E-commerce search and recommendations is the software behind an online store's search box and its "you may also like" rows. It understands what a shopper types, ranks the store's products for that search, and suggests other products. It is also called site search, product search or product discovery.
Shoppers rarely type what the catalogue says. Consider a kitchenware store, and a shopper in Kolkata who types "non stik tawa 28". The catalogue calls the product "Non-Stick Flat Dosa Pan, 28 cm". A simple search that matches only exact words finds nothing, and the shopper leaves. Search software corrects the spelling, knows that "tawa" and "dosa pan" mean the same thing here, and reads "28" as a size in centimetres. It shows the pan as the first result.
To match meanings and not just words, the software turns each product and each search into vector embeddings. An embedding is a list of numbers that an AI model makes from a piece of text or an image. Texts with similar meanings get similar lists of numbers. So the search "pan for dosa" gets a list of numbers close to the dosa pan's list, even though the two share no word. This also helps with searches typed in Hindi words but English letters, such as "steel ka dabba".
Finding the matches is the first step. Ranking them is the second. The software puts the matching products in order by how closely they match, how often shoppers bought them after the same search, and whether they are in stock. It also follows rules that the store sets, such as showing the brand's new range first. Products that are out of stock are hidden or placed last.
Recommendations work from what shoppers do. "Frequently bought together" comes from past orders: shoppers who bought the dosa pan often bought a wooden spatula too. "Similar products" comes from the embeddings. Both use the store's own records of what shoppers searched, viewed and bought, which are called first-party data. A customer data platform (CDP) can add more about each shopper. When the results change for each shopper, the software is doing e-commerce personalisation.
Search is only as good as the product data behind it. If the catalogue does not record that a pan is 28 cm, no search can find it by size. So many stores use AI product descriptions to fill in such details. Newer stores add an AI shopping assistant, which answers a shopper's question in a conversation and uses the search software to find the products. Shoppers can also search with a photo instead of words, through visual search.
A marketplace with millions of products builds its own search. A D2C brand with 400 products usually buys search as an app for its store.