Problem: While filing a complaint, customers are asked to choose the complaint category or theme. However, customers are unaware of business terms so they often choose the wrong category. This was a major issue for us with more than 20% of complaints being misclassified and routed incorrectly. This led to long wait times and incomplete resolution of customer complaints.
Solution: We leveraged topic algorithms such as TF-IDF and LDA to reclassify customer complaints based on the exact language used in the complaint. As seen in the chart below, there are several differences between original (customer-led) and new (NLP-recommended) themes. Especially, NLP-themes identified ‘Late Fees’ as the main reason (17%) for customer complaints, which was considerably higher than the previous estimate.
