior, workflows, sensor records, and expert decisions, AI can project industry knowledge into a latent capability space. The paper calls this latent task essence.
Once knowledge is represented in this space, a firm can ask a new question: where else might our capabilities be useful?
This is why the paper describes AI as an essence extractor. AI can document what a firm currently does and identify the deeper capability structure behind that activity, making cross-industry redeployment more feasible.
Why Labor-Flow Data May Underestimate AI-Era Relatedness
This has an important implication for policy and research. If economists continue to measure relatedness only through labor flows, they may underestimate the real transferability of SME knowledge in the AI era.
Labor-flow data answer the question: where have workers historically moved?
AI-revealed relatedness answers a different question: where could the underlying capability structure be redeployed once it is codified?
Those questions can produce different rankings. A food craft SME may look close to restaurant management under labor-flow measures, yet AI may reveal that its deeper process-control capabilities are closer to functional foods, biomanufacturing, or precision production. A traditional workshop may appear isolated in conventional industry statistics while having high latent transferability once its tacit knowledge is codified.
This is one of the paper’s central policy insights: AI may expand the feasible industry set for SMEs.
The Unified View: One Investment, Two Uses
The most important part of the paper is the unified model. In real life, SMEs make one investment in codifying knowledge, and that investment can serve multiple purposes.
The same AI codified asset can help the firm preserve knowledge for the next generation and help the firm redeploy capabilities into a new industry. These two uses interact.
If a firm has low cross-industry transferability, AI mainly serves as a succession tool. It helps preserve knowledge across time. If a firm has high AI-revealed transferability, the same asset becomes more powerful. It can support succession and create new market options.
This is where the paper’s policy message becomes sharp. The strongest target for AI codification support is an SME with three features:
- It faces a serious succession gap.
- It still has some chance of attracting or developing high-quality successors.
- Its knowledge has high AI-revealed potential for cross-industry redeployment.
Such a firm is valuable because AI codification does two things at once. It protects knowledge that might otherwise disappear, and it creates new economic options beyond the original industry.
What This Means for Taiwan and Similar Economies
Taiwan is a natural setting for this idea. Many traditional industries are rich in tacit knowledge: precision machining, casting, food processing, leather, textiles, specialty materials, and craft manufacturing. At the same time, many of these sectors face labor shortages and succession challenges.
The usual policy response is to subsidize digitization or automation. That may help, while remaining too broad. The paper suggests a more targeted approach: subsidize AI codification where the knowledge is both at risk and redeployable.
Consider a precision casting workshop. Its current business may be small, while its expertise in materials, tolerances, process stability, and anomaly detection may be relevant to higher-value manufacturing services. If AI can help codify that expertise, the firm may preserve its knowledge for successors and build a path into new industrial applications.
Or consider a traditional food producer. Its knowledge of fermentation, temperature, timing, and sensory evaluation may have potential in functional foods or biotech-adjacent production. AI codification could make that hidden capability easier to preserve, teach, and redeploy.
This is a different way to think about industrial policy. The unit of analysis becomes the firm’s tacit knowledge and latent transferability, rather than the industry category alone.
Why Private Investment May Still Be Too Low
The paper also shows that firms with high cross-industry transferability may already have stronger private incentives to invest in AI. That may seem to weaken the case for subsidies. If these firms already want to invest, why should policy intervene?
The answer is that SMEs often face capital constraints, uncertainty, and risk aversion. A small firm may recognize the value of AI codification and still hesitate because the upfront cost is high, the payoff is uncertain, and the required organizational change is difficult. In addition, part of the benefit is social. Preserving tacit knowledge can protect local industrial capabilities, support supplier ecosystems, and create future options that the individual firm cannot fully capture.
This shifts the subsidy argument. The strongest case is that private investment may remain below the social optimum even when the private incentive is positive.
A Better Question for AI Policy
Most AI policy begins with the question: which firms should adopt AI?
This paper suggests a better question: which knowledge should society help preserve and redeploy?
That question leads to a more precise policy screen. Governments, industry associations, and development agencies could identify SMEs with:
- high tacit knowledge intensity,
- aging ownership or weak succession pipelines,
- meaningful potential to attract high-quality successors,
- high AI-revealed similarity to growing industries,
- and capital constraints that limit investment.
AI support for these firms would finance the conversion of fragile tacit knowledge into durable knowledge capital, rather than merely buying software.
The Bigger Point
AI is often framed as a force that replaces work. For SMEs, the more interesting possibility is that AI can preserve work that would otherwise disappear and transform it into new economic possibilities.
Some knowledge is too valuable to let vanish with retirement. Some capabilities are more transferable than traditional industry categories suggest. AI can help reveal and preserve both.
The future of SMEs may depend less on whether they automate individual tasks and more on whether they can turn tacit knowledge into a reusable, teachable, and redeployable asset. That is the real promise of AI as a knowledge intermediary.
To see more detail, please read the full paper here: [Link to the paper]
