Customer Segmentation

The project aimed to introduce the new segmentation approach where the customer behavior is analyzed and are segmented based on how they use the institution's products, how they engage with the institution and ultimately, taking into account of the customers' financial needs.

Current customer tagging is not a realistic representation of the number of business customers, with 99% of customers tagged as persona. This current tagging has become the road block in a lot of projects as this is the very first level of customer segmentation.

For the analysis, machine learning model is used, asking algorithms to collectively look at the banking behavior of the customer and predict which behavior pattern is a closer match for which group, business or personal. The datasets covered the demographic data, the savings behavior, and transaction behavior and time with the institution.

The predicted composition of business-personal customers ratio from the algorithm reflected very similarly to the results of independent qualitative research carried out in prior, which claimed about half of the customer base are using the accounts with financial institution for business purposes.

K-Nearest Neighbors (KNN) for classification algorithm was used.

Model is trained on 70% of the dataset and accuracy is tested on remaining 30%, unseen from the training process.

Accuracy of the prediction is 97%.

This project had begun with the product owner asking if there's a way to segregate business customers from personal tagging using data for forecasting purpose. After the analysis, the insights generated and algorithm predicted tagging has become the backbone of major business initiatives such as the base dataset for loan funnels for up-selling and cross-selling purposes, for targeted sales, and further customer segmentation work that aims to slice down further on sub-segments on each business and personal customer group.

Disclaimer: All the information described are based on the real-life projects at the financial institution, however, the detailed figures are altered due to the sensitive nature of the information.