Data-Driven Artificial Intelligence (AI) for Churn Reduction
ConferencesResearchAPAC 2020churn modelLogistic Regressionposted by ODSC Community November 23, 2020 ODSC Community
A telecommunications company was losing customers (churn rate was 49.9%) and wanted to identify why customers were leaving them. Using data-driven Artificial Intelligence (AI), the key reasons for customers leaving the business (churn) was identified and a proactive retention campaign was developed to prevent customers from leaving the business.
Example: Churn Classification
The Logistic Regression Model was used to build a data-driven churn model. The dataset had 2404 customers and twenty-three predictor variables were identified as relevant customer data to investigate to determine the drivers of churn and which customers were likely to churn.
1. First, we import the python libraries for training our churn model.
2. Next, we read in the relevant Churn Data using a training dataset and testing dataset
3. Next, we explore the data by looking at the summary statistics
4. There were only three continuous input variables, the other twenty input variables were all categorical variables. We next checked to see whether there was high multi-collinearity between ’Total Charges’, ‘Monthly Charges’ and ‘Tenure’. As the correlation between ‘Total Charges’ and ‘Tenure’ was high (0.858664), ‘Total Charges’ was not included as an explanatory variable for the Churn Model.
5. Next, we fit the Logistic Regression Model, to determine the significant factors that drive churn
6. The summary of the Logistic Regression Model Coefficients helps us to determine the key drivers of churn. The higher the variable coefficient, the more influence that variable has on churn.
As we can see, FiberOptic, DSL and MonthtoMonthContract have the highest coefficients and therefore are the top 3 influential variables driving churn.
Results: The accuracy and performance evaluation of the Churn Model (using the training data) was determined by running the Churn Model on unseen test data.
7. We next fit a Confusion Matrix, to evaluate the performance of our training model on the test dataset.
The overall accuracy of the Churn Model was 0.7415, this means that 74% of customers were accurately identified as churners or no-churners.
8. The ‘Sensitivity’ of the Churn Model was 0.7504, this means that for every 100 customers that were churners, the model identified 75 churners. The ‘Specificity’ was 0.7316, this means that for every 100 non-churners, the model correctly identified 73 non-churners. The ‘Area Under the Curve (AUC)’ was 0.7416. The higher the AUC, the better the Churn Model is at distinguishing between customers who will churn and customers who will not churn. An AUC of 0.74 is close to 1, this means that we may accept the Churn Model as a good Artificial Intelligence Tool for identifying churners.
Conclusion: The telecommunication was able to reduce customers from leaving them by investigating their ‘FiberOptic’ and ‘DSL’ services and through the discovery that the ‘FiberOptic’ and ‘DSL’ speed was extremely slow and often caused interruptions to TV shows and Moving Viewing. The telecommunication company quickly fixed the FibreOptic and DSL services problem and offered their customers who were highly likely to leave them retention packages giving free ‘FiberOptic’ and ‘DSL’ services for the first six months of a 2-year contract. All in all, with data-driven AI, the telecommunication company was able to reduce their churn from 49.9% to 9.99% and increase their profitability.
Carol is a speaker for ODSC APAC 2020. Check out her talk, “Empower your Organization with Data-Driven AI,” there to learn more about AI and churn reduction, among other relevant topics.
Many organizations have lots of data, but how do we make sense of the massive data. There is a clear need to innovate and digitize many work processes to make better sense of data. With data science, companies can apply scientific algorithms to systematically find patterns in the data so that business decisions are smarter and faster, providing the company with an advantage over its competition.
In this talk you will learn:
– How different industries are using AI so that business decisions are smarter and faster.
– Learn how to apply the data science framework for solving business problems
– The 8 Principles that are essential for moving towards a data-driven organization
About the author/ODSC Speaker: Carol Hargreaves
A born leader with a passion for solving business problems using big data analytics, machine learning & artificial intelligence to build data-driven solutions that deliver growth & enable informed decision making resulting in revenue growth allowing business processes to become smarter & faster while keeping customers engaged & delighted. Analytics and Business Intelligence Professional with over 30 years of experience, with leading roles in the Pharmaceutical, Healthcare & Fast Moving Consumer Goods industry and in the Education Industry. An excellent Analytics Instructor for Solving Hands-On Real-World Business Problems.