Guides · 10 min read

Best product matching tools and APIs (2026), with measured accuracy

Most matching vendors say "95%+ accuracy" and don't say how they measured it. We measured ours on hand-checked pairs from five UK stores and published the method. Without a shared EAN, 313 of 314 matches were right.

Product matching finds the same product in different online stores, so you can compare prices, find gaps in your assortment, or merge catalogues. It sounds simple until you try it: stores write names differently, many don't publish EAN codes, and the ones that do sometimes publish the wrong one.

Most vendors in this space quote an accuracy of "95%" or higher. Few say how they measured it. We measured ours on real store data, checked the pairs by hand, and put the method and the numbers below.

The short answer

  • EAN codes alone aren't enough. In our data, 12% of pairs where both stores publish an EAN had different EANs for the same product. Stores also reuse one EAN for different colours, and put a product's EAN on bundles that include it.
  • Matching without codes works when it checks the details. Where the two stores didn't share an EAN, our matching was right on 313 of 314 checked pairs, including finishes, colours and model generations.
  • A shared EAN isn't proof of the same offer. A piano with a stool and headphones can carry the same EAN as the piano alone. If you compare prices, check bundles and multi-packs separately.
  • Ask vendors how they measured. A percentage without a sample, a method, and a rule for bundles doesn't tell you much.
  • Test on your own stores before you pay. A free or cheap sample on your categories tells you more than any claimed percentage, including ours.

Want to see matches on your stores?

Add two stores in the console, or ask Claude through our MCP server. The first matching sample of a project is free, and you check the pairs yourself before you pay.

Product matching tools at a glance

These tools don't all do the same job. Some scrape any store you name, some cover marketplaces and Google Shopping, one matches against retailers it already crawls, and one only matches data you bring. Most match competitors against your own catalogue. DeepScout matches all the stores in a project against each other, so you also see what your competitors share. Check the "Which stores" and "How pairs are made" columns first: they decide whether a tool fits before accuracy does.

Accuracy is what each vendor publishes, unless marked "measured". No vendor below publishes how its figure was measured. API prices are for matching only; scraping is billed on top. Details for each tool are below the table.

Product matching options, November 2026

ToolWhich storesHow pairs are madeAccuracyPrice
DeepScoutAny store, also competitor vs competitorCandidates verified by AIMeasured: 313 of 314 without a shared EAN€7 / 1k products, AI check included. Free sample
ScrapewiseAny store + marketplacesEAN, then text; unsure pairs reviewed by handNone published€0.15 / 1k text, €1.50 / 1k pictures, + server time
Apify AI Product MatcherOnly data you bringML model on pairsNone published$10 / 1k pairs checked
IntelisGoogle Shopping + URLsAI, barcode, manualNone published$49 / mo for 100 products
PrisyncURLs you add + marketplacesBarcodes or URLs you addNone published$99 / mo for 100 products
Price2SpyURLs + marketplacesEAN, ML + humans, or manual99.5–99.8%Not public
Intelligence NodeOwn index + your URLsNot published≤ 1 wrong per 100On request
DataWeaveNot statedAI + human validation99%+ (AI alone ~80%)On request
CompeteraSites you choose + marketplacesAI + human reviewUp to 99%On request
SkuuudleSites you choose (managed)Two analysts per matchNone published$10k+ / year
ProWebScraperSites you choose (managed)Their team (managed)95%+ after tuningAfter a pilot

How product matching works

There are three ways to decide that two listings are the same product. Most tools combine them.

1. Codes: EAN, GTIN, UPC, MPN. If both stores publish the same barcode, it's a strong signal and costs nothing to compute. The catch is coverage and quality:

  • Many stores don't publish an EAN at all. In our UK hi-fi test, one of three stores published none.
  • The same product can have different codes: a US UPC in one store, a European EAN in another, or a newer code after the manufacturer reissued it.
  • Stores make mistakes. One store listed a Yamaha digital piano with the code 9457812717175. Yamaha's codes start with 4957812.
  • Stores reuse codes. One shop used the same EAN for the black and the white version of a speaker. Piano shops put the piano's EAN on bundles with a stool and headphones.

2. Names and attributes. Compare the brand, model number, colour, size, capacity and pack size written in the name and the product details. This finds matches without codes, but naive fuzzy matching confuses near-identical names: "HP704 White" and "HP704 White Value Package", or "B20 SC3 PE" and "B20 SC3 PEC" (polished ebony vs polished ebony with chrome fittings).

