The four levels of product data validation
Strong validation uses more than a completed-field check. A record can be structurally complete and still be technically implausible or supported by the wrong source.
- Structural validation: required fields, formats, data types and identifier patterns.
- Controlled-value validation: approved units, categories, attribute names and enumerated values.
- Cross-field validation: logical relationships between dimensions, quantities, variants and related specifications.
- Source validation: provenance, source priority, recency and agreement between supporting documents.
Validation rules should match product risk
Not every field carries the same consequence. A missing merchandising phrase is different from an uncertain voltage, size, material or certification. The review path should reflect the impact of being wrong.
High-risk attributes can require a named source and human approval. Lower-risk fields may proceed when confidence and validation conditions are satisfied.
Exceptions are part of the design
A real product catalog contains incomplete files, duplicate identifiers, discontinued items, unusual variants and supplier-specific conventions. A workflow that only handles clean inputs pushes the difficult work back onto employees.
An exception queue should show the proposed value, supporting source, failed rule and decision required. The reviewer’s correction should then improve the mapping or validation logic when appropriate.
Measure validation as an operational system
Useful measures include records passing without intervention, exceptions by rule, time in review, corrections after approval and repeat exceptions that indicate a missing rule. These measures help prioritize workflow improvements without pretending that every exception should disappear.