The Pattern Between Industries

I have spent a long career moving between regulated industries — legal, health, insurance, financial services — and the most useful thing I have learned is not specific to any of them. It is that they keep solving the same problems, separately, at enormous expense, without noticing that the problem next door is theirs.

A claims-adjudication workflow in insurance and a clinical-coding workflow in health are, structurally, the same machine: take an unstructured human account, map it onto a controlled vocabulary, attach evidence, and produce a decision that has to survive audit. A know-your-customer process in finance and a conflicts check in a law firm are the same machine again. Each industry believes its version is sui generis. None of them is.

Why the pattern stays hidden

If the structure is so similar, why doesn’t everyone see it? Because the costume is industry-specific and the costume is what people are trained on. A health professional learns clinical coding as a clinical problem. An underwriter learns risk classification as an insurance problem. The vocabulary, the regulators, the failure modes, the war stories — all of it is local. The shared skeleton underneath is invisible precisely because everyone is an expert in the surface.

This is not a failure of intelligence. It is a structural consequence of specialisation. Depth in one domain is bought with the time you would otherwise spend noticing that the domain next door has the same bones.

Pattern recognition as a non-replicable asset

In a world where models can absorb the substance of any single domain quickly, the durable advantage is not knowing one industry deeply. A model will know it too. The durable advantage is the connective tissue — the cross-domain pattern library that lets you look at a new problem in one field and recognise the solved version of it in another.

This is hard to replicate because it cannot be acquired by study. You cannot read your way to it. It is the residue of having actually worked the same structural problem in genuinely different contexts, with different stakes and different regulators, until the shared skeleton becomes the thing you see first and the costume becomes the thing you see second. That inversion — skeleton first, costume second — is the asset.

What it is good for

Two things, mostly. It makes you fast: a problem that looks novel to a domain specialist is often, to the pattern-matcher, a known shape with new labels, and you can skip straight to the part that is genuinely new. And it makes you a translator: you can carry a solution that matured in one industry — where the pressure forced it to get good — into another that is still solving it the hard way.

The intelligent-systems era rewards this more, not less. As models flatten the cost of domain knowledge, the premium moves to the thing models are worst at: the lateral leap between contexts that no single training distribution covers well. The pattern between industries is exactly that leap. It is the part of the work that does not get automated, because it is made of having been there.