Shopping recommender System
Recommender System for e-commerce: how to help your customer find what he really needs?
What was before?
Online shopping has become an integral part of our life and with the pandemic the share of online shoppers has dramatically increased and will continue to grow. Therefore, it is extremely important to ensure a seamless shopping experience and keep sales rates high. In the extremely competitive environment, the e-commerce platforms need to find a way not only to increase the purchase conversion and the average ticket, but also to make the customers happy and thus retain them longer. Brands delivering rapid, on-point offers capture the most benefits.
What was the issue?
Very few people come to an online shop with a clear goal to buy a concrete product. Others would wander around not being able to find what suits them. If they do find it, they will most likely just leave after purchasing a product, without buying something else to go with it.

The good news is that the online shops have tons of data about their customers that they can use to boost sales and make the shopping experience more comfortable. Based on this data the shops can provide every customer with the personalized recommendations on products that he or she will most likely buy. This feature can be implemented based on classic algorithms, but as the number of products and metrics grows, the system becomes too complex and unmanageable and only Artificial Intelligence can handle it.Due to fast growth of IT industry market dictates large and small companies must compete for talents and constantly sharpen their hiring process. Besides competition, hiring in IT has higher requirements to recruiters who also must demonstrate enough of IT knowledge to be able to evaluate applicant's CV and support screening interview to detect and disqualify unskilled applicants.

In most cases proper candidate evaluation requires professional analysis of sample computer code provided by candidate. This step requires expensive time of senior developer and there may be hundreds of code samples. It means that resolving shortage of software developers requires more hours of software developers for candidate evaluation. It's a vicious circle.
What did we do?
VisionSystems have designed and developed a complex recommender system for the online North American shopping platforms. It captures data from the customer's personal data, preferences, past and current shopping sessions, purchases of the other customers similar to him and transform it into dynamic offers. It then scans through the entire product catalogue and lines up the best products for the individual customer.
What was the result?
Recommender system has allowed the online platforms to finally take the most advantage of the data that they collect and recommend the customers the goods that they are most likely to buy. 20 % of recommendations that our system has generated have proven to be correct. One of the major concrete benefits that the solution brings is the growth of the average ticket. One more benefit that one cannot underestimate is the increase the customer satisfaction and consequently retention, since no customer will leave the online shop that understands his needs and forecasts them accurately. Additionally, increasing the upsale has been a great way to save on the shipping costs, since it is cheaper for the online-shop to ship when a customer purchases several items instead of just one.
How it works?
The core of AI the system is a software that uses machine-learning algorithms. The Machine Learning model analyzes the historical data on sales and the personal data of the users and generates the shopping recommendations to every user visiting the online shop. There may be several models like that running simultaneously and later they unite in an ensemble to provide better quality recommendations.

Within the user-based approach, the model analyzes the profile of the customer and his preferences. Once it has defined the profile, it finds the profiles of the other customers that he is similar to and recommends to him the products that they bought.

Within the item-based approach, the model analyzes the customer's shopping session and recommends the products equivalent to the products that he's already viewed and the products complimentary that he has viewed, put in the basket or bought earlier. E.g. it would recommend the coffee capsules to the customer who is viewing the coffee machines or the cleaning products for the carpet he has bought a few months ago.