API-only product matching is a fully automated product matching process, meant for service providers (price / assortment monitoring companies) who do have capabilities to perform Automatch of their own, and need a service to perform match-scoring, but need no absolute guarantee on returned results.
How does API-only product matching service work?
The process is as following
- You find the matching candidates (in some cases this is search for given product, on targeted Website, while in others it can be a site crawl)). In any case, it’s important that you prepare a list of matching candidates.
- You perform Product Matching AI API request. Your request should consist of
- Product that needs to be matched, with following product fields
- Product name
- Product price (including any discounts)
- Product URL (the product page)
- Product image (as binary object)
- Product description
- All of product matching candidates
- Each matching candidate needs to come with same set of fields (5 fields listed above) as product to be matched
- Product that needs to be matched, with following product fields
- Product Matching AI will respond, for each matching candidate, with matching score, which can also be interpreted as matching probability. It will be up to your software to decide whether a particular matching candidate is good
How about the matching accuracy?
Product Matching AI’s model has been trained on nearly 1 000 000 of product matching examples, from various industries. Our tests have shown 97% accuracy. We are working on further improving it.
Real-life example
Once again, we will illustrate the above process on product matching for Dove Daily Care Shampoo 400ml
If you perform Automatch on targeted website, looking for Dove Daily Care Shampoo 400ml, your Automatch would result in 3 matching candidates.
Again, you submit 1 API request to our service. This request will have
- Original product, with it’s 5 fields (name, URL, price, image, description)
- 3 matching candidates, all of them with their 5 product fields (same fields as listed above)
- Product Matching AI will apply it’s machine learning model on each of the above candidates, and return it’s matching score
In the above example
- Matching Candidate 1 will return matching score of 0.534
- Matching Candidate 2 will return matching score of 0.957
- Matching Candidate 3 will return matching score of 0.650
As you can see, matching scores indicate that Candidate 2 is the right one (while other two are not).
Who is this service for?
This service is intended for companies who can come up with matching candidates of their own (some of them do it by Automatch process, and some of them by crawling targeted websites) and need a service which will assign product matching probability for each of these candidates.
In most cases, this service is used by price monitoring companies, who need to provide product matching as essential part of their service.
API-based product matching: a continuous or one-off process?
Since this service is API-based, it’s meant to be used continuously – and that’s what we would strongly recommend to all our clients.
Our recommendation would be an automated data flow which
- Takes new products from your client’s accounts
- Performs Automatch (finds matching candidates)
- Interfaces Product Matching AI API – which will get these matching candidates scored
- Implement a logic within your own system which will proclaim the matching candidate with highest score (if above certain threshold) as a good match – as opposed to other candidates