Where should we start if most of our content is unstructured?

The instinct is to begin with documents because they are tidier. The result is a system that works well on the least valuable half of the estate and has no proven pattern for the half where the decisions actually live.

Starting with a conversational corpus proves the harder case first: classification against the organization's dictionary at volume, entitlement on material people spoke casually into, and retention on content whose status was previously ambiguous. Once that pattern holds in Centralpoint, extending to documents is straightforward — the reverse order rarely is.