Product Match Confidence Scorer

Paste two product records — a supplier line and a catalog line, or two rows you suspect are the same SKU — and this tool returns a single product matching confidence score from 0 to 100, plus a plain-English verdict on whether they describe the same physical product. It is built for anyone reconciling catalogs without a data team: distributors, retail and marketplace teams, and API-first platforms.

What it checks

The scorer compares the two records field by field, then blends the signals into one weighted confidence number. It evaluates:

• Identifiers — exact and normalized matches on GTIN/EAN/UPC, MPN, and SKU, which carry the highest weight when present and valid.
• Brand and manufacturer — token-level similarity that tolerates abbreviations (for example “Mfg Co.” vs “Manufacturing Company”).
• Product name and description — fuzzy string similarity using edit-distance and token-set comparison, so reordered or partial titles still match.
• Key attributes — size, color, voltage, material, pack quantity, and unit of measure, with mismatches on hard specs penalized heavily.
• Numeric and unit normalization — values expressed in different units (for example 0.75 in vs 19 mm, or a 12-pack vs “case of 12”) are reconciled before comparison.
• Conflict flags — contradictions such as matching MPNs but incompatible voltages, which lower the score and are surfaced as warnings.

How it works

Matching real-world product data is rarely a clean equality check. Two suppliers describe the same bearing, the same case of cleaning chemical, or the same office chair with different word order, abbreviations, and unit conventions. This is why catalog reconciliation relies on probabilistic matching rather than exact joins.

The scorer applies the same logic production matching engines use:

1. 1
   Normalize
   Both records are lowercased, de-punctuated, and unit-normalized so superficial formatting differences do not count as conflicts.
2. 2
   Compare per field
   Each field is scored independently — identifiers via exact/normalized equality, text via fuzzy similarity, specs via tolerant numeric comparison.
3. 3
   Weight and blend
   Field scores are combined with weights that reflect how decisive each signal is. A valid shared GTIN counts for far more than a similar description.
4. 4
   Verdict
   The blended 0–100 score maps to a plain-language band — likely match, needs review, or unlikely — with the reasons that drove it.

The bands above are illustrative starting points; the right thresholds depend on your tolerance for false merges versus manual review, which the auto-merge playbook below walks through.

All processing happens client-side in your browser. Nothing is uploaded, stored, or sent to a server — you can run it on confidential supplier price files or unreleased CPG and furniture lines without a data-handling review.

Related resources

FAQ