Product quality scoring and automatic category classification

Overview

Product quality scoring and automatic category classification are the foundation of channel-ready product content. Every product in your catalogue receives a quality score from 0 to 100, with a dimensional breakdown that shows exactly which data gaps or errors are holding it back. Vendably also matches each product to a canonical taxonomy category automatically, flags uncertain matches for review, and infers missing fields such as brand or GTIN from the data you already have. When scores fall below your configured threshold, you receive quality alert notifications so you can act without constant manual checking.

This article explains how these capabilities work, where to find them, and what success looks like in practice.

Product quality scoring: dimensions and thresholds

Every product receives a quality score calculated from multiple dimensions of data completeness and correctness. The score ranges from 0 to 100, and each dimension contributes to the overall result. You can see the dimensional breakdown for any product to understand which areas need attention.

Quality score dimensions

The following dimensions make up a product's quality score:

Each dimension is scored independently, then weighted to produce the overall quality score. A product missing images might score 45 overall even if its description is excellent, because image completeness carries significant weight in channel compliance.

You can view the dimensional breakdown in the DataHub, Product Hub, or the source detail view for any product. The breakdown shows not just the score, but the specific fields or attributes causing the gap.

Configuring quality thresholds

You set a quality threshold (typically 70 to 85 out of 100) below which you want to be notified. This threshold is the trigger for quality alert notifications. Rather than checking scores manually every day, alerts come to you when a product falls below your standard or when a previously approved product degrades.

Quality alert notifications

Quality alerts notify you when a product's score drops below your configured threshold. Alerts are triggered in two scenarios:

1. A new product is added to your catalogue and scores below the threshold on first assessment.
2. An existing product's score drops below the threshold due to data changes, missing enrichment, or channel-specific validation failures.

Alerts include the product identifier, current score, the dimensions dragging the score down, and a direct link to the product detail view so you can make corrections immediately. You can configure alert delivery via email or in-platform notification, and set different thresholds for different product categories if needed.

Alerts reduce the overhead of quality management. Instead of running a daily report and manually scanning for problems, you respond only to products that fall below your standard. This is particularly valuable in large catalogues where hundreds of products might be in flight at any given time.

Automatic category classification

Vendably matches each product to a canonical taxonomy category automatically using text-matching techniques. The system reads your product title, description, and existing category data, then compares it against the taxonomy structure you are distributing to (for example, Google Product Taxonomy or a channel-specific taxonomy).

How automatic classification works

The classification process follows this logic:

1. Exact match attempt: The system searches for an exact match between your product data and known taxonomy categories. For example, a product titled 'Wireless Bluetooth Headphones' might match exactly to 'Electronics > Audio > Headphones > Wireless Headphones'.
2. Partial match with fallback: If an exact match cannot be found, the system falls back to a broader parent category. A product with incomplete or ambiguous data might be placed in 'Electronics > Audio > Headphones' rather than a more specific subcategory.
3. Confidence scoring: Every classification receives a confidence score. High-confidence matches (90+) are applied automatically. Lower-confidence matches are flagged as candidates for manual review.
4. Audit trail: The system records which data fields were used to make the classification decision, so you can see why a product was placed in a particular category.

Automatic classification saves time and ensures consistency across your catalogue. Without it, you would need to manually assign categories to every product, a task that becomes impractical at scale and introduces human error.

Reviewing and correcting classifications

Products with low-confidence category matches appear in a review queue within the Product Hub or DataHub. You can accept the suggested category, override it with a manual selection, or request re-classification if product data changes. Accepted classifications are locked unless you explicitly change them; this prevents accidental reclassification on the next import.

Incorrect category assignments directly affect channel performance. A product in the wrong category will not appear in the correct search results, will not match the right buyer intent, and may be rejected by channels that validate category against product attributes. Reviewing low-confidence matches early prevents these downstream problems.

Product enrichment: inferring missing fields

Vendably can infer missing fields such as brand, GTIN, or product type from the data you already have. For example, if you provide a product title that includes a well-known brand name, Vendably can extract and populate the brand field automatically. If you provide a product description that mentions a GTIN or barcode, the system can parse and validate it.

