One Bank One Flow
One Bank One Flow A bank’s primary income is earned by the interest it charges on the loans and to give loans, the bank needs deposits. In OBOF, at the primary level, the goal was that the money which is going out from one of the customer’s bank accounts should flow into another ICICI customer’s account ——-
Project Overview For any transaction that a customer of the bank does, there could be two outcomes on the side of the counterparty: Money is flowing to an ICICI customers account Money is not flowing to an ICICI customers account Our main target was to minimize the transactions happening in the second category. We only looked at corporate customers since both their transaction size and volume are huge. The main challenge in this project that I faced was the absence of a unique key to identify whether the counterparty was an ICICI customer or not. Because if the counterparty is an ICICI customer, the account manager can route the payment to the existing ICICI account. If they are not ICICI customers, this becomes a new customer acquisition opportunity.
The only way this distinction could be made was on the basis of name. We needed to match the name that was appearing in transaction data with the ICICI customer base name, to see if they already had a relationship with us. But the issue with names is that they are not standard. For example, ABC Private Limited can also be written as ABC P. Ltd. or as ABC Pvt. Lt. and many more. I would take out the transaction data for a particular quarter for ICICI corporate customers. Then I would move ahead with standardizing the name that was appearing in the transaction using a list (For example P. should change to Pvt. Private should also change to Pvt.). I would also do this standardization for ICICI customer base data. Then I would get the Jaccard similarity index for transaction name and ICICI customer name. Wherever the result for the similarity index was more than 65%, that would be considered a match. Rest would be treated as new customer acquisition leads.
The performance of the project was measured by looking at how the balances for these customers were changing, how much change did old customers contribute, and how much new customers did. For new customers, we would also see how many products they have with us so that the stickiness would increase.
Tech Stack Data Extraction: SQL Data cleaning(name standardization): Python Data analysis(Name matching): Pyspark Measuring performance: Initially SQL and SAS then moved to SQL and PowerBi