Match confidence

Product-match confidence checklist

Knowing a competitor price is useful only after the offer has been matched to the right product and commercial basis. A correctly captured number attached to the wrong capacity, suffix, bundle or seller is a confident error.

15 September 2026

7 min read

By Alexey Skryabin

Match-control method

Two similar component records compared with technical drawings and a magnifier

Check product identity first

  • Brand and exact model or MPN.
  • GTIN, EAN or UPC where present.
  • Model suffix and intended market version.
  • Product family versus the exact sellable variant.
  • Every conflict or missing decisive identifier.
A shared family name is not an exact match. A missing GTIN increases the evidence required from model and attributes.

Verify the decisive technical variant

For HVAC and appliances, confirm capacity/output and test basis, performance variant, power supply and market, refrigerant/version where relevant, controller/connectivity package, and every commercially distinct size, colour or configuration.

Separate bundle, condition and offer identity

  • Complete system versus partial unit.
  • Installation, delivery, accessories or service inclusion.
  • Pack quantity, conditional rebate or promotional gift.
  • New, refurbished, used or clearance condition.
  • Seller, fulfilment, tax, delivery, membership, coupon and basket conditions.

Use five confidence classes

  • Confirmed exact: decisive identifiers and variant agree; no material conflict remains.
  • Probable exact — review: strong agreement, but one non-material field needs human confirmation.
  • Comparable alternative: intentionally non-identical and used only for a stated benchmark dimension.
  • Conflict: a decisive field disagrees; block exact price and availability comparison.
  • Insufficient evidence: identity cannot be established; retain only in the source-gap log.

When the match must fail

A trustworthy method does not force every listing into a match. It records conflicts, separates comparable alternatives and blocks downstream exception logic when identity is not established.

Do not publish a match-accuracy percentage until there is a labelled evaluation set, error taxonomy, representative sample and review of false positives as well as false negatives. Vendor claims are not evidence of Pulse performance.

Bound the signal question

Start with one market, a small model set and one decision.

Share the sources, products and commercial question. We will assess whether a managed 30-day Pulse pilot can produce a useful evidence brief inside the published boundary.

With permission, Google Analytics and Yandex Metrica help us understand page use and enquiry paths. No advertising pixels or session replay are enabled. You can continue without analytics.

Read the privacy details