How enrichment works

Enrichment uses pattern matching, known product databases, and cross-reference data to fill gaps:

• Brand extraction: Recognised brand names in titles or descriptions are extracted and populated in the brand field.
• GTIN inference: Barcodes or product codes mentioned in descriptions or attributes are parsed and validated as GTINs.
• Product type inference: Product category, title keywords, and attribute data are combined to infer product type (for example, 'clothing', 'electronics', 'home and garden').
• Attribute inference: Missing colour, size, or material attributes are inferred from title or description where possible.

Every inferred value is written back to the product record with a visible audit trail. You can see exactly what was inferred, which field it was written to, and when. This transparency is critical for quality control: you can accept inferred values that are correct, reject or correct those that are wrong, and use the pattern to improve your source data going forward.

Enrichment increases your quality scores by filling gaps without requiring manual data entry. It also improves channel compliance: a product with a valid GTIN and brand is more likely to be approved by marketplaces and less likely to be flagged for deduplication.

Reviewing enriched fields

Enriched fields are marked as 'inferred' in the product detail view. You can review them in bulk, accept them to lock them in place, or reject them if they are incorrect. Rejected inferred values are not re-applied on the next data refresh, so the system learns from your corrections.

Where to access quality scoring and classification

Product quality scores, dimensional breakdowns, category classifications, and enriched fields are accessible from three locations:

DataHub

The DataHub provides a catalogue-wide view of quality metrics. You can filter products by quality score range, view aggregate quality trends over time, and export a list of products below your threshold. The DataHub is your starting point for understanding catalogue-wide quality health.

Product Hub

The Product Hub shows individual product detail, including the quality score card, dimensional breakdown, category classification with confidence score, and enriched fields with audit trails. You can edit data, accept or reject enrichments, and recalculate scores from here.

Source detail view

When you import product data from a file or ecommerce platform, the source detail view shows how that product was scored and classified at import time. You can see the original source data alongside the enriched version and any category suggestions that were not yet accepted.

Practical workflow: improving quality at scale

A typical workflow for improving product quality looks like this:

1. Set your threshold: Configure a quality threshold (for example, 75 out of 100) and enable alert notifications.
2. Review alerts: When alerts arrive, open the product detail view and see which dimensions are dragging the score down.
3. Prioritise by impact: Fix products in high-traffic categories or high-volume SKUs first, as these have the biggest impact on channel performance.
4. Address gaps systematically: Common gaps are missing images, short descriptions, or missing identifiers. Batch-fix these using feed rules or enrichment rather than editing one product at a time. See feed-rules-filtering-and-transforming-products-at-export for details on bulk transformations.
5. Review enriched fields: Accept inferred brand and GTIN values where they are correct. Reject or correct those that are wrong.
6. Review low-confidence categories: Accept or override category suggestions for products flagged for manual review.
7. Re-import and re-score: When you next import product data, scores are recalculated and alerts are triggered only for products still below threshold.
8. Monitor trends: Use DataHub reporting to track average quality score over time. A rising trend indicates that your enrichment and correction efforts are working.

Success metrics

Success in product quality and classification looks like:

• Rising average quality score across your catalogue: You should see the median score increase from, for example, 62 to 78 over a quarter as you address gaps and enrich data.
• Fewer quality alerts: As products improve, fewer products fall below threshold. Alerts become less frequent and more actionable.
• Correct category assignments: Products are assigned to the right taxonomy categories on first classification. Low-confidence matches are rare.
• Enriched fields accepted: The majority of inferred brand, GTIN, and attribute values are correct and accepted. Rejection rate is low, indicating good inference accuracy.
• Improved channel approval rates: Products with higher quality scores and correct categories are approved by channels more quickly and are less likely to be flagged for data issues.
• Better search and discovery: Correct categories and complete attributes mean products appear in the right search results and facets, leading to higher visibility and click-through rates.

Next steps

Once your product quality is solid, you are ready to distribute that data reliably to marketplaces and affiliate networks. See distributing-product-data-to-marketplaces-and-affiliate-networks for guidance on channel distribution. If you need to improve image quality or compliance, see product-image-hosting-and-channel-compliance-via-the-cdn. For more advanced data transformations, see feed-rules-filtering-and-transforming-products-at-export.

If you are importing data from a new source, importing-product-data-from-a-file-or-an-ecommerce-platform explains how to set up imports so quality assessment begins immediately.