Recommender system as the central component of a successful web store
The number of web stores and products offered by them is growing rapidly, but people’s attentiveness hasn’t improved. Therefore, the critical challenge for every web store is to guide users to products that they are interested in and to do so in a way that would help to achieve business objectives. Today, all of the 10 largest e-commerce companies in the world have developed their own artificially intelligent recommender systems, and the pioneer in the field, Amazon carries out 30% of the sales with the help of their recommender system. Recommender systems have clearly become a central component of successful web stores. Most online merchants, though, can’t afford their own machine learning team who would develop a recommender system for the web store. Therefore, the question arises: how to implement a recommender system? For the most popular e-commerce platforms (Magento, WooCommerce, etc.), the so-called OOTB (out-of-the-box) solutions are available as modules, which enable displaying recommendations to users when installed. Unfortunately, it’s not possible to create a well-functioning recommender system like that. Anyone who has experience with OOTB modules is well aware of the following three issues:
- OOTB modules don’t enable setting a business objective for the recommender system;
- Business rules can’t be applied to OOTB modules, and therefore, “silly” recommendations are given;
- The performance of the recommender system offered by the OOTB modules can’t be monitored and these systems don’t improve in time.
STACC recommender systemSince displaying “bad” recommendations to users brings more damage than benefits to the company, we took up the challenge of developing a recommender system that would be easily deployed in a web environment and would use machine learning for making recommendations. Today, we have developed a product that enables the following: 1. Recommender model is guided by business objective, data and business sector During the years of development, we have implemented all of the most successful recommender models into our system. Different models are being tested in live environment and system automatically selects a model that maximizes the business objective that is set for the recommender engine (e.g. increasing sales revenue). 2. Different filters are applied to the recommendations To avoid “silly” mistakes, our system has the total flexibility to include business-specific rules and filters. The following examples help to clarify what we mean by rules:
- Can the item X be recommended when the user has already bought it before?
- Example 1: If you are selling bicycles, there is no point to recommend the same bicycle to the user again.
- Example 2: If you are selling milk, you can recommend that to the user every day.
- What are the restrictions for certain items?
- Example: Intimate care products should never be recommended.
- Are there any special rules to be applied for certain user groups?
- Example 1: Never recommend field player’s gear to a goalkeeper.
- Example 2: Never recommend meat products to vegetarians.
- Additional rules:
- Example 1: Recommend only products with higher price from the same product category.
- Example 2: Don’t recommend more than two products from the same product category.
If you have a web store and you are interested in implementing an artificially intelligent recommender system, order a free recommender system demo by filling out the form below!
8.12.2020 | NEWS
STACC Recommender System through the lens of architectureActivities 2016-2017 STACC built the first recommender system back in 2016. It was an e-mail recommender that was built to…
31.08.2020 | NEWS