Retail: trusted analytics with Data Products in the Unified Catalog

Analysts at a retail chain did not trust the numbers in their own reports: the same term, "active customer", meant three different things across three departments, and nobody could trace a table back to its source. Every new report started with a hunt for the right source instead of with analysis.

The Data Map scans and classifies the underlying sources (sales, inventory, CRM) and builds lineage automatically. On top of that, the Unified Catalog sets up Governance Domains per business area (Sales, Inventory, Customer), each with its own domain owner. Within each domain, Data Products are assembled: bounded, described collections of tables with an owner, a quality score and access policy. A business glossary ties terms like "active customer" to one, organization-wide definition.

The alternative, a generic data catalog without ownership, had already been tried and stalled into a list of tables nobody maintained. Data Products force an owner per collection, and that owner is accountable for quality and freshness, not the platform team.

A spreadsheet-based data dictionary partially solved the definition confusion, but had no link to the actual data and went stale within months. A homegrown catalog based on metadata tags gave findability, but no lineage and no quality monitoring.

• Data Products need an owner per domain; without dedicated capacity the catalog stays half empty.
• Splitting into too many small Data Products raises maintenance load; overly large Data Products lose clarity of ownership.
• The Unified Catalog's pay-as-you-go pricing calls for deliberate scoping: not every domain at once.

Microsoft Purview (Data Map, Unified Catalog with Governance Domains and Data Products), Microsoft Fabric, Power BI.

• Start with the one or two domains with the biggest trust problem, not the whole organization at once.
• Give every Data Product a measurable quality score, not an optional description.
• Let the business own and maintain the business glossary, not IT.

The biggest acceleration came not from the technology but from ownership: once a Data Product had a name and a face, quality visibly improved within weeks. Before owners were assigned, the catalog stayed largely unused.