Saurabh Batra’s research while affiliated with University of Delhi and other places

Customer journey analytics is the process of monitoring and analyzing how customers use combinations of points of contact, services, or products to interact with an organization. Companies use customer journey analytics because it is one of the most effective ways to increase long-term customer value, improve customer loyalty, and drive revenue growth. Customer journey analysis provides teams with a window on customer behavior that provides valuable information that they can then use to inform their decisions. These points of contact are called events and they define the customer’s behavioral model across an organization. In a typical big bank, about 100,000 events are carried out per second. Several dynamic events are associated with a bank such as an ATM withdrawal, a POS transaction, a cash deposit. This series of customer journey events can be used for different use cases such as cross-selling, account reactivation, hard rolling, soft rolling to name a few. This paper uses deep neural network-based, recurrent neural network (RNN) algorithm to capture these customer journey events across a bank and how these events can be used to predict cross-sell propensity for other bank products. We developed various RNN models using both time series and static data layers to estimate the likelihood of cross-selling credit card facility to existing customers on hypothetical dataset. A highly predictable model is developed with an AUC of 0.92 on training and 0.90 on validation sample. This model can capture around 91% of the cross-sell events in first two deciles for training sample, indicating that by targeting a small proportion of the portfolio, a bank can achieve maximum conversions from cross-sell programs.

Computer vision technology can be used for instant car damage recognition by analyzing images of damaged vehicles to detect and identify the location and severity of any damages. Technology can accurately classify damage into categories such as small, medium, or high severity. This can help insurance companies and other relevant stakeholders quickly process claims, reduce fraudulent claims, and improve the overall claims process efficiency. The conventional car damage assessment process is time-consuming, labor-intensive, and prone to errors. Computer vision models offer a new solution to detect car insurance fraud by identifying the damage severity and streamlining the claims process. AI can automate the process by analyzing images of damaged cars and generating a breakdown of the damage. The authors propose a unique computer vision process that can help identify small, medium, and high severity of damages and validate investigators' recommendations to detect anomalies in real-time.

Investors increasingly non-financial factors as part of their risk analysis process and growth assessments of corporates. Machine learning (ML) models for predicting ESG scores are an extremely useful tool to help investors make more informed decisions on their portfolios. Such a tool with wide-encompassing alternative data can be useful to the investors. The use of such datasets and machine learning models for ESG ratings can continuously improve the accuracy and reliability of those models. Using machine learning algorithms to identify key drivers of ESG ratings is an effective way of improving portfolio performance. Although the current state of ESG ratings is relatively static, data collection and mapping methodologies are evolving. As more data becomes available, the noise in ESG factors will become less important. This unique document provides a machine learning algorithm for predicting an ESG rating based on a company's financial and non-financial attributes. The financial and non-financial attributes of corporations are extracted from Moody's Orbis and Ratings from S&P. The objective here is to predict the ESG rating of companies where the ESG rating is not easily accessible. At the same time, this approach would allow investors to have a suitable framework for investments based on ESG ratings. With the latest financial and non-financial disclosure by a corporate an ESG score can be predicted which can be used to identify its riskiness with a corresponding increase/decrease of ESG score.