The same product is listed under a different title, a different SKU and a different spec sheet on every site that sells it. ScrapeWise matches them, then delivers the matched rows to your pricing team — no matching model to build, train or maintain.
Most retailers do not publish a GTIN or EAN, and the ones that do often publish the wrong one. Matching on identifiers alone quietly misses most of the catalog.
Every retailer writes its own title. Pack size, colour name, model year and bundled accessories all move around, so exact string matching finds almost nothing.
A 500 g bag and a 2 kg bag share the title, the image and often the SKU stem. Match them together and your price comparison is wrong, not just incomplete.
Tools that return a match without a score hide their own uncertainty. You cannot review what you cannot see, so bad matches reach the pricing decision unchallenged.
Building matching in-house means someone owns the training data, the thresholds and the drift — permanently, not as a one-off project.
Per 1,000 delivered pages — matched and validated, no plan and no compute units
Ready-made scraper endpoints you can call directly for marketplace and retailer data
Free requests on every new account — test the match quality before you commit
We scrape the retailers and marketplaces you care about, capturing titles, attributes, images, identifiers, price and availability from each product page.
Listings are normalised and matched to your SKUs on several signals at once — text, structured attributes, images and identifiers — rather than a single brittle rule.
Matched pairs land in the portal and in your systems by REST API or CSV, on your schedule, with uncertain matches flagged for a human to confirm.
This is the actual sequence a single SKU goes through — WHEY-VAN-500G, a 500 g vanilla whey protein, against four competitor listings that all look like it.
"Whey Protein Isolat Vanille 500g Beutel", "Vanilla Whey 0,5 kg Pouch" and "Whey Protein Vanilla 2 kg Dose" are parsed into structured attributes: brand, flavour, net weight in grams, container type, language. 0,5 kg and 500 g become the same number. 2 kg does not.
Each listing is scored on the signals it actually publishes. The amazon.de listing carries a GTIN, an image and a full spec block, so it scores 0.98. The idealo.de listing publishes a title and nothing else, so it scores 0.56 — not because it is a bad match, but because there is nothing to confirm it with.
Two listings clear the threshold and are delivered as matches. The 2 kg pack is held as a separate variant rather than folded in. The title-only listing goes to review, where a person confirms or rejects it — before it can move a price.
Every matching approach that fails in production failed because it trusted one signal. These are the four we combine, and the specific blind spot each one has on its own.
The strongest signal when present and correct. The problem is coverage: most retailers publish no identifier at all, and a meaningful share of the ones that do publish the manufacturer's code for a different pack size or a superseded model. Identifier-only matching is high precision and low recall — it is right when it fires, and it rarely fires.
The only signal that is always there, and the least trustworthy. Retailers pad titles with bundled accessories, drop the pack size, translate the colour name, or keep last year's model number. Fuzzy string similarity alone will happily match a 500 g bag to a 2 kg tub because the strings are 90% identical.
Spec tables, JSON-LD blocks and breadcrumb categories give you net weight, colour, size, material and model year as actual fields. This is what separates variants. Its blind spot is availability — thin retailer pages publish nothing structured, which is exactly where the title-only cases come from.
Image similarity catches products whose titles share no useful words — a rebranded listing, a translated title, a marketplace seller writing their own description. It cannot separate variants: the 500 g bag and the 2 kg tub are routinely photographed with the identical manufacturer press shot.
A matcher that returns a match with no score is asserting certainty it does not have. Ours reports the score, and the borderline cases stop in a queue instead of flowing straight into a price change.
Multiple independent signals agree, including at least one that can separate variants. These rows go straight into your pricing system. In the worked example above, amazon.de at 0.98 and kaufland.de at 0.94.
The evidence points one way but nothing confirms the variant. The pair stays visible in the portal with both listings side by side, so a person confirms or rejects it in seconds. idealo.de at 0.56 — a title and nothing else.
The listing is clearly the same product line but a different pack size, colour or model year. It is not a match and it is not a rejection, so it is held as a separate variant rather than silently folded in. otto.de, 2 kg against your 500 g.
Folding variants together does not make your coverage look better. It makes your price comparison confidently wrong, which is worse than having no comparison at all.
A 2 kg tub at EUR 49.90 next to your 500 g bag at EUR 16.90 reads as a competitor undercutting you by 66% per kilo. Fold them together and your repricing rule chases a price that does not exist.
Colourways share the title, the spec table and often the SKU stem. Some sell at a premium, some at clearance. Matched as one product, the clearance colour drags your entire price position down.
Last year's model and this year's, or the body-only listing and the one with the accessory kit, are different products at different prices. Treated as one, you are benchmarking against a discontinued SKU.
You supply your catalog and the sites you want matched against it. You receive rows that are already linked to your SKUs.
SKU, title, brand and whatever attributes you already hold. CSV upload or API push.
WHEY-VAN-500G, Whey Protein Vanilla 500 g, BrandCoRetailers and marketplaces by domain. Category pages, search pages or a URL list.
amazon.dekaufland.deidealo.deotto.deGTIN, EAN or MPN improve precision where they exist. Matching does not depend on them.
4012345678901On by default. Pack sizes, colourways and model years are never folded into one match.
Matches are re-checked on your cadence, so retailer title changes do not quietly rot yesterday's matches.
11 columns each
Delivered by REST API, CSV or scheduled export. Every row carries its confidence score and the signals that produced it.
One row per confirmed competitor listing, keyed to your SKU. This is the WHEY-VAN-500G example as it lands in your system.
| your_sku | competitor_url | matched_title | confidence | variant_status | price | availability |
|---|---|---|---|---|---|---|
| WHEY-VAN-500G | amazon.de/dp/... | Whey Protein Isolat Vanille 500g Beutel | 0.98 | exact | EUR 18.49 | in stock |
| WHEY-VAN-500G | kaufland.de/product/... | Vanilla Whey 0,5 kg Pouch | 0.94 | exact | EUR 17.95 | in stock |
| WHEY-VAN-500G | otto.de/p/... | Whey Protein Vanilla 2 kg Dose | 0.71 | variant_held | EUR 49.90 | in stock |
| WHEY-VAN-500G | idealo.de/offer/... | Vanilla Whey Protein | 0.56 | in_review | EUR 16.20 | unknown |
You receive the competitor product already linked to yours, with price and availability attached. No parsing step, no reconciliation step, no spreadsheet in the middle.
Matches carry a confidence signal and stay visible in the portal, so your team can confirm the uncertain ones instead of discovering them through a mispriced SKU.
We own the scrapers, the proxies, the anti-bot handling and the matching itself. When a retailer changes its layout or its titles, that is our problem, not your sprint.
If your pricing team is comparing the wrong products, the pricing decision is wrong before anyone looks at it. Get matched competitor rows delivered instead — no plan, no compute units, just a balance you top up.
How product matching works at ScrapeWise, what it costs, and when you should build it yourself instead.
Product matching software identifies when two listings on different websites are the same physical product, even though the title, SKU, spec sheet and images all differ. It is the step that makes competitor price comparison possible: without it you are comparing your product against something that merely looks similar. Good matching combines several signals — text, structured attributes, images and identifiers such as GTIN or EAN — because any one of them on its own misses too much of the catalog.