Churn Forecasting HLS
HLS is a major independent distributor of beverages and catering products in Belgium. It supplies around 1,800 products (such as beer, wine, spirits, soft drinks, water, coffee and dry products) to over 3,000 customers in the catering industry all over Belgium.
The sales team could not understand why customers would stop buying white beer and churn after six months when they had purchased white beer several times per month in the past. Calculating the churn rate involves formulas and interpretation and HLS struggled to do this using traditional spreadsheets and business intelligence methods.
HLS contacted ML2Grow to create a custom analysis, development and visualisation platform for customer churn which could become an integral part of their sales team.
We used Machine Learning to recognise patterns in customers’ purchasing behaviour. We analysed huge volumes of data and identified the characteristics shared by different types of customer behaviour. This enabled us to identify the kinds of customer actions that led to churn. Sometimes these actions were very subtle, or difficult to measure, such as when customers purchased smaller volumes of certain products or left longer intervals between purchases.
This approach is beneficial for companies as it enables them to take action with that information. For instance, a company could choose to actively reach out to a customer and find out why that customer is about to change his or her purchasing behaviour (before they actually do this).
ML2Grow solved this problem by designing an algorithm that detects changes in the buying behaviour of every single customer by looking at their past purchases. The system produces a brief and readable Excel report for each sales manager so that sales teams can take the initiative to contact their customers and understand what is happening.
There are advantages to being able to predict customer churn. It helps you stay focused and avoid losing customer acquisition outlays. ML2Grow has extensive experience in the retail and wholesale sectors. In particular, it creates recommender systems (which recommend specific products to customers or recommend customers who are open to a particular product) and churn prediction systems (which identify customers with the highest risk of churn).
Although we cannot reveal too much about the secret sauce in our state-of-the-art solution, our approach uses churn cutoff calculations together with a classification algorithm. We combine several algorithms which create more accurate results than current models available.
How we added value
HLS significantly reduced its annual customer churn
Insights gained from customer churn were seamlessly integrated into existing sales follow-up workflows
Weekly reports were produced showing customers with a risk of churn so that the sales team could take action