Match Confidence Score

A match confidence score is a value indicating how certain a matching system is that two listings from different retailers describe the same or an equivalent product. Scores rise when strong evidence agrees, such as a shared GTIN alongside matching brand, model and pack size. They fall when signals conflict or go missing, which is common with private label goods, multipacks, regional variants and refurbished stock. The practical purpose of a score is triage. High-confidence matches flow into reporting automatically, low-confidence matches route to a person for review, and the threshold between the two is a business decision rather than a technical setting. A score without visible supporting evidence is hard to act on, since a reviewer needs to see which attributes agreed and which did not.

Why it matters

  • Makes uncertainty explicit, so a report distinguishes between a match that is solid and one that merely passed a threshold
  • Governs how much manual review a matching programme requires, which is usually the largest hidden cost in running one
  • Carries particular weight in MAP enforcement, where acting on a wrongly matched listing means contacting a retailer about a product that was never yours

How it is used

  • Threshold setting, where a team decides which score routes to automatic use and which routes to review
  • Review queue prioritisation, working from the lowest-confidence matches upward
  • Match quality reporting over time, tracking how confidence shifts as retailers change their listings

A confidence score describes certainty about a single match. It says nothing about how many products were matched overall, which is match rate or coverage, and a high average confidence across a small matched set can look healthier than it is. Confidence is also not accuracy. A score reflects how well the available evidence agrees, so a system can be confidently wrong when two different products share most of their attributes, which happens often with sequential model numbers and near-identical variants.

Learn how AI is transforming web data into enterprise intelligence.