3. Images and AI. Modern matchers use embeddings to find candidates and a language or vision model to verify them: is this the same model, the same variant, the same pack? This is what lets a matcher handle names in different languages or formats. It is also slower and costs more per pair, so good systems use it only on uncertain pairs.

The hard part isn't finding candidates. It's rejecting the near-misses: the right model in the wrong colour, the previous generation, a pair instead of a single speaker, a bundle instead of the product.

We measured our accuracy: method

We ran DeepScout matching on two sets of UK stores in November 2026 and checked the results by hand. We report the pairs where the two stores don't share an EAN: either one store publishes none, or the codes differ. These are the matches a simple EAN lookup can't make, so they show what a matcher adds.

The two test runs

SetupRun 1: hi-fiRun 2: pianos and keyboards
StoresRicher Sounds, Peter Tyson, Sevenoaks Sound & VisionBonners Music, Rimmers Music
Products2,206 (14 brands that all three stores sell)4,222 (full catalogues)
EANsTwo stores publish them, Sevenoaks publishes noneBoth publish them on most products
Matched pairs without a shared EAN77119 (free sample: 691 random Bonners products against all of Rimmers)
Of those, checked by hand295 (all 45 with different EANs, 250 of 726 with no EAN, at random)19 (all)

How we checked:

  • Rules set before we looked at results. A pair is correct when it is the same brand, model, generation, colour or finish, and quantity. A bundle and a single item, or a pair and a single speaker, are different products. If only one store names a colour, the pair isn't counted wrong for colour.
  • Every pair checked against the product data and, where unclear, the live product page.
  • Recall too. For 100 random Richer Sounds products, we searched the other two stores' full exports for the same product and counted how many the matcher found.

Every checked pair, its label, the rules and a scoring script are on GitHub. If you run a matching tool, test it on the same products and open an issue there with your results. We'll check them with the same rules and publish them next to ours.

Measured precision without a shared EAN

Share of matched pairs that are correct, DeepScout, November 2026

PairsRun 1: hi-fiRun 2: pianosBoth runs
EANs on both stores, but different45 / 454 / 449 / 49
No EAN on one store249 / 25015 / 15264 / 265
All without a shared EAN294 / 29519 / 19313 / 314

Run 1 had 726 pairs with no EAN on one store, so we checked a random 250 of them. All other groups were checked in full. The one wrong pair: a pair of Sonos Five speakers (£878) matched to a single Sonos Five (£549).

What a match looks like

One of the checked pairs: a row of the matches export, joined with both stores' product exports. Sevenoaks publishes no EAN and calls the finish Macassar. Peter Tyson calls it Wood. Macassar is a wood veneer, so it's the same speaker.

What you can do with it

Both stores sell it at £1,699. Across thousands of pairs, this is your price position.

matches-joined.json
{
"store": "petertyson.co.uk",
"name": "Monitor Audio Gold 50 6G Bookshelf Speakers - Wood",
"variant": "Wood",
"price": 1699.0,
"currency": "GBP",
"ean": "5060565778863",
"availability": "Stock due 3 to 5 days",
"product_url": "https://petertyson.co.uk/monitor-audio-gold-50-6g-wood",
"match": {
"store": "sevenoakssoundandvision.co.uk",
"name": "Monitor Audio Gold 50 6G Bookshelf Speakers Macassar",
"variant": "Macassar",
"price": 1699.0,
"currency": "GBP",
"ean": null,
"availability": "Out of Stock Online",
"product_url": "https://www.sevenoakssoundandvision.co.uk/p-68284-monitor-audio-gold-50-6g-bookshelf-speakers.aspx"
}
}

What the data tells you about any matcher

  • Codes help less than you'd think. In run 1, 45 of the 366 pairs where both stores published an EAN had different codes for the same product. A matcher that only uses EANs misses all of them.
  • A shared EAN can still be wrong. One store used the same EAN for two colours of a speaker. Piano stores put the piano's EAN on bundles with a stool and headphones. If you compare prices, a bundle matched to the plain product shows a price gap that is really the accessories.
  • Accuracy depends on the category. Hi-fi stores rarely bundle; piano stores bundle almost everything. A single percentage for "all categories" hides this, whoever quotes it.
  • Look at the pairs, not the percentage. Twenty minutes with a sample of pairs from your own stores shows which mistakes a tool makes and whether they matter to you.

