Data quality metrics for merchandising and catalog

This lesson shows you how to measure it.

Why catalog quality is a data problem, not a content problem

Merchandisers often treat missing product data as a "someone forgot to fill it in" issue. It is actually a measurable, trackable data-quality problem with direct downstream effects.

Two systems consume your catalog data:

• The customer-facing system (site search, filters, product pages). Missing attributes mean products do not appear in filtered results. A dress with no "sleeve length" value is invisible when a shopper filters for "long sleeve."
• The allocation and planning system. Duplicate SKUs (Stock Keeping Units, the unique code for each sellable variant) split demand signal across two records, so your replenishment model under-orders both.

Bad catalog data breaks discovery and it breaks the math that decides where inventory goes.

The four core metrics

We track four metrics. Each is a simple ratio you can compute today from a product export.

1. Attribute completeness rate

The share of required attributes actually populated across your catalog.

First, define a required attribute set per category. For dresses that might be: color, fabric composition, sleeve length, neckline, dress length, care instructions, country of origin. For handbags it differs: material, dimensions, strap type, closure.

Then:

Worked example. You have 500 dresses, each needing 7 attributes. That is 3,500 required cells. Your export shows 2,940 are populated.

Benchmark note: retailers commonly target 95 percent or higher for search-critical attributes. Anything below 90 percent usually means visible gaps in filtered navigation. (Treat "95 percent" as a widely cited operational target, not an official standard.)

Do not average all attributes equally. Weight the ones customers filter on. A missing "care instruction" hurts less than a missing "color" that removes the item from every color filter.

2. Image coverage

Images drive fashion conversion more than any single text field. Measure two things.

Minimum image count compliance: the share of SKUs meeting your image standard. If your standard is "at least one on-model shot plus three detail shots," a product with only a flat-lay fails.

On-model coverage specifically, because apparel converts far better with a body shot than a flat product-only image.

Example: 500 dresses, 430 have an on-model image.

The 14 percent gap is often new-season arrivals or marketplace-sourced items where you inherited only the vendor's flat image.

3. Size-chart accuracy

The most under-measured and the most expensive. Wrong size charts drive returns, and apparel return rates are high: for online apparel, returns commonly run in the 20 to 30 percent range, with a large share driven by fit (widely cited industry estimates, as of 2025). Fit-driven returns are pure margin leakage plus reverse-logistics cost.

Size-chart accuracy is harder to compute than the others because "correct" is not a filled-or-empty cell. Practical measures:

• Chart-to-product linkage rate: share of SKUs mapped to the correct size chart for their brand and category. A US brand's chart applied to an EU-sized item is a defect.
• Measurement consistency check: flag charts where measurements do not increase monotonically across sizes (an S chest larger than an M chest is a data error).

Run this across every chart weekly. It catches the "M and L swapped" errors that generate return spikes.

4. Duplicate-SKU rate

Duplicates arise when the same physical product enters your system twice: once from a vendor feed, once from manual entry, often with slightly different titles.

Detecting duplicates needs matching logic, not just exact-match on title. Match on combinations: brand + style number + color + size, or use GTIN (Global Trade Item Number, the barcode standard) where available. If two records share a GTIN, they are the same product.

Example: 12,000 SKU records, matching logic finds 360 that are duplicates of an existing record.

Even 3 percent matters: those 360 split inventory and demand across two records, corrupting your allocation model.

🎬 [VIDEO: "Product Data QualityData QualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → Fundamentals" - youtube.com - a clear walkthrough of completeness, consistency, and accuracy dimensions applied to retail catalogs]

Where this data lives

Your metrics are only as good as your sources. The typical apparel catalog data flowdata flowAn automated sequence of steps that moves data from source to destination: ingestion, transformation, validation, and loading, so it arrives clean and ready to use.View full definition →:

• PIM (Product Information Management): the master system holding attributes, descriptions, and image links. This is where completeness is measured and enforced.
• DAM (Digital Asset Management): stores the actual image and video files that image-coverage metrics reference.
• ERP / merchandising system: holds SKU records, GTINs, and inventory, the source for duplicate detection.
• Vendor / marketplace feeds: inboundinboundA strategy that attracts prospects organically via valuable content (blog, SEO, social) rather than interrupting them.View full definition → data of variable quality, the most common source of gaps and duplicates.

The governance principle: define required attributes and image standards inside the PIM as validation rules, so bad records cannot be published in the first place. Measuring quality after publication is triage. Blocking at ingestion is prevention.

For a solid vendor-neutral primer on the discipline behind these dimensions, see the DAMA data quality dimensions overview, the industry body for data management.

Turning metrics into a scorecard

Combine the four into one catalog-health view, refreshed weekly, sliced by category and by data source.

| Metric | Example value | Target | Status |

|---|---|---|---|

| Attribute completeness (weighted) | 84% | 95% | Below |

| On-model image coverage | 86% | 95% | Below |

| Size-chart linkage | 91% | 99% | Below |

| Duplicate-SKU rate | 3% | Under 1% | Below |

Slicing by source usually reveals that marketplace and dropship feeds carry most defects, while first-party in-studio products score high. That tells you exactly where to invest: vendor onboarding rules, not more internal staff.

Connecting quality to outcomes

To get budget for data-quality work, connect metrics to money using data, not guesses.

Discovery lift. Run a before/after test. Take 200 dresses missing "sleeve length," populate it, and measure their impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition → in filtered search over four weeks versus a control group. Products newly appearing in filters should gain impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition →. This is a clean, measurable causal test.

Return reduction. Track return rate for SKUs before and after size-chart correction. If a corrected chart on a jeans style drops its fit-return rate from 28 percent to 22 percent, that six-point drop is directly attributable and quantifiable per unit.

Allocation accuracy. After de-duplicating, compare forecast error (measured as MAPE, Mean Absolute Percentage Error, the average size of forecast misses) for merged SKUs. Consolidated demand signal typically tightens the forecast.

The point: never argue for data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → on principle. Instrument it. Every one of these metrics has a measurable downstream effect you can test.

A simple cadence

• Daily: validation rules block incomplete records at PIM ingestion.
• Weekly: run the four-metric scorecard, sliced by category and source.
• Monthly: run the non-monotonic size-chart check and full duplicate scan across the whole catalog.
• Quarterly: run one before/after discovery or return test to keep the money story fresh.

Key Takeaways

• Four metrics cover most catalog risk: attribute completeness, image coverage, size-chart accuracy, and duplicate-SKU rate. Each is a simple ratio from a product export.
• Weight completeness by what customers filter on. A missing color removes an item from search; a missing care label rarely does.
• Size-chart accuracy is the costliest and least measured. Fit-driven returns (a large slice of the commonly cited 20 to 30 percent online apparel return range) trace straight back to wrong or mislinked charts.
• Prevent at ingestion, not after publish. Enforce required attributes and image standards as PIM validation rules; slice defects by source, since marketplace feeds usually carry most of them.
• Always connect quality to a measured outcome: discovery impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition →, return rates, or forecast error. Instrument the effect, do not assert it.