Product Matching for Price Monitoring: Exact vs. Similar, and How to Measure Accuracy

Product matching pairs each of your SKUs with the same product on a competitor's site, and every competitor price you act on depends on it. Compare prices only on exact matches, keep variant and similar matches in separate groups for assortment and positioning work, and measure accuracy with two numbers: precision and recall, scored against a sample you labeled yourself. This guide explains the match types, how vendors define their accuracy figures, and how to test product matching software or a managed service on your own catalog.

Ficstar has collected web data since 2005 and processes more than 1 billion product prices every month. Product matching is part of how we deliver competitor price data, and this guide also covers the cases where another option fits better.

What product matching does in price monitoring

A competitor price is only useful when it sits next to the right product of yours. Matching links your SKU to the competitor listing that sells the same item, or a close substitute, so that a price gap means something.

Barcodes help when both sides have them. A GTIN, also known as an EAN or UPC, is a globally unique product number, and GS1 UK describes it as the number under a barcode. In practice many listings carry no GTIN, and Actowiz notes that marketplace sellers often enter them incorrectly or reuse one across variants. Private-label products share no identifier with the branded items they compete against. DataWeave makes the same point: barcodes only take you part of the way.

Matching therefore has to use brand, model, specifications, titles, images and human review as well as identifiers.

Exact, variant and similar matches

Vendors use different labels. Intelligence Node offers exact, similar and variant matches, and DataWeave describes identical, similar and private-label matching. What matters for pricing is which matches are safe to compare directly.

Match type	What it means	Use it for	Avoid using it for
Exact	Same product, same variant, same pack size	Price comparison, MAP checks, inputs to repricing rules	Nothing specific; this is the pricing-ready set
Variant	Same product family, different size, colour, pack or model year	Coverage checks and context for price gaps	Direct price comparison, unless normalized per unit
Similar, including private label	A comparable substitute from another brand or a store brand	Positioning, assortment gaps, private-label benchmarking	Saying you are over or under a competitor on the same product
No match	No valid equivalent found	Knowing a competitor does not carry the item	Concluding a gap exists without checking recall

Intelligence Node's own example is two ketchup bottles on Amazon from different national brands, both 20 oz squeeze bottles in a pack of 6. Their volume, product type, ingredients and flavor line up, so they count as similar matches. That is useful for assortment analysis and misleading as a price comparison.

Variants cause the most quiet errors. In tire pricing, one tire model can split into dozens of real SKUs by size, load and speed rating, and run-flat or OE codes. Matching at the model level while pricing at the SKU level produces wrong comparisons by construction. In our tire pricing case study, we monitored 20 competitors and more than 50,000 SKUs, with about 1 million pricing rows per weekly crawl.

Why a wrong match costs more than a missing one

A missing match leaves a visible gap. A wrong match produces a clean-looking price that is not comparable, and teams act on it. Typical causes are a different model year, a different size or colour, a bundle matched to a standalone item, or a 10-pack matched to a 20-pack. The results are price cuts you did not need and, in MAP work, enforcement notices about products that are not the same, which Actowiz also lists among the consequences.

Scale makes this worse. A 95% accurate match set sounds high, but on 10,000 matched SKUs it leaves about 500 wrong comparisons, and they repeat on every delivery until someone finds them.

How to measure product matching accuracy

Matching accuracy is not one number. Two measures describe it:

• Precision: of the matches returned, the share that are correct. Low precision means false matches and bad pricing decisions.

• Recall: of the matches that exist, the share the system found. Low recall means competitors you cannot see.

• 
  

Which one to favor depends on the use. Pricing and MAP enforcement need high precision, because acting on a wrong match costs money and retailer relationships. Market research can accept lower precision in exchange for wider coverage.

What a vendor's accuracy number means

Vendors publish figures such as 99%, and the definitions differ. Intelligence Node's published SLA shows what a precise definition looks like: if 100 SKUs are matched, at most one can be a false positive, and this applies to exact matches. A separate clause caps false negatives among unmatched SKUs. Competera reports an accuracy of up to 98% from a combination of manual and automatic matching. Both figures are vendor-reported, and they are not directly comparable.

Ask every vendor, including us, these questions:

1. Does the figure cover precision, recall or both?

2. Does it apply to exact matches only, or to variant and similar matches too?

3. What happens to products that cannot be matched automatically: human review, or they are dropped?

4. Was it measured on your categories or on a demonstration set?

5. How often are matches re-checked as catalogs change?

Run your own test

1. Pull a sample of a few hundred SKUs from your catalog. Weight it toward hard cases: variants, bundles, multi-packs, private label and products without barcodes.

2. Label the correct competitor match for each SKU yourself, before you see any vendor output. This is your ground truth.

3. Give every shortlisted provider the same sample and the same competitor sites.

4. Score each result. Precision is correct matches divided by matches returned. Recall is correct matches divided by the matches that exist in your ground truth. Score exact, variant and similar matches separately.