Recall: how many matches does it find?

Precision tells you whether the matches are right. Recall tells you how many real matches the tool found. For 100 random Richer Sounds products, we searched the other two stores for the same product by brand and model and checked every candidate.

  • Peter Tyson: the matcher found 60 of 61 real matches.
  • Sevenoaks: 51 of 52.
  • Combined: about 98%. The one product missed in both stores was the Sonos Five pair.

Two more cases come from one-to-one matching: Sevenoaks has one page for all colours of a product, which can link to only one Richer Sounds colour. Counting those as misses, recall is 96.5%.

What the piano test cost

Run 2: two full stores, 4,222 products, free matching sample

Store analyses2 stores, €0.50 each (the first 10 each month are free)€1.00
Scrape Rimmers Music763 products, easy store, €0.40 per 1,000€0.31
Scrape Bonners Music3,459 products, medium store, €0.70 per 1,000€2.42
Matching sample691 Bonners products against the full Rimmers catalogue€0.00
TotalWhat we paid for run 2€3.73
Optional: full matchingAll 4,222 products at €7 per 1,000, instead of the free sample+€29.55

Tools by category

A note on vendor percentages. Most vendors quote precision on matches their staff have already reviewed, and none publishes the sample or the method. Academic benchmarks measure something harder: F1 (precision and recall together) on test sets full of near-identical products. On the WDC Products benchmark from the University of Mannheim, the best systems score an F1 between 0.64 and 0.89 (Peeters et al., 2024). Zero-shot GPT-4 reached 0.90 on the same benchmark (Peeters et al., 2025). A vendor's "99%" and an academic "0.89" aren't the same measure, so don't compare them directly.

Matching APIs for developers

You send or scrape the product data and get matched pairs back as data. Best when you build your own pricing tool, feed or analysis.

  • DeepScout. Scrapes the stores and matches them in one run, through an API, an MCP server for Claude, or a console. It finds candidate pairs, and an AI model verifies them. Matching costs €7 per 1,000 products for the first 50,000 (€3 from 50,000 to 250,000, €2 above), and the first sample in each project is free. Scraping costs €0.40–1.20 per 1,000 products, depending on how hard the store is. Our two test runs above cost €15.44 and €3.73.
  • Scrapewise. The closest to DeepScout: scraping plus matching, prepaid, with an MCP server. Matches by EAN, then text, then images, and puts unsure pairs in a review queue for a person to approve. Text and EAN matching costs €0.15 per 1,000 products plus server time (at least one hour per run, from about €0.11 an hour); picture matching is €1.50 per 1,000 pictures. The prices aren't like for like: ours includes an AI model that verifies the candidate pairs, theirs covers EAN and text matching, with unsure pairs left for a person to review. No accuracy figure is published, so test both on the same stores.
  • Apify AI Product Matcher. An actor on the Apify store that matches two datasets you've already scraped, with a model trained on labelled product pairs. It costs $10 per 1,000 product pairs checked, matched or not, so the bill depends on how many candidate pairs it checks, not on how many products you have. Scraping is billed separately. Its page says that in the recall-focused mode about half the suggested pairs were wrong in their tests. No other accuracy figure is published.

Price-monitoring SaaS with matching built in

You get dashboards, alerts and repricing rules, and matching is part of the setup. Best when the goal is pricing decisions, not data.

  • Intelis. Shopify-friendly price monitoring and repricing. AI, barcode and manual matching. From $49/month for 100 products; API from $249/month.
  • Prisync. Matches on the barcodes you provide, or on competitor URLs you add yourself. From $99/month for 100 products; API access costs 20% extra.
  • Price2Spy. Automatch on EANs, hybrid ML with human checks, or manual matching. Claims 99.5% (automatch) and 99.8% (hybrid and manual). Prices are no longer public.

Enterprise platforms and managed services

Sales-led, with onboarding measured in weeks and humans checking matches. Best for large retailers who want the matching done for them.

