Data quality audits - finding gaps before they cost you sales

Every team believes their product data is "mostly fine." It's unfalsifiable until you measure. An audit is just measurement made repeatable: which fields, which products, filled or not, consistent or not. The output is a number per attribute - and numbers end debates.

The simple version, this week

Export your catalog (Shopify's product export, or Matrixify for metafields). In a spreadsheet, take your top category and the required attributes you defined in the data modeling lesson. For each attribute, two checks:

Completeness: what percentage of products have any value at all? A blank material field on 30% of your garments is a hard number your team can act on.

Consistency: how many distinct values exist, and are they legitimate? Sort the column, eyeball it. "Cotton", "cotton", "100% Cotton", "Coton" - four values, one material, three problems. For closed-list attributes, any value outside the list is a flag.

Two hours of work, and you'll know more about your catalog than most brands ever do.

Where gaps actually cost money

Prioritize by exposure, not by tidiness. A gap costs most where the attribute is load-bearing: attributes your filters run on (empty field = product invisible in filtered results - a stockout you can't see), attributes your feeds map (gaps become Merchant Center warnings or weak Shopping performance), and attributes on your revenue-driving products. A missing care instruction on a slow mover is cosmetic. A missing material on a bestseller is a leak.

So: weight your audit by traffic and revenue. Twenty perfect long-tail products matter less than one incomplete hero product.

Make it a rhythm, not a project

The failure mode of audits is heroics: a giant one-time cleanup, followed by decay, followed by another giant cleanup two years later. The alternative is boring and effective - the same audit, quarterly, same spreadsheet, tracking the same numbers over time. New products are where gaps enter (suppliers, seasonal launches, rushed uploads), so completeness on products added in the last quarter is the most telling number of all. If new products arrive complete, the catalog heals itself. If they don't, no cleanup will ever stick.

At real catalog scale, running this manually stops being fun - that's what the Data Operations phase is for. The rhythm is what matters; the tooling can grow with you.