Act before it’s too late!

Subscribe to emnos news!

Florian Baur

VP Client Engagement Northern Europe, USA

writing about trends, developments & innovation.

, created by Florian Baur

Act before it’s too late!

Customer retention costs a fraction of acquisition - new emnos statistical model predicts potential churn

With consumer choice and expectation at an all-time high, the retail sector needs to adopt a truly customer-centric approach to nurture and retain customers. emnos has developed a Customer Retention Model which uses transaction data from loyalty cards to identify early signs of customers who are shopping less and predict those in danger of moving on. The same data establishes what those shoppers really want and provides the foundation for an engaging and personalised experience in time to strengthen individual loyalty before it’s too late.

Traditionally, the retail sector has tended to put more weight behind the search for new customers rather than focusing on those it already has – a surprising trend considering acquisition costs around five times more than retention. Also, according to Gartner Group, 80% of future revenue will come from just 20% of existing customers.

 

Test and Learn

The emnos Customer Retention Model has three key phases.

  1. Definition of potential churners based on purchasing patterns and spend. For example, very low frequency shoppers are excluded from the analysis so the Model can focus on the most regular and valuable customers.  This definition provides the basis for the Model and will vary from company to company.

  2. The most relevant variables are isolated from a comprehensive list of factors. So key points may include frequency, frequency evolution, mission of last basket, sales volume, basket and number of branches visited.  This data helps discriminate between churners and non-churners.

  3. Finally, the Model is tested using a range of statistical indicators (such as ROC curve or confusion matrix) to assess its validity given the established churner definition. This definition can, if necessary, be modified and/or further relevant variables added. Having passed all the quality checks, it can now be used to predict customer behaviour and rank it based on ‘likelihood to leave’.


Early success

Designed specifically to address the issue of retaining customers, the new Model has already proven its worth for a French chain of over 900 convenience stores which was starting to lose its most frequent shoppers.   Frequency was, therefore, the principal factor followed by potential based on total spend over a six month period. Having identified the customers most at risk of leaving the chain, 49,000 were targeted with marketing campaigns designed to encourage loyalty and retention. 

Mailshots, emails or coupons were sent to each customer depending on their individual retention scores and average baskets over the previous six months.  Using an approach proven to have stronger appeal to potential churners, the offers did not focus on products but gave a discount based the amount spent – for example, spend 15€ and get 3€ back. The value of the offers varied depending on the customers’ average spend and ‘likelihood to leave’.  Over 300,000€ in additional revenue was generated and there are plans to run similar campaigns on a monthly basis throughout 2016.

Loyalty card data provides a wealth of information on customer behaviour but the challenge lies in providing a meaningful interpretation. When analysed correctly, it can deliver the depth of understanding needed to accurately identify customers showing signs of slipping away.

Back