AI Doesn't Just Miss Tacit Knowledge. It Can Destroy It.

The Standard Framing Gets Something Important Wrong

The dominant logic behind most AI adoption is straightforward: find the task, automate the task, reduce the cost or time. It is a reasonable starting point. The problem is what it misses.

Tasks do not exist in isolation. Behind most of the tasks worth automating is a body of expertise: judgement built over years, pattern recognition that operates faster than it can be explained, contextual awareness that no process document has ever fully captured. This is tacit knowledge, and it is almost never accounted for in the standard AI adoption model.

The typical response to this, when it is considered at all, is to treat tacit knowledge as a documentation problem. Interview the experts, write it down, encode it in the system. But that approach has a structural flaw: tacit knowledge resists documentation precisely because it operates below the level of conscious articulation. The engineer who knows when something is about to fail does not always know how they know. The analyst whose read on a situation proves consistently correct often cannot fully explain the reasoning. The knowledge is real and operational; it just does not transfer well through structured capture.

The Risk Nobody Talks About

The consequence of ignoring this is widely understood: AI systems that miss the nuance, fail to generalise to edge cases, and produce outputs that experienced practitioners immediately distrust.

The less discussed consequence is more serious.

When AI is designed to handle decisions that previously required expert judgement, it does not just fail to capture that expertise. Over time, it removes the conditions under which that expertise develops. Practitioners who no longer need to exercise a skill stop developing it. Organisations that route decisions through automated systems stop building the institutional knowledge that informed those decisions in the first place. The system performs the task. The expertise behind the task quietly degrades.

This is not a hypothetical risk. We have seen it in practice: organisations that adopted AI to handle complex operational decisions and, some years later, found they had a generation of practitioners who had never developed the depth of expertise their predecessors had, because the system had never required it of them.

What Good Design Looks Like

The alternative is not to avoid automation. It is to start from a different question.

Rather than asking which tasks can be automated, ask what your most experienced people actually know, including the things they have never been asked to articulate. Map the expertise that sits behind the tasks, not just the tasks themselves. Understand which knowledge is most at risk of being lost, through retirement, turnover, or simply through the attrition that comes when a skill is no longer regularly exercised.

Then design AI that works with that expertise rather than around it. Systems that surface expert judgement rather than replace it. Systems where each interaction is an opportunity to encode reasoning that would otherwise remain implicit. Systems that become more valuable over time rather than more brittle, because they are building organisational knowledge, not just processing information.

Our research in this area, including work on voice-based knowledge capture in operational environments and contributions to discussions at events including Big Data London, informs how we approach this problem in practice. The design principles that emerge from that work are different from the principles that govern conventional AI adoption, and the outcomes, for organisations that apply them, tend to be meaningfully different too.

The starting point is the same in every case: understand the expertise first. Everything else follows from that.

For more on what this means in practice, the piece on AI for knowledge transfer goes deeper on the mechanisms. For a framework to assess where your organisation stands, the five questions before any AI project is a useful place to start.