3 ways to approach a recommendation system for an online store

assets/files/images/08_12_21/3-ways-to-approach.jpg

This article is brought to you by Retail Technology Review: 3 ways to approach a recommendation system for an online store.

By Richard Grant,freelance writer.

Providing e-commerce customers with recommendations of other products to add to their basket based on their behaviors and buying habits is a great sales tactic.

There are a few options when it comes to implementing a recommendation system on your own shopping site, so you have a certain amount of flexibility available to you.

To help you choose, let’s go through three of the main routes available and explore the pros and cons each offers.

Filtering through collaboration

A simple yet effective option, collaborative filtering for product recommendations simply means that you suggest products to customers based on what other users with similar interests have purchased in the past.

Users can be linked together in this way if both have snapped up a specific item from your store previously, indicating that their personalities and tastes align with one another.

Product ratings, not just purchase histories, can be harnessed almost identically in this context, and so long as you adhere to restful standards, you should be able to extract and analyze customer data to determine this.

Obviously there is the potential for this type of filtering to misfire from time to time, and its accuracy is reliant on having a decent amount of data relating to a customer available to you already.

The longer a person has been using your store, the better you will be able to leverage collaborative filtering to recommend products.

Filtering through product descriptions

Next up we have content-based filtering, which matches customers to products according to the way that the items on your store are described.

Drawing connections between features and functionalities of a product and the preferences of a user can be done in advance.

You can also improve the accuracy of recommendations for new customers by getting them to pick out their preferences when they create an account, or simply wait until they have made a few purchases to build a profile on them instead.

Again, the more data at your disposal, the better content-based filtering for product recommendations will work. However, since you do have the aforementioned ability to encourage users to personalize their own choices up front, this could be valuable.

Filtering through hybridization

As you might expect, product recommendation systems do not need to be adopted with a binary approach to filtering.

Lots of retail sites blend a few different elements, meaning that you can narrow down suggestions through collaborative and content-oriented factors without being beholden to just one metric.

This sounds good in principle, but in practice there are some complications and limitations to take onboard.

For example, a dearth of data is the biggest bugbear for online stores, and if your site is newly launched then you will be in quite the predicament.

Collaborative filtering will take a while to get up to speed, but can deliver precise results further down the line. Content-focused filtering does have a means of starting off stronger, but isn’t always enough on its own to bring home the bacon.

Looking to the long term

As with all aspects of running an e-commerce site, it is not helpful to expect that a recommendation system will work wonders immediately. Instead you have to be patient to see the desired results, and also be willing to track how the system performs and tweak it if necessary.

You could even discover your own unique approach to pitching recommendations to customers that fits in better with your target demographic, so it is worth drilling into sales data and on-site behaviors to see if there are unique solutions to your sales conundrums.

Add a Comment

No messages on this article yet

Editorial: +44 (0)1892 536363
Publisher: +44 (0)208 440 0372
Subscribe FREE to the weekly E-newsletter