  • Intelligence Node. The most specific claim in the market: at most 1 wrong match per 100 exact matches, and at most 2 missed per 100 unmatched products. No method published.
  • DataWeave. Claims "99%+". Its own blog says AI alone gets "around 80%" before human validation, which is an honest number and close to the academic results.
  • Competera, Minderest, Omnia Retail, Dealavo. Pricing platforms with AI matching and human review. Claims range from "up to 99%" (Competera, Dealavo) and ">99% certified precision" in one Minderest case study, to none at all (Omnia).
  • Skuuudle. Fully managed: two analysts sign off every match. From about $10,000 a year.
  • ProWebScraper. Managed scraping and matching. Claims 95%+ after tuning, priced per site after a pilot.

Build it yourself

  • EAN databases. Verified by GS1 tells you who registered a code and what it is, with a free lookup limit of 30 queries a day. UPCitemdb and Open Food Facts (food only) have APIs. They help you identify products, but they can't match stores that don't publish codes.
  • Embeddings + an LLM check. The usual DIY pipeline: normalise the data, find candidates with embeddings or search, ask a model whether each candidate is the same product, and review the uncertain ones. It works, and it's how most of the tools above work inside. The cost is the engineering: scraping the stores, keeping the data fresh, and handling bundles, variants and wrong codes. LLM checks cost about half a cent per pair at 2024 GPT-4 prices.

When to pick something else

  • You only need price monitoring with a dashboard and alerts for a few hundred products. A price-intelligence SaaS gives you repricing rules and reports. We give you the data and the matches through an API.
  • All your products and your competitors' have reliable EANs. Match on the code yourself and check bundles. You don't need AI for that.
  • You need a human to approve every match. Some vendors include manual review in the price. Our matches come with the product data so you can review them, but we don't review them for you.
  • You need marketplaces. Amazon, eBay and Allegro aren't supported.

How to test a matching vendor yourself

  1. 1
    Run a sampleUse your own stores.Pick two or three stores you compete with, including one that doesn't publish EANs. Run the vendor's sample or trial and ask for the matched pairs as a file, not a dashboard.
  2. 2
    Check by handSplit the pairs by EAN.Same EAN, different EAN, no EAN on one side. Check 50–100 pairs per group against rules you write down first: colour, generation, pack size, bundles.
  3. 3
    Check recallCount what it missed.Take 50 of your products, find them in the competitor's store yourself, and count how many the vendor found. Ask whether its accuracy figure counts bundles as matches.

Try it on your stores

Create a project, add your store and a competitor, and run the scrapes. Then start the free matching sample and check the pairs yourself.

Open the console

Questions

What is a good accuracy for product matching?

For price comparison, aim for at least 95% of matches correct, measured on your own categories, with bundles and multi-packs counted as wrong when they're matched to a single product. Look at what the wrong ones are: a wrong colour can matter less than a bundle compared with a single product.

Can I match products without EAN codes?

Yes. Matching by brand, model, attributes and images works without codes. In our tests, pairs without a shared EAN were right 313 times out of 314. Codes help, but they're missing or wrong often enough that a matcher that only uses them misses a lot.

Why do some stores have different EANs for the same product?

Regional codes (a US UPC vs a European EAN), reissued codes, and plain mistakes. In our hi-fi test, 45 of 366 pairs where both stores published an EAN had different codes for the same product.

How much does product matching cost?

SaaS tools usually charge a monthly fee by the number of products you track. APIs charge per product matched. Self-serve price-monitoring tools start at $49–99 a month for 100 products. Matching APIs charge per product or per pair: €0.15 per 1,000 products for text and EAN matching only (pictures and server time cost extra), €7 per 1,000 with an AI check included, or $10 per 1,000 pairs checked. Enterprise and managed services start in the tens of thousands a year. On DeepScout, matching costs €7 per 1,000 products, and the first sample in each project is free.

How long does matching take?

On DeepScout, about an hour for 1,000 products and 4 to 15 hours for 10,000, after the stores are scraped. Our piano sample of 1,454 products took under 40 minutes.

Is DeepScout's matching always right?

No. It got 1 of 314 pairs wrong in our test, and results vary by category. Test it on your own stores with the free sample before you rely on it.

Get started with the API

Scrape any e-shop into clean product JSON, then match it against the shops you pick. No EAN or barcode required. Usage-based, from €0.40 per 1,000 products.

To make it all work

We use marketing and analytics cookies to learn how you use the product. That helps us improve it and show you relevant ads.