Catalog Adoption, Lineage Gaps, and Glossary Duplication
Calder Manufacturing has scanned 15,800 tables and 62,000 columns into Enterprise Data Catalog across Oracle, SQL Server, Snowflake, and S3, plus 340 PowerCenter mappings and 180 Tableau workbooks. Six months in: 34% of assets have no assigned steward, 1,200 glossary terms were imported from three business units with heavy duplication ('Customer', 'Client', 'Account Holder'), and lineage from PowerCenter to Tableau breaks at 60% of workbooks because they use custom SQL. The catalog's nightly profiling job now runs 14 hours and competes with the ETL window. Business users complain that search returns stale assets nobody uses, and the CDO wants to demonstrate value in one quarter.
1. The CDO wants value in one quarter. How do you prioritize?
Read the worked answer
Prioritize by value and risk, not by coverage percentage.
First, fix the operational hazard: a 14-hour profiling job that competes with ETL is a production risk. Move profiling out of the ETL window, restrict it to critical assets, and sample large tables. That is a one-week change that removes an incident waiting to happen.
Second, pick the top 200 assets by query frequency and downstream dashboard usage, assign stewards to those, certify them, and surface them ahead of uncertified results in search. Certification of 200 high-use assets is worth more than steward coverage on 15,800 tables.
Third, freeze glossary imports and consolidate terms with a small governance council - three business units arguing over 1,200 terms resolves faster than a tool rollout.
Fourth, fix lineage for the twenty most-used Tableau workbooks, not all 180.
Report certified assets, percentage of critical assets with a steward, canonical term reuse, and search success rate. Coverage of the full estate is a year-two metric.
2. How do you close the lineage gap for workbooks built on custom SQL?
Read the worked answer
Use a layered approach and be explicit about confidence per edge.
First, ensure the Tableau metadata is being harvested through the scanner or API so workbook-to-database connections are known. Then run the catalog's SQL parsing on the embedded custom SQL to extract table references. Where parsing succeeds, you get table-level lineage automatically - schedule it to re-parse on change.
Where parsing fails, manually stitch at table level and mark the relationship as inferred rather than parsed, so impact analysis can warn users that the edge is asserted. Do not fake column-level lineage; a wrong column-level edge is more damaging than a declared table-level one.
Validate the result by running an impact analysis on one change you know the answer to: pick a column, see whether the catalog lists the right workbooks. Repeat quarterly, and track the percentage of workbooks with parsed versus inferred lineage as the program metric.
3. How do you consolidate 1,200 duplicate glossary terms without a political deadlock?
Read the worked answer
Make it a term ownership exercise, not a taxonomy project.
Define a canonical term per concept with a named business owner, and treat 'Client' and 'Account Holder' as synonyms of 'Customer' rather than as competing entries. Synonyms preserve search behavior, so users who type the old word still find the right asset - that removes most of the resistance to merging.
Prioritize the 50–80 terms that appear in certified assets and regulatory reports; leave long-tail terms for later. Route changes through a lightweight approval workflow with a single accountable owner per term, and require new terms to be justified against existing ones before creation.
Map assets to canonical terms, not to the duplicates, so the catalog's business-facing view stabilizes. Measure success as canonical terms defined, percentage of certified assets mapped to a canonical term, and reduction in duplicate terms - not the raw term count, which can rise legitimately as coverage improves.