Product recommendation, an integral part of the strategy of personalizing the user experience, enables e-tailers to demonstrate their in-depth knowledge of consumer preferences and needs, as well as their ability to provide them with the products that are just right for them.
This ability to understand and anticipate customer expectations relies on sophisticated product recommendation systems. Before exploring how these systems manage to propose relevant products to each user, it is essential to define what they are.
The concept of recommender system is deployed under various expressions, but the one that particularly catches our attention is “recommender system”. This is a holistic IT system, designed to integrate and deploy various recommender modes, algorithms and use cases. This complex system combines advanced techniques to offer personalized suggestions, responding precisely to each user's expectations. These techniques include :
Content-based filtering
Content-based filtering is characterised by its focus on the intrinsic characteristics of products. This type of filtering is widely used in product recommendation by suggesting items that are similar in attributes to the products the user is exploring - such as their colour, material or other specific features.
Sensefuel has been using content-based filtering for over 5 years through a vector stack, a key element of its solution. This technology assigns semantic vectors to each product, facilitating the calculation of similarities between products. Thanks to these vectors, we can identify which products are close to each other based on their attributes.
This approach is used in various aspects of the Sensefuel platform, including the search engine, result ranking, product selection and many other applications. The same vectors are also used to generate personalised recommendations.
Collaborative-based filtering
Unlike content-based filtering, which focuses on product attributes, collaborative-based filtering extracts meaning from user behaviour. By following each user's path, it is possible to infer preferences and potentially significant correlations between different products.
So if 'User A' visits a series of product pages and 'User B' follows the same path, these products are considered related, regardless of their nature. Whether it's books or shoes, collaborative-based filtering uses the statistical average of a community of users' behaviour to infer correlations and recommend a product when one user consults another.
The data is collected in two ways:
Explicit: the user provides information directly by rating a product or adding it to a favorites list.
Implicit: each visit to the product page, each addition to the shopping cart, and the final purchase are processed to identify consumer trends and preferences.
It is clear that the e-merchandiser has a wealth of information at his or her disposal and that algorithms play an indispensable role in recommending products. However, their effectiveness depends largely on the use cases to which they are applied.
These recommendations can be integrated into various parts of an online store (home page, product listings, search results, etc.) Let's look at some concrete examples.
- Recommendation from a product sheet
There are several possible usage scenarios for a product sheet.
The first approach is to suggest products with similar attributes, such as colour and texture. For example, if the user is on a page about jeans, the e-commerce site will suggest other jeans with the same colour, texture and size, but in a higher price range, with the aim of upselling.
Another strategy is based on buying behaviour: suggesting items that other consumers have bought frequently, in addition to the item currently being viewed. This makes it possible to recommend products that have been validated by others at the time of purchase. In the DIY sector, for example, this analysis can be used to recommend kitchen furniture from the same range or compatible accessories, without having to qualify the information beforehand.
It is also possible to use the purchase path to identify the other products that users have viewed during the same session. Users who have seen this product have also seen these other products. Although this type of product recommendation is noisier, it encourages the user to discover other products.
What's more, it's possible to combine these different approaches: analysing the products viewed prior to the purchase of a particular item to suggest potential alternatives, even if those products were not purchased. For example, if a user is considering buying a velvet jacket, it is relevant to know that buyers of this jacket have also explored other velvet jackets. Presenting these product vignettes encourages diversification of choice and opens up new avenues of exploration.
- Recommendations from the home page, based on the contents of the user's shopping basket
The e-merchandiser could consider recommending additional items based on the user's current cart contents and recent purchases. For example, if the user has already added five or six items, what might be the seventh item to recommend, in line with behaviors observed in the shopper community? Based on the analysis of the different items in the user's cart, this method goes beyond a simple recommendation based on a specific product sheet via an algorithm.
When exploring recommendation strategies on e-commerce platforms, it is important to note that Sensefuel does not limit itself to traditional methods such as collaborative-based filtering or content-based filtering. Sensefuel integrates these approaches with its core AI and advanced personalization system to increase the relevance of its recommendations.
Moreover, its approach to product recommendation does not rely solely on AI algorithms. Sensefuel also provides recommendations based on product selection. E-merchant customers can create specific assortments to promote flagship products, support promotions, and more. The recommendation algorithm not only selects the first items in the selection, but also carefully analyzes users' individual preferences to select from product assortments that best match their specific tastes. This approach takes advantage of the personalization system built into the entire Sensefuel solution, which offers the e-merchandiser total freedom of configuration thanks to a turnkey solution and widgets that can be integrated directly in the desired location on a page.
Sensefuel offers both Plug & Play and API deployment modes, designed to give the e-merchandiser easy control over what is displayed where on the site, without necessarily relying on a technical team.
Conclusion
Product recommendation is much more than a technical feature of e-commerce platforms. It is a strategic tool that allows retailers to demonstrate their deep understanding of individual consumer needs.
By integrating different approaches such as content-based filtering and collaborative-based filtering, combined with personalized selection methods, Sensefuel provides a flexible and powerful solution for optimizing the online shopping experience. This approach not only increases sales by offering relevant products, but also strengthens customer loyalty by ensuring a personalized and engaging shopping experience.