Glossary
Data profiling is the examination of a dataset to understand its structure, content and patterns before deciding how to use or improve it. In product data, profiling can reveal missing attributes, unusual values, repeated identifiers, inconsistent units and unexpected formats. It provides evidence for designing mappings and quality checks; it does not itself correct the records.
Profiling examines what is actually present, rather than what an import template says should be present. Useful observations include the number of populated values, distinct options, value lengths, ranges and patterns within each attribute. Results can be grouped by supplier or product category to expose differences hidden by catalogue-wide totals.
Keep examples of unusual values alongside the statistics. A count of distinct strings is easier to interpret when the reviewer can see whether the differences concern spelling, units or genuinely different specifications.
A supplier file contains entries such as "500 mm", "0.5 m", blank cells and "see drawing" in one length column. Profiling reveals mixed representations and the proportion of records that need another source. It can also show whether these patterns are concentrated in particular product families.
The team can then propose a numeric length attribute with a defined unit, a transformation for recognised values and a review route for unresolved rows.
Profiling describes observed patterns. Validation tests explicit acceptance rules. An unusual value can be correct, while a common value can be wrong, so profiling findings need interpretation. After implementing corrections or new rules, repeat the profiling against a documented scope to see what changed.
Explore practical support for organising, improving and maintaining product information.