USING A PREDICTIVE CHURN MODEL TO ENSURE CUSTOMER RETENTION
CHURN PREDICTION FOR A GLOBAL SUPPLIER OF SUSTAINABLE RAW MATERIALS
Businesses often suffer investment losses when customers begin to churn, such as when they unsubscribe, buy from a competitor or cancel a contract. It is therefore crucial that companies figure this out early on, as acquiring new customers often costs more than retaining existing ones.
With our expertise in machine learning projects, our experts can develop or optimize an existing predictive churn model. For example, our Data Scientist recently helped one of our clients by converting their existing model into a smart and fully automated model running entirely on Amazon Web Services (AWS). With an accurate forecasting tool, our client could take immediate action and prevent churn from happening. For a better and clear overview, all prediction insights are visualized in a Power BI dashboard that shows the key factors that could contribute to churn, what the churn risk is per customer and which one should be prioritized. During this project, we assisted our client with:
- Machine learning
- Power BI
- Cloud deployment
To properly optimize the model, we had to define what constitutes a churn event. During a churn prediction project, typically socio-demographic data, historical transactions, client-company interaction or e-commerce behaviour and so on is used. For our client, we analyzed a huge amount of historical data collected over four years and broke the data in six-month time frames to identify churn events. We then examined the period when order volume declined and churn occurred, along with other internal & external factors that could contribute to churn. For example, complaints about deliveries or current inflation affecting the market. We also had to clean up data and break down the code into many functional units that work together. That way, when a problem occurs, we immediately know where it is and how to fix it, rather than searching for a needle in a haystack.
With the current market turbulence, our client wanted to have a clear picture of the potential business risks. The most ideal scenario would be, of course, to prevent churn from occuring by identifying the key factors that can help stakeholders take immediate action. This means they need to gain insights into the key factors that can contribute to churn, what the likelihood of customer churn is and which key factor should be prioritized. Furthermore, to ensure our client gets a better and clear overview of the predictions, all insights had to be visualized in a user-friendly Power BI dashboard.
MEDIAAN IN ACTION
During this project, our Data Scientist worked closely with our client, specifically with their Data Engineer and a stakeholder. They used the following methods and technologies:
AWS Sagemaker -> Jupyter Notebook
AWS cloud formation and Step functions
AWS S3 buckets and Athena
AWS Glue Jobs
Not only have we supported our client in fully automating and rebuilding the existing predictive churn model, but we also helped deploy it in the AWS cloud environment. In addition, stakeholders can now gain insight into the key factors contributing to churn, when churn will occur and the churn rate of customers, visualized in a Power BI dashboard. The next step is to further optimize the model and make it more robust so that it can provide even more accurate predictions.
By having an accurate churn predictive model, companies can analyze client behaviour and measure their probability of churning. Our experts always ensure that the solution is future-proof and scalable.
WANT TO SEE MORE?