Catalog Everything Versus Make Metadata Useful
Atlas Health Systems has 600 BI reports built over eight years. Logs show 240 were viewed in the last 90 days. The business glossary contains 900 terms; only 130 were viewed last quarter, and 60 have no owner. A regulatory request for lineage on a claims report took 19 days to answer, assembled manually from three ETL tools and email. Report owners are unknown for about half the reports. The CIO has funding for a metadata platform and wants to catalog everything at once. Business teams say the glossary is an IT dictionary and do not use it.
1. The CIO wants to catalog all 600 reports and 900 glossary terms at once. Critique this scope and propose a defensible alternative.
Read the worked answer
Cataloging everything treats metadata as inventory rather than a service. The usage data already segments the estate: 240 reports are used, so about 360 are candidates for archive or retirement; cataloging them first would spend effort on assets that may disappear. The glossary has 130 viewed terms and 60 unowned, so most terms are not validated by use.
Alternative: define a metadata service with users and SLAs. Phase 1, 60 days: certify the 240 used reports with owner, definition, source, and freshness; resolve the 60 unowned glossary terms by assigning owners or retiring them; connect the glossary to the reports people actually open. Phase 2: retire or archive unused reports with owner sign-off, which shrinks the scope. Phase 3: extend to remaining reports on demand. Success is time-to-answer and use, not record count.
2. How should Atlas approach lineage so the next regulatory request does not take 19 days?
Read the worked answer
Do not attempt full enterprise column-level lineage up front; it is expensive and often stale. Tier the estate by regulatory and financial materiality. For the claims report and similar critical outputs, starting with 30-50 reports, build column-level lineage across the three ETL tools, automating where parsers work and documenting manual gaps. Store lineage as a maintained asset with an owner and review cycle. For the rest, provide table-level lineage and a request playbook.
Then rehearse: run the next request through the playbook and measure days to answer, aiming for two days for in-scope reports. Use the 19-day failure as the baseline and publish the improvement.
3. Business teams call the glossary an IT dictionary. What would make it adopted?
Read the worked answer
Adoption follows usefulness in the workflow, not completeness. Embed definitions where decisions happen: in BI tool tooltips, report headers, and certification badges; make the glossary the place where a disputed term is resolved, with an accountable owner and a decision date. Recruit three or four business-led domains, such as claims and finance, to own their terms and publish them with examples and exclusions. Retire unused terms rather than defending them.
Measure searches that end in a report or definition used, and the number of disputes resolved through the glossary; report those to the CIO instead of term counts.