Definition
Catalog quality scoring is the instrument that converts a catalog's quality into a repeatable number — a weighted, per-category composite of checks such as required-attribute coverage, value validity, and duplication, computed the same way every run so scores are comparable over time. The score is not the quality; it is a deliberately lossy summary of it, designed to be tracked, compared, and used to trigger work.
Key points
- A score is a weighted composite — attributes do not matter equally, and a scoring model that treats them equally produces a number nobody trusts.
- Weights should follow consequence: an attribute that blocks a feed or a filter outweighs a nice-to-have descriptor.
- Scores are computed per category, because required attribute sets differ; a single global score averages away the categories actually in trouble.
- Thresholds turn the number into a decision — publish, hold, or route for enrichment — and without them the score is decoration.
How is a catalog quality score constructed?
A team defines a check set per category — required attributes present, values inside an allowed vocabulary, units normalized, identifiers unique, images meeting spec — and assigns each check a weight reflecting what breaks when it fails. Each product yields a sub-score per dimension; those roll up to a product score, then to category and catalog roll-ups. Thresholds are then attached: below one line a SKU is held from syndication, below another it is queued for enrichment, and a per-category average that drops week over week raises a flag. The weighting scheme and the thresholds are versioned, because changing either silently makes historical scores incomparable.
Common pitfalls
- A score nobody acts on. If no threshold routes work, blocks a publish, or owns a target, the dashboard becomes a vanity metric that improves only when someone games the checks.
- Flat weighting, which lets ten trivial descriptors outrank one missing GTIN and hides the failures that actually cost money.
- Changing weights or check definitions mid-stream without versioning, so a jump in the score reflects a changed ruler rather than a changed catalog.
FAQ
How is scoring different from measuring catalog quality?
Scoring sits on top of measurement and adds three things measurement does not have: weights that declare which failures matter more, roll-up rules that combine per-attribute results into product, category, and catalog figures, and thresholds that turn a figure into an action. Take those three away and what remains is a measurement report — which plenty of teams produce and never score.
What is a good catalog quality score?
There is no portable answer, because the number depends entirely on your own weights and check set. A score is meaningful only against your own baseline and thresholds; comparing your composite to another organization's compares two different rulers, not two catalogs.
Source
Most scoring models take their required-versus-recommended attribute distinction from a published specification; Google's Merchant Center product data specification marks attributes as required, optional, or conditionally required by category, which is exactly the structure a weighted, per-category score is built on.