Teaching AI How People Work Is Fraught With Problems

For AI to be genuinely useful inside an organisation, it needs to understand the specific context in which it operates. That context is partly made up of explicit rules and guidelines — policies, process documentation, procedure manuals — that can be codified relatively easily. But a large and often overlooked portion of workplace knowledge is tacit: the hard-to-articulate, experience-based know-how that people absorb over years of doing a job. A recent essay in The Business Times explores why this tacit knowledge gap is one of the most stubborn obstacles to effective enterprise AI adoption.

The article highlights two primary routes to giving AI the context it needs. The first is straightforward: document explicit rules upfront. The second is data-driven: technologies like Celonis, a German process-mining firm, ingest raw data from enterprise systems to map how workflows like invoices and procurement actually unfold in practice — rather than how they are supposed to unfold on paper. But even the best process-mining tools struggle with the deeply contextual, intuitive decisions that experienced workers make without conscious deliberation. That gap is where AI deployment often stalls.

For Singapore businesses racing to deploy AI across finance, logistics, and professional services, this is a practical concern. Many companies have invested heavily in AI platforms only to find that the tools underperform because they lack the granular understanding of how work really gets done. The risk is that organisations underestimate the effort required to surface and encode tacit knowledge — or worse, assume that AI can simply learn it from raw data without human guidance. The essay warns that the harder, slower work of translating unspoken expertise into something machines can act on remains a fundamental bottleneck.

Why it matters for Singapore: Singapore's economy is built on services, and services run on tacit knowledge. From a relationship manager at DBS who reads between the lines of a client's request, to a logistics coordinator at PSA who instinctively knows when to escalate, the expertise that makes Singapore companies competitive is often the hardest to encode for AI. As organisations push forward with automation and AI adoption, those that invest in surfacing this unspoken knowledge — rather than just throwing data at their models — will see the strongest returns. The challenge is not technological; it is deeply human.