Staff Activity Monitoring project aims at monitoring suspicious inquiries/activities done by the Bank staff and avoid fraudulent transactions to keep the money safe.

Data is fetched to Big Data Lake from 4 sources, 2 being real time and 2 hourly sources, Spark logic is executed on this data and alerts are generated on an hourly basis. Some of the alerts includeOdd hour inquiries, Cross zone account inquiries, money transfer from Office accounts to Staff's account, etc.

These aletrs are then pushed using Sqoop export feature to the Transaction Monitoring team who further investigate these suspicious activities and take necessary actions.

In this project, we are analysing the communication (SMSs and E-mails) sent by the bank to all the customers.

The communication data is used to generate multiple insights like business-wise/division-wise distribution of SMSs/E-mails, segregation of Financial/Non-Financial SMSs/E-mails, Repeated(duplicate) SMSs/E-mails being sent, etc. These insights are showcased to all the departments of the Bank on a daily basis, which helps them to manage the costs associated with the communication.

This project also highlighted that the vendors were charging the Bank on "Submitted" SMSs model, which was then changed to "Delivered" SMSs model.

ReKYC project aims at creating a customer list who are ReKYC due and contacting them in advance to get the ReKYC done.

Data is fetched from around 8 sources across the Bank and Spark logic is applied to exclude customers who have done deemed KYC by buying some of the Bank's products like - buying a credit card, ordering a cheque book at home, courier delivery at home address, loans etc.. The final list is then sent to RBI for verification and customers are contacted for doing ReKYC.

As this was an RBI compliant project, there was a rigorous UAT involved.

FraudPOC project aims to find a common Point of Compromise(POC) where the suspected fraud must have taken place.

We have created a utility in Apache Hue to get inputs like Credit/Debit card numbers, Start and End Date, threshold, etc. Whenever customers report a card misuse, these card numbers are entered in the utility, spark logic is applied on the transaction data to find the top N suspected points (Merchants/ATMs/etc.) where these cards might have been misused. The investigation team then manually investigate such compromise points with the help of Police and take it to closure.

Once top N suspected points are under investigation, as a pro-active measure we find all the cards used at these Merchants/ATMs in that time frame and the customers are requested to change their PIN or a new card is issued.

Adaptive Cognizance is an ensemble model which is acts as a Next Best Offer Recomendation model for Prepaid customers.

Telecom data viz. recharges data and the KPI metrics were used to analyze and develop machine learning models which will predict the next best prepaid product/recharge (from the available master products) for the customer based on his usage-profile matrix and the product-profile matrix.

We implemented this model in our campaigning product as a Packaged Analytics model, so that a non-data-scientist can also execute this model and analyze the results. The model output was used by the Analytics team to run campaigns and revenue comparison was done on the basis of Target/Control groupsin order to judge the efficiency of the model.

Churn analysis in Telecom is to predict the Customer Churn and provide an aggressive offer so that the customer is retained.

Telecom usage metrics data was used to study the customer behaviour and this data was used to create machine learning models which would predict if a customer will churn or not. The accuracy of the model was calculated using precision/recall so that an optimum solution was developed which will help the client to retain customers more rather than reducing spends on false positive churners.

This model was also incorporated as a part of the company's campaining product and the model output was used to run multiple campaigns. The Churn model was used by ~20 global clients

Customer Genome project aims to study the genomics(behaviour), i.e. the transaction patterns of the Bank's customers and provide them with relevant offers to elevate the revenue.

Customer transactions were churned from different sources like Finacle, Debit cards system, Credit cards system, Loans, etc. and using set of rules the customers were tagged to specific events. For e.g. a customer can be tagged as a "Heavy Electronics Buyer" and "Likes Chinese Cuisine", etc. There were 25 daily and 200 monthlytags/events.

Based on the customer tagging, multiple campaigns were organised to provide best offers to the customers. The campaign results were used to fine tune the tagging logic so as to reduce the false positives.