Main idea

Looking for information in a company currently presupposes knowing where to look, in which tool, with which filter, under which vocabulary, on which page or in which ticket. That knowledge is written nowhere: it is in the heads of the people who have been around long enough, and it is what takes a newcomer months to acquire.

What that knowledge covers is not the question — it is everything around it: what you are trying to do, what is already known, what has been decided, what is reliable, what is uncertain, what is connected, what is off topic today but must not be lost. The human supplies that with every search, free of charge and without noticing, and the system receives only the keywords that come out of it.

A system that understands the context of the question therefore does not merely make search more pleasant: it takes over a share of work that was carried by the person, and that disappeared with them.

Why it matters

It names a cost that appears in no budget: the time spent learning the structure, paid by every newcomer, and paid again at every change of tool or naming scheme.

It also gives a more demanding criterion of progress than a conversational search interface. The question is not whether you can ask your question in natural language, but whether you can ask it without knowing in advance where the answer lies.

Nuances and limits

Part of that structural knowledge is business, not plumbing: knowing that a warranty claim is handled differently from a production incident cannot be outsourced to an engine.

And shifting the load onto the system presupposes that the system has what the person had in mind — hence that the context has been written. Without that, the load is not shifted, it is merely lost.

Open questions

• How do you measure the share of structural knowledge a team still carries, so as to know what a departure would cost it?