Product Matching That Makes Every Comparison Trustworthy

Product matching identifies and classifies relationships between product records across catalogs, retailers, marketplaces, sellers, or internal systems. Relationships may be exact, variant, pack equivalent, comparable, substitute, private label, duplicate, rejected, or unmatched.

Catalogs are profiled and normalized, plausible candidate pairs are generated, identifiers and product signals are scored, a relationship type is assigned, confidence controls are applied, ambiguous pairs are reviewed, and approved results are delivered and maintained.

A retailer listing titled “X2 Wireless Buds, White” may be matched with an approved “Wireless Earbuds X2” record when the GTIN, brand, normalized model, color, specifications, and pack quantity agree.

An exact match represents the same underlying product. Similar matching covers defined relationships such as variants, comparable products, substitutes, or private-label equivalents. Similar products should not automatically be used for exact price comparison.

Yes. Inferred matching can combine brand, manufacturer, model, title, description, attributes, specifications, pack data, taxonomy, imagery, and contextual checks. Missing identifiers usually increase uncertainty and review requirements.

The fields depend on the category and may include model, size, color, flavor, capacity, material, dimensions, technical specifications, formula, generation, pack count, weight, volume, and other category-defining properties.

Images can provide supporting evidence through product, packaging, shape, label, and visual-feature similarity. They should be combined with other signals because images may be reused, outdated, incomplete, or visually similar across different products.

Pack count, unit quantity, weight, volume, and multipack structure are parsed and normalized. One 24-pack may be identified as pack-equivalent to two 12-packs while preserving that the listed pack structures differ.

Products from the same family can be linked while retaining variant fields such as size, color, flavor, capacity, model generation, or configuration. Variant relationships remain separate from exact-SKU relationships.

Yes, when a category-specific comparability framework is defined. Private-label products should be labeled as comparable or equivalent for the intended analysis, not misrepresented as identical branded products.

A confidence score summarizes supporting signals, conflicts, source quality, category rules, and model or ruleset outputs. It should map to an operational band such as verified, high confidence, review required, rejected, or unmatched.

Evaluation should include precision, recall, false-positive rate, false-negative rate, coverage, review rate, and confidence calibration. Results should be segmented by relationship type, category, source, identifier availability, and confidence band.

The pair can be routed to a review queue with side-by-side records, supporting signals, conflicting fields, suggested relationship type, and downstream impact. It should not be silently accepted.

Yes. Human review can support ambiguous, conflicting, high-value, private-label, pack-complexity, and category-specific cases. Review scope is designed around risk and required precision.

Multilingual matching can use translated or normalized text, identifiers, brands, models, attributes, specifications, taxonomy, images, and market-specific rules. Feasibility depends on source quality and category complexity.

Yes. Product Matching establishes which products and pack quantities are valid for exact or comparable price analysis. Match relationship, confidence, and comparison rules should remain attached to the price record.

Delivery can include CSV, Excel, JSON, APIs, webhooks, databases, cloud storage, warehouses, BI platforms, or integration with approved internal product systems.

Share the source and target catalogs, category, record counts, identifiers, desired relationship types, downstream use case, required precision, review expectations, and preferred output format. Kvetoiq will profile the data before defining the production workflow.