Bridging the gap with customers using machine learning on the NPS dataset

Abstract

One of the key tools companies worldwide use to measure customer satisfaction is via the Net Promoter Score (NPS). This thesis tackles the problems in analyzing NPS data due to open text feedback, low response rates, and class imbalance in customer churn prediction. This research applies machine learning techniques such as natural language processing (NLP) and regression models to classify open-text feedback, fill in missing NPS scores and predict customer churn. To that end, SpaCy is used specifically to identify key drivers from open text feedback responses and categorize them to help draw meaningful insights. Customer satisfaction and loyalty were explored using sentiment analysis and predictive analysis models. The study also handles class imbalance indirectly through appropriate resampling techniques used as a condition to guarantee balanced and effective identification of churn rates. These results show that classifying feedback and predicting missing scores enhance the accuracy of NPS analysis and customer churn prediction yields actionable insights to inform retention strategies. In this work, we demonstrate how NLP and machine learning can solve some of the common challenges of customer satisfaction metrics and bolster machine learning based NPS analysis.