Rupeek’s Japan Doshi shares on Data Science & Asset Backed Lending

Dec 22, 2021

Shares

On Asset-backed lending

Japan shares:

The way asset back lending works is whenever you give credit, you actually look at the value of the asset. Let’s take gold as an asset class, since gold is our current focus, and we will certainly have more assets in the future. When we look at gold and specifically at jewelry since Indians are sitting on a large amount of gold jewelry, you have to look at the purity of the gold. We take a lot of images to figure out how the color changes, how the transformation happens, we collect net weight. We apply the image visual processing to figure out what parts of the jewelry are stone, what part of jewelry is actually the gold. We collect a lot of data, and we use all of this data to figure out the appraisal value of gold. And getting the right value is very important from the regulation standpoint and to manage your risk. You can’t give loans beyond certain values. If we don’t get that right, two things could happen if we don’t appraise it right, either we end up giving more exposure meaning, eventually, customers would not be able to pay back. There’s a lot of emotional value attached to gold, if you don’t basically appraise it properly, the other thing that could happen is you may not end up giving enough value to the customer, so you don’t want to be in either spot. The goal is to be balanced so that the customer gets the maximum value. 

On Data Science

Japan shares:

We have a lean data science team, it is extremely crucial to our success. We have invested in data science across the spectrum in many areas. We use data science to figure out which of the leads are coming to us. We have a higher probability of converting or getting a loan. We are using data science to figure out what is the probability of customers renewing the loan using data science models to figure out what is the probability of a loan going to auction? While it’s unsecured lending, 1% of the customers do get into auctions. It may be a small number but anything that goes into auction is a loss to the business and we try to work on that. We are using data science to automate the appraisal capability. So, we are investing heavily. I think there are very interesting things we’re learning. One of the things we are debating internally is, if we can predict the gold price. It’s really difficult. Gold prices have fluctuated a lot. So what happens is, when the goal prices go down, it really impacts us because customers are not going to go on the sideline. They want to try to come back so that they can get more value. We are trying to figure out at least the moment the gold price goes up to a certain fraction so that we can give customers better products and better loan value ratios. 

To find out more, watch the entire podcast here:

by