5. Check what happened to the SKUs the provider did not match, and repeat the test a few weeks later to see how it handles catalog changes.

We describe the same approach, a proof-of-data test instead of a software demo, in our guide to price intelligence services.

A demonstration set will not predict results on your catalog. The WDC Products benchmark from the University of Mannheim uses 11,715 product offers from 3,259 e-shops that describe 2,162 products, and its evaluation found that all tested matching systems struggled with products they had not seen before, to varying degrees. New products appear constantly in e-commerce, so ask for results on SKUs added after setup.

What a good matching process includes

Strong matching combines several layers instead of relying on one signal: identifier matches where both sides have one, attribute matches on brand, model and key specifications, text similarity on normalized titles, image comparison for ambiguous listings, and human review of borderline cases. Each candidate gets a confidence score. High-confidence matches are accepted, and low-confidence ones go to review instead of being guessed.

Approaches that hold up at scale typically add NLP similarity modeling rather than plain fuzzy text matching, token weighting for terms such as size and count, blocking rules by brand or category, and learning from approved and rejected matches. We cover the operational side in our guide to managed scraping vs. in-house.

Software, managed service or in-house: what each is built for

The options below are grouped by what they are built for, not ranked.

Option	Built for	Who does the matching work	Examples and notes
Self-service price monitoring tools	Small to mid-size catalogs with standard e-commerce competitors	You configure, run and review matching	Price2Spy offers automatic, manual and hybrid matching
Enterprise pricing and intelligence platforms	Large catalogs that need exact, similar and private-label matching, governance and contractual SLAs	You operate the platform; some add human-assisted verification	Intelligence Node (99% SLA on exact matches), DataWeave (human-assisted verification)
Fully managed collection and matching	Requirements outside standard platform coverage: specific sources, protected sites, custom fields, difficult matching, delivery to your schema	The provider builds, runs and checks the pipeline and delivers matched data	Ficstar
In-house build	Stable, limited scope with dedicated engineering and QA capacity	Your team	See the managed vs. in-house guide above

For a wider platform comparison, see our guide to the best price intelligence services.

How Ficstar handles product matching

Ficstar matches similar or identical products across multiple competitor sites, even when they are named and described differently. Our product data scraping service pairs your SKUs with competitor equivalents across thousands of items, even when identifiers, naming conventions and descriptions differ. Complex datasets go through 50+ quality checks before delivery, and collection is rerun when a problem is found. Data arrives in the format your systems need.

Example: matching menu items across delivery apps

One of the largest quick-service restaurant franchises in North America needed competitor prices from three delivery platforms across hundreds of locations. In our restaurant chain case study, the same item appeared under different names: "Crispy Chicken" on the brand's own site, "Chicken Sandwich" on DoorDash, sometimes with a size descriptor, sometimes without. We used natural language processing to compare word similarity, order, size descriptors and synonyms, matched high-confidence items automatically, sent edge cases to manual review and kept a reference map of approved matches for later crawls. Location matching worked the same way: addresses were normalized, cross-checked against phone numbers and ZIP codes, and low-confidence matches were reviewed by a person.

You tell us which sources and fields you need, and we build, run and check the pipeline.

When not to choose Ficstar

Ficstar is a managed data collection service, not a self-service tool. Another option fits better in these cases:

• Small catalog, simple sources. If you track a few thousand standard products or fewer against straightforward e-commerce competitors that carry barcodes, a self-service price monitoring tool is likely the better starting point.

• You want to operate the software yourself. If your team wants a dashboard, its own matching rules and day-to-day control of match decisions, a platform gives you that. Managed collection involves more upfront discovery and configuration than opening a self-service account.

• You want repricing in the same product. Ficstar delivers matched competitor data and is not a repricing platform. If you want monitoring and automated repricing in one tool, look at platforms that include it.
  

FAQ

What is product matching in price monitoring?

It is the process of linking your SKUs to the same or equivalent products on competitor sites, so that price comparisons are like for like.

What is the difference between exact and similar product matching?

An exact match is the same product in the same variant and pack size, while a similar match is a comparable substitute such as another brand with the same key attributes. Use exact matches for price comparison and similar matches for assortment and positioning.

What does "99% matching accuracy" mean?

It depends on the vendor's definition, so ask what it measures. Intelligence Node's published SLA, for example, allows at most one false positive per 100 matched exact SKUs, and figures from different vendors are comparable only when precision and recall are defined the same way.

Can product matching work without barcodes or GTINs?

Yes, by using attributes, text similarity, images and human review, but it is harder to get right, so test it on a sample of non-barcoded products.

Should I prioritize precision or recall?

Favor precision for pricing and MAP enforcement, and recall for market research.

Test matching on your own SKUs

Ficstar's free trial runs on your real SKUs and competitors, so you can score the matched output against your own ground truth. Start your free trial.