Measuring product data quality: six dimensions and their metrics

How do you measure product data quality?

Product data quality is measured on six dimensions, each with a formula, a sampling method, a target and an owner. Measurement turns quality from an opinion into a number that a steering group can act on. The What is a PIM system pillar places quality inside the governance work that precedes any PIM selection.

Electronic data exchanged equals physical data. GS1's Data Quality Framework for brand owners puts the principle in those five words. The framework measures quality on completeness, timeliness and accuracy, and checks accuracy against the physical product. This hub adds three dimensions that PIM projects need: consistency, conformity to standards and uniqueness.

The six dimensions do not overlap. Each one answers a different question about a value:

• Completeness: whether the value exists.
• Accuracy: whether the value matches reality.
• Consistency: whether the value matches the same attribute elsewhere.
• Timeliness: whether the value is current where it is used.
• Conformity to standards: whether the value follows the required format, unit and code list.
• Uniqueness: whether each product and identifier exists once.

Each dimension needs a named owner. The product data governance guide explains how owners and stewards divide that work.

Which six dimensions define product data quality?

The six dimensions each come with a formula that a PIM or a simple script can calculate. The table shows the definition, the metric and the usual measurement method.

Targets differ per dimension and per channel. The sections below give the measurement details and the PIMvendors estimates for typical targets.

How do you measure completeness and accuracy?

Completeness is measured automatically per channel and locale, and accuracy is measured by sampling against reality. The two belong together, because a field filled with a wrong value counts as complete.

Completeness needs a definition of "required" per channel. Google Merchant Center, for example, requires id, title, description, link, image link, availability and price, plus brand for most new products. A retailer or data pool adds its own list. A PIM calculates completeness against each list and blocks publication below the threshold.

PIMvendors estimate: aim for 100 percent on required attributes and 80 to 90 percent on recommended attributes. Report per channel and per locale, because averages hide gaps.

Accuracy needs a sample. GS1's framework audits attributes such as dimensions, gross weight, net content and GTIN against physical products. It sets sample sizes on a 90 percent confidence rate with a 10 percent margin of error. For technical attributes, the source document is often the datasheet or the test report.

PIMvendors estimate: aim for at least 98 percent accuracy on sampled safety, dimension and compliance attributes. Lower targets can apply to marketing attributes.

How do you measure consistency and timeliness?

Consistency compares the same attribute across systems, and timeliness measures how fast an approved change reaches every channel. Both dimensions expose integration problems more than enrichment problems.

To measure consistency, take a sample of 50 SKUs and 10 attributes. Compare the values in ERP, the PIM, the webshop and one external channel such as a marketplace or retailer site. Report matching values as a percentage of compared values. Mismatches usually point to a second master, such as corrections made in a feed tool or directly in the webshop.

To measure timeliness, record the approval timestamp in the PIM and the publication timestamp on each channel. Report the median in hours per channel. PIMvendors estimate: a common target is changes visible on all owned channels within 24 hours. External channels such as retailers and data pools may follow their own cycles.

Consistency matters more since online offers carry legal information. GPSR requires manufacturer details, product identification and warnings on online offers. An inconsistent warning across channels is a compliance gap as well as a quality issue. The PIM vs feed management comparison explains how to prevent a second master in the channel layer.

How do you measure conformity and uniqueness?

Conformity checks that values follow the required format, and uniqueness checks that each product exists once. Both can be automated with validation rules in a PIM.

Conformity covers four types of rules:

1. Data types and formats, such as numeric values and date formats.
2. Units of measure, such as millimetres instead of a mix of mm, cm and inches.
3. Code lists, such as country of origin codes or packaging material codes.
4. Classification rules, such as the features required by an ETIM or ECLASS class.

Classification standards make conformity measurable at scale. ETIM is the international classification standard for technical products. ECLASS describes itself as an ISO/IEC-compliant standard, and Release 16.0 alone added 995 classes and 1,253 properties. A class defines which features apply and which values are allowed.

Uniqueness starts with identifiers. A GS1 GTIN exists in four lengths, GTIN-8, GTIN-12, GTIN-13 and GTIN-14, and each must be unique per trade item and packaging level. Run duplicate checks on GTIN and SKU first, then on combinations of brand, model and key attributes.

How does a PIM enforce data quality?

A PIM enforces data quality through validation rules, completeness gates, workflows and dashboards. Software enforces the rules, and governance decides which rules apply.

Four PIM capabilities matter most:

• Validation rules that reject non-conforming values at entry or import.
• Completeness gates that block publication per channel and locale.
• Workflows that route failed checks to the responsible steward.
• Dashboards that report all six dimensions per product line and owner.

AI adds speed in some areas. AI attribute extraction and auto-classification can raise completeness and conformity quickly. AI remains unreliable for regulatory data and cross-attribute consistency, so those values need human review. The AI attribute extraction guide covers how to test those features.

In PIMvendors selections, quality capabilities become scenarios. The vendor imports a supplier file with deliberate errors, shows which rules catch them and reports the result per dimension. The data model and data quality requirements page turns this guide into testable requirements.