You are looking for a pair of blue pants online. Unfortunately, they do not have your size in stock. What can you do? Recommendations based on image similarities help you find appropriate alternatives. That translates into exciting applications for operators of online shops.
When you are looking for a pair of blue pants in a store and your size is not available, staff at the store offer you similar pants, preferably blue, but perhaps in other muted colors as well. Increasingly, online shops are also using recommendations based on image similarities such as “similar products” in order to show customers alternatives when products are out of stock. This way retailers can improve the shopping experience for customers while reducing purchase cancellations. They also increase the conversion rate and sales from recommendations at the same time.
Images in particular are always a huge challenge for recommendation engines because “machines” have no access to the illustrated “data” such as color or shape. Current technologies in the field of artificial intelligence now make it possible for personalization solutions to access this data.
Image similarities – how do they work?
An AI service regularly analyzes all existing product images of a retailer for similarities. A multi-layer neural network trained in image analytics also analyzes the images and detects similarities. The detected similarity is then mapped based on the strength of the relationship and sent to the personalization solution.
A recommendation type for image similarities can also take into account non-visual settings to generate relevant recommendations. Here it makes sense to have, for example, only products that are in stock or that, depending on the intended purpose, are from the same category or only from other categories and that are similar in price or of higher quality.
Advantages of image to image recommendations
Out of stock situations cannot always be avoided. In the absence of relevant recommendations, customers will often leave the shop. If, however, they are actually presented with an alternative similar to the desired product, they do not have to continue the laborious task of searching and clicking through lists of products. Are there other pants in my color and size? Shop owners already know what the customer is looking for. If customers are genuinely interested in buying, the probability that they will look at these recommendations is very high.
Here’s how to use recommendations based on image similarities
In addition to the relevant alternatives to sold out products just described, there are two other exciting applications.
Customers can receive recommendations for complementary products with the same look, e.g. “Shop the look”. This is especially exciting in the fashion and accessories, jewelry, home decor and furniture sectors. A sought-after item is complemented by an item with a similar look: a belt to match the shoes, a scarf to match the blouse or the earrings to match the bracelet. Product images contain important information that the shop operator would otherwise have to tag using language, requiring a lot of effort.
The next step for this technology is the visual search in the shopping app or using a chat bot. You see a cool watch in a magazine, take a picture of it, send it to your favorite retailer via messenger or using a shopping app and you get real-time recommendations of visually similar items in the shop. This makes shopping even easier for us as customers, guiding us to products that we like more quickly.