You bought a data catalogue eighteen months ago. It is populated and nobody uses it. What went wrong?

Populated is not the same as useful

Two failures produce this outcome and they need different fixes.

The content went stale. Technical metadata was loaded once at implementation and never refreshed, so it lists tables that no longer exist and omits ones that do. People trusted it, were wrong twice, and stopped.

It never answered the question people actually have. The catalogue lists schemas and column types, which is not what anyone is looking for. The real questions are: is there a dataset about X, can I trust it, who owns it, and is anyone else using it.

Fix the first with harvesting

Anything requiring manual entry will be incomplete within a quarter. Schemas, sizes, partitions and last-modified times are read from the platforms continuously. Lineage is derived from query logs and orchestrator definitions rather than drawn. Usage comes from access logs.

Fix the second by adding the three fields that matter

Owner — a person, resolved automatically from the repository or the team directory rather than typed in.

Usage — who queries this and how often. It answers "can I trust it" more convincingly than any quality score, and it is free from the access logs.

A description with the caveats, which is the only genuinely manual part. Get it by making the cost tiny and the moment natural: a description field in the transformation model, reviewed in the same pull request as the logic. Not a portal somebody is asked to visit.

The measure

Whether people search it before asking a colleague. If they still ask in chat, the catalogue is not answering their question, and adding more datasets will not change that.