AI Product Matching › Accuracy
How do you measure the accuracy of a product matching tool?
Take a few hundred of your own products, find their correct matches by hand, and run the tool on the same set. Measure precision, the share of the tool's matches that are correct, and recall, the share of correct matches it found, at the threshold you plan to use. For price comparison, precision matters most.
Updated 10 November 2026 by the ScrapeWise team.
Quick picks
For pricing teams
Optimise precision
Wrong matches create fake price gaps.
For assortment analysis
Balance with recall
Missing products hide gaps.
To test on your own catalogue
ScrapeWise (our product)
Run a sample; matching is currently free.
Compared
| Metric | What it measures | Formula | When it matters |
|---|---|---|---|
| Precision | Share of accepted matches that are correct | correct matches ÷ accepted matches | Pricing decisions: wrong matches cost money |
| Recall | Share of real matches found | correct matches ÷ all real matches | Market coverage |
| F1 score | Balance of precision and recall | 2PR ÷ (P + R) | Comparing tools with one number |
| Review rate | Share of pairs sent to a person | review pairs ÷ all pairs | Your team's workload |
| Coverage | Share of your products with any match | matched products ÷ all products | Whether the data is usable |
A fair test, step by step
- 1. Sample
- Pick 200–500 products across your main categories, including hard ones.
- 2. Label
- Find each product's true match in each competitor by hand, or mark “not sold”.
- 3. Run
- Give every tool the same products and competitors.
- 4. Measure
- Count correct, wrong and missed matches at the same threshold.
- 5. Inspect errors
- Look at the wrong matches; they show whether the tool fails on sizes, variants or bundles.
Questions to ask vendors
- Measured on what?
- Accuracy figures from other catalogues don't transfer to yours.
- At what threshold?
- Precision and recall change with the threshold.
- Who reviews?
- Some “accuracy” includes human review you pay for.
Frequently asked questions
How do you evaluate the accuracy of a product matching tool?
Label a sample of your own products by hand, run the tool on the same set, and measure precision and recall at the threshold you'll use.
What is precision in product matching?
The share of matches the tool accepted that are actually correct.
What is recall in product matching?
The share of all real matches that the tool found.
What is a good product matching accuracy?
It depends on use. For pricing, aim for very high precision and review uncertain pairs; for assortment analysis, recall matters more.
How many products do I need to test?
A few hundred products across your main categories usually show the differences between tools.