Risk assessment and credit profiling for new to credit customers is tricky. Along with it, there is great scope for improving the accuracy of risk prediction for credit-served customers too. For NAVI, we discovered location attributes that indicate the risk appetite of prospect users. Next, we augmented their user database with these location variables/ attributes to predict risk better and help them identify new customers, which would otherwise not be considered for credit products. Here’s how we did it.
The lack of data for credit profiling NTC customers has led to a significant proportion of the adult Indian population being unserved or underserved regarding credit. A TransUnion report, “Empowering Credit Inclusion: A Deeper Perspective on Credit Underserved and Unserved Consumers”, highlights that more than 160 million consumers had insufficient access to credit in India at the end of 2021. The report also highlights the percentage population that falls under credit served and underserved from a global perspective.
For the ‘new to credit consumers’, risk assessment is a challenging feat simply because of the absence of credit history or score. This becomes an impediment for such users to avail credit opportunities. Such users are traditionally termed as unscorable, and lenders face a “chicken or egg” conundrum of how to provide the first credit product without a credit history.
“Although India has made great strides in increasing levels of credit inclusion across the country in recent years, the current reality highlights the importance of incorporating enriched credit data into the lending ecosystem, so that fewer consumers find themselves as credit unserved. Once these consumers can be evaluated by financial institutions, lenders can better determine where there might be new opportunities for growth and how they can expand credit inclusion further,” said Kumar, MD & CEO of TransUnion CIBIL.
So how could these ‘new to credit’ consumers be evaluated better for risk and credit profile? That’s where GeoIQ comes into the picture.
Tapping Into the New To Credit Segment
NAVI has made an important place for itself in the credit market, especially in the quick cash loans and personal loans segment. With more than 17 lac customers, 10 million plus app downloads, and more than INR 9000cr worth of loans disbursed, the fintech major was struggling with evaluating the ‘new to credit’ consumers.
The problem statement that we were trying to solve for NAVI were:
- Cold Start Problem — they wanted a method/ basis to evaluate the first-time credit buyers to understand their credit risk
- Improve Risk Prediction Accuracy — They wanted to augment their existing data with location attributes to improve the accuracy of their risk prediction models
Solving Risk Prediction Woes with Hyperlocal Data
Location attributes are a great indicator for understanding user behavior, for example, information such as average rentals, presence of brands, and other variables define the characteristics of the user segment in the target location.
Now, this location data lies external to the organization, so how do we harness it?
We source data from more than 600 data sources (government data sources, public data sources, satellite data, and others), but this data is unstructured and not directly consumable. We then transform this data into a structured and usable format for direct consumption or as input to our ML models.
For NAVI, we designed a custom ML model to augment their existing user database with location attributes (based on their addresses).
The first step is data discovery, i.e., identifying what are the variables that impact the risk assessment for a user. Listed here are some of the attributes or variables identified by the model to have a direct impact on users’ credit profiles:
- Presence of offices
- Number of residential apartments
- Population density
- Presence of gyms
- Average rental
- Average meal cost for two
- Distance to the nearest bank
- Presence of ATMs
- Presence of tertiary roads
- Presence of electronic stores
Now, when the existing risk assessment process is augmented with such attributes at a granular level, the impact is huge as is shown below:
By enabling location insights on top of the existing risk prediction model, NPAs reduced by 10% and book size increased by 20%. NAVI has processed more than 30 million applications on this new model augmented with location data.
Conclusion
Apart from risk prediction or credit profiling for lending firms, location insights could also impact collections and help understand user affluence for targeting different fintech products. With current models and processes built over data internal to the organization and mostly through online channels, there is scope to improve prediction accuracy across fintech use cases. And this could be done by augmenting the existing models with location data like we have seen here for NAVI.
For more information on our products and services, you can visit www.geoiq.io or get in touch with us at hello@geoiq.io.