bonprix: Artificial intelligence for optimised product ranking

With its product potential forecasting model, international fashion retailer bonprix is using a new forecasting model developed in-house that is based on artificial intelligence (AI) and forms the basis for the daily updated product rankings in all product categories of the bonprix online shop. This enables the company to offer its customers an even better service, as the products that are currently the most popular and at the same time sufficiently available across all sizes are displayed at the top. To achieve this, bonprix innovatively uses a convolutional neural network – a technology that originally comes from machine image recognition.

bonprix has been investing in the in-house development of AI models for years or uses systems from service providers. The latest application is an optimised product ranking in the online shop www.bonprix.de. Product ranking here means that all products in the various collections or categories are re-sorted and displayed hierarchically on a daily basis. The product potential forecast model is designed to display the products that are particularly popular with customers and in inventory / stock at the top of the list. The model was developed internally and, after extensive A/B testing, has successfully replaced the previous application in Germany. In addition, the product potential forecast has already been implemented in Austria, Switzerland and ten other countries.

Sascha Netuschil, Head of Data Science at bonprix, is responsible for the development: "At bonprix, we are always looking to further improve the customer journey on our sales channels and want to offer our customers exactly what they are looking for – virtually at the click of a button. The use of AI to process the useful, complex interaction data of our customers is a significant step forward in this regard. With the product potential forecast, we can significantly refine the product ranking in the bonprix online shop for our customers and, above all, ensure that the top-ranked products are available in sufficient quantities. This also has a positive effect on conversion rates," he summarises the advantages and goals of the new application.

Innovative combination of multiple time series and convolutional neural network

In the bonprix online shop, product rankings in various categories, which are updated daily, help customers quickly find the latest trends without having to click through countless pages. They are based on previous click and order data for individual products and their associated items, and display those for which a high order volume can be predicted on a daily basis at the top of the list.

The product potential forecast model works with multiple time series. These are created by daily observation of user interaction parameters for a product over the past seven days: how often this product was clicked on, how often it was added to the wish list, how often it was placed in the shopping basket and how often it was ultimately purchased.

This creates a matrix in which patterns can be identified that are relevant for the order forecast and ranking control of a product. To do this, bonprix has made use of AI technology that has previously been known primarily from image data recognition: a convolutional neural network (CNN). Since the data matrix for a product is structured like an image, the CNN is able to read the data and calculate a sales forecast for the coming day. In addition, inventory levels are also taken into account in the ranking: if the availability of a product falls below a specified value, it is automatically ranked lower. If it becomes available again later, the model recognises it as a former top product based on historical data. This means that the product potential forecast provides far more reliable predictions than the machine learning model previously used at bonprix.

In order to display even more suitable products to its customers in the first place, bonprix refines the product ranking for registered users in a further step by means of clustering based on five different age groups. The products that are particularly popular in the respective age group are then displayed hierarchically.

Artificial intelligence as a driver of innovation

bonprix sees the ever-growing range of AI-based applications as an important driver of innovation for the e-commerce industry. For years, the fashion company has therefore been focusing on the continuous technological evolution of its processes using AI and machine learning – whether in the context of fraud prevention, better size advice for customers or optimised product range design through the so-called Learning Collection. All analyses of customer-related data are always carried out in strict compliance with the General Data Protection Regulation (GDPR).

‘We want to show our customers the relevant part of our product range in an increasingly targeted manner. Product potential forecasting is a good example of an application that makes this possible. In the future, we plan to further personalise product ranking by utilising behaviour-based data,’ explains Markus Fuchshofen, Managing Director of bonprix and responsible for e-commerce management, domestic sales and branding. He also points to the great future potential that bonprix sees in the ongoing implementation of AI: ‘We have been using artificial intelligence across the company for years. The migration of all processes to the Google Cloud, which was completed this year, provides us with the ideal basis for continuing to expand the range of AI applications at bonprix in the future.’