How an e-retailer employs machine-learning to reduce customer churn


How an ecommerce company used predictive analytics to improve customer retention.

Name of Organization: Showroomprivé

Industry: Ecommerce

Location: La Plaine Saint-Denis, France

Business Opportunity or Challenge Encountered:

Acquiring new customers is much more expensive than retaining current ones. Keeping customers from unsubscribing from a company’s services or from choosing another company’s solution is therefore a challenge that should be at the top of every corporate agenda. A way to address this challenge is through predictive customer churn prevention, in which data is used to find out which customers are likely to churn in order to win them back — before they are gone.

Showroomprivé.com, an ecommerce company founded in 2006, sought ways to employ machine learning approaches to retain more customers. The challenge is huge: The e-tailer is one of the leading ecommerce players in Europe with more than 24 million members, has about 15 flash sales and over two million visitors per day. Showroomprivé registered gross turnover of over €600 million in 2015, meaning €443 million in net sales, up 27 percent versus the previous year. The company employs more than 800 people. specializes in fashion, offering a daily selection of 1,500 brand partners on its mobile app or online.

How This Business Opportunity or Challenge Was Met:

In order to counter churn, Showroomprivé had been using static rules to trigger marketing actions. These rules were common to all customers and no prior qualifications were made to determine the value of each individual client.

To anticipate, prevent, and reduce customer churn rates and improve customer loyalty, Showroomprivé first sought to detect clients with a high potential of no longer buying from the website based on individual purchase rates. Then, they intended to refine targeting of marketing campaigns for each potential churner to win them back.

To accomplish this, the e-tailer employed Dataiku Data Science Studio (DSS) to develop a solution that predicts whether or not a buyer will return to the website to make a purchase.

Showroomprivé also leveraged technologies including HP Vertica, Microsoft SQL Server, Python, and used models including Gradient Boosting Tree, Logistic Regression, Stochastic Gradient Descent, Random Forest and Decision Tree. It took one month and a half, one data scientist, and a CTO to deploy the first version of the application.

The integration and enrichment of a variety of data sources — customer data, catalogue data, order and delivery data, and web logs — was completely automated. Showroomprivé’s team created more than 690 features derived from this data, based on variables such as clicks on sales, orders, litigation, customers. The new system also enabled the team to test multiple machine learning algorithms to achieve the best predictive model.

Measurable/Quantifiable and “Soft” Benefits from This Initiative

A previous analytics solution relied on descriptive analytics on past data, says Damien Garzilli, strategy and business intelligence manager for Showroomprivé. The current in-house prediction system, which employs real-time capabilities, now automatically detects future churners with about 77 percent accuracy. The team can now detect potential churners with an AUC of 0.819 (Area Under Curve, a metric for binary classification in predictive algorithms) and act accordingly to win them back.

(Source: Dataiku)


Case studies on customer experience management

Industry insights: retail analytics

Leave a Reply

Your email address will not be published. Required fields are marked *