- Product
- Solution for
For Your Industry
- Partners
- Plans & Pricing
- Resources
Product matching is the process of identifying that a product in your catalog and a product on a competitor’s website are the same item, so their prices can be compared. Price monitoring platforms match products using unique identifiers such as EAN, GTIN, and MPN, seller SKUs, structured attributes, and AI models that compare titles, specifications, and images. Match quality determines the accuracy of everything downstream: if the match is wrong, every price comparison, alert, and repricing decision built on it is wrong too.
Product matching (also called product mapping or product linking) is the process of connecting each product in your catalog to the corresponding product listings on competitor websites and marketplaces. The output is a set of verified links: your SKU on one side, and every competitor URL selling the identical item on the other.
Once those links exist, price monitoring becomes simple arithmetic: collect the competitor prices behind each link, compare them to yours, and calculate your price position. Without correct links, the arithmetic is still simple, but it is being performed on the wrong numbers.
Match quality decides every pricing decision because match errors propagate silently through the entire workflow. A wrong price is visible. A wrong match looks exactly like a correct one until someone checks the competitor URL.
A consumer electronics retailer may see a competitor selling “their” smartphone 20% cheaper. If the matched listing is actually the 128GB variant compared against their 256GB model, the team will believe they are overpriced on that product permanently, and may cut margin for no reason.
Typical damage from poor matching includes:
1. Unique identifier matching: EAN, GTIN, UPC, MPN
Global identifiers such as EAN and the broader GTIN family are assigned per product variant, so two listings sharing an EAN are, in principle, the same item. This is the gold standard: near-certain precision when the identifier is present and correct on both sides.
In practice, coverage is the limitation. Many retailers do not publish EANs, marketplace sellers enter wrong or recycled codes, bundles and retailer-exclusive variants carry their own identifiers, and private label products have codes that exist nowhere else.
2. SKU and MPN matching
Manufacturer part numbers frequently appear in competitor titles, spec tables, and URLs. Matching engines extract these codes and combine them with brand verification to avoid collisions between manufacturers with similar code formats. Precision is very high in code-driven categories such as electronics, appliances, DIY, and automotive, and near zero in categories where codes are never displayed, such as much of fashion and beauty.
3. Attribute and title matching
When identifiers are missing, the engine compares what the listings say: brand, model, size, color, capacity, and pack quantity, extracted from titles and specifications. A good attribute matcher normalizes units (1000ml vs 1L), works across languages, and treats variant-defining attributes as hard constraints. A naive one matches “iPhone 15 case” to “iPhone 15” because the words overlap.
4. AI and image matching
Machine learning models trained on millions of confirmed matches learn which combinations of textual and visual signals indicate identity rather than similarity. Image models are especially powerful in fashion and furniture, where photos carry more identity information than titles. AI produces probabilities, not certainties, which is why responsible platforms route low-confidence matches to human validation.
A platform that quotes only “AI-powered matching” without identifiers, extraction, and validation behind it is describing a guess, not a method.
Exact matching links identical items and supports one-to-one price comparison. Similar product matching finds the closest comparable competitor item when no identical product exists, and supports positioning decisions rather than direct comparison.
A furniture retailer with a private label sofa has no EAN twin anywhere in the market. The relevant question is not “who sells my sofa cheaper” but “what do comparable sofas with equivalent dimensions, materials, and quality tier cost at my competitors.”
Similar matching matters most for:
The critical discipline is separation: an exact match can drive an automated repricing rule, while a similar match should inform decisions made by a human. Platforms that blur the two feed similarity-based prices into workflows built for identity-based prices, and margin damage follows.
Teams should measure matching quality with two numbers together: match rate, the share of the catalog linked to at least one competitor listing, and match precision, the share of created links that are actually correct.
A sporting goods retailer evaluating two vendors may see one promise 95% coverage and the other 85%. The first number is meaningless alone: any vendor can inflate coverage by accepting looser matches, which silently destroys precision.
Questions that expose real matching quality:
A vendor quoting only an impressive coverage number is telling you half the story.
Product matching is never finished because the market keeps changing underneath the links. Matching is a continuous process, not a one-time setup task.
A fashion retailer refreshing 40% of its assortment every season needs thousands of new matches each quarter just to keep monitoring coverage stable, before counting competitor catalog changes.
What breaks matches over time:
When a vendor demo shows perfect matches, the real question is what the same data looks like six months later.
A consumer electronics retailer starts monitoring 15 competitor webshops and 2 marketplaces for a 10,000 SKU catalog. Here is how layered matching typically plays out.
Identifier matching links the majority of branded products quickly, because electronics listings usually expose EANs or MPNs. SKU and MPN extraction covers listings where the identifier is buried in titles or spec tables. Attribute and AI matching pick up the remainder: listings with missing codes, translated titles on foreign marketplaces, and accessories with messy naming.
High-value verification cases the process must catch:
Each of these, matched wrongly, would generate a false undercut alert or a wrong repricing trigger on a high-visibility product. This is why low-confidence matches go to human validation instead of straight into the data.
Product matching is the process of identifying that a product in your catalog and a listing on a competitor website are the same item, so their prices can be compared accurately. Every comparison, alert, and repricing decision depends on the match being correct.
EAN matching links products across websites using the European Article Number, a unique identifier assigned per product variant. When both sides expose a correct EAN, the match is near-certain. Many listings lack clean identifiers, so EAN matching is combined with SKU, attribute, and AI matching for full coverage.
Similar product matching identifies the closest comparable competitor item when no identical product exists, for example for private label products. It supports positioning and assortment decisions, and should be kept separate from exact matching rather than driving automated repricing on its own.
AI matching can achieve high accuracy at scale, but it produces confidence scores rather than certainties. Reliable setups pair AI with identifier matching where possible and route low-confidence matches to human validation instead of accepting every algorithmic match automatically.
It depends on category and market: identifier-rich categories like electronics support higher exact match rates than fashion or private label ranges. What matters is the combination of coverage and precision, and a vendor’s willingness to commit to both and let you audit the links.
The most common cause is match errors: a variant, bundle, multipack, or different model matched as if it were your product. Check the competitor URL behind the suspicious price. If comparisons are frequently wrong, the matching process behind your data is usually the problem.
Missing an important marketplace?
Send us your request to add it!