Look at where AI has landed first and you see a pattern. Coding, support, search. Tidy, text-based, easy to verify, easy to automate. That is where everyone crowded, because it is the easiest work to point a model at. Which is exactly why it is not the opportunity. The opportunity is the opposite corner. The messy, regulated, relationship-heavy industries that run on undocumented process and people’s memory. Trades, aged care, allied health, accounting, construction, strata, compliance. High friction, low adoption, and almost no one has built the operating layer.

Why messy is the moat

Friction looks like a problem. It is the moat. A messy industry is hard to enter, hard to codify, and hard to copy, which is precisely why the business that does codify it is so defensible. The friction creates the workflow. The workflow creates the data. The data creates enduring enterprise value. The tidy work has none of that, which is why it is already commoditised. The mess is the barrier, and the barrier is the point.

The research agrees

The same studies that show most AI failing also show where it succeeds. The wins are in deeply integrated, vertical, often regulated workflows, the unglamorous back office that everyone overlooks. Generic tools fail. Specialists who know one industry succeed at roughly twice the rate. The pattern is clear. The return is in the mess, for whoever is willing to go into it.

Everyone wants the tidy work because it is easy. The value is in the work that is hard.

Why almost no one will

Most software people will not touch these industries. They are slow, regulated, relationship-driven, and they cannot be served from a distance with a generic product. You have to go inside, run the work, and earn the right to codify it. That is unattractive to a company that wants to ship a tool and scale. It is exactly the opening for an operator who is willing to do the work.

Where to start in your industry

You do not boil the ocean. You find the one workflow where the most time, margin and intelligence is leaking, and you run it on a layer that captures the work. Quoting in trades. Intake in legal. Rostering in aged care. The close in accounting. Fix that one, own the data it throws off, and the rest of the play follows.