Income plays a highly influential role in a lender’s assessment of a customer’s loan repayment ability. However, as we move into a full-fledged digital lending space, where decisions are taken in a few clicks and without the hassle of manual interventions for document verifications, accurate data for assessment of customer’s income is very limited.
In the current lending landscape, lenders are pushing hard to go digital and are using various bank statement scraping tools to identify an individual’s salary. There are two major problems associated with this approach:
1. Bank Statement Scrapper API based income data is available for hardly 10% of the customer base.
2. Bank Statement Scrapper API based income assessment also hampers the customer’s user experience, thus leading to significant drop out rates in the app journey.
To overcome these challenges, CreditVidya has developed Income X – an alternate data product that leverages advanced AI-based algorithms and is powered by state of the art alternate data feature extractions. Additionally, it’s an SDK based solution having zero impact on the user experience as it relies on a service that runs in the background and takes less than 5 mins of integration time with any Android app.
In the current case study, we are showing how Income X solution fairs with traditionally used Bank Statement based salary identification.
1. True Salary: Algorithm that identifies the exact salary that a customer earns much like that being done by bank statement scraping tools.
2. Group Behavior Income: A Deep Neural Net Model that predicts likely income earning band of a customer at a 90% accuracy.
True Salary vs salary obtained from bank statement scrapping tools:
We assessed the accuracy of True Salary against one of the leading Bank Statement Extraction API. As expected we observed 99% correlation between True Salary & Bank Statement Salary. The below chart highlights the strong performance of True Salary.
Having established the accuracy of True Salary against Bank statement Salary, we used it as a benchmark to test the performance of Group Behavior Income. The findings were significant:
1. We predicted the exact Group Behavior Income for 86% of the users.
2. We ensured that for 90% of the users the predictions were below the Income Group if not the exact, thus ensuring low over prediction rates.
Fig 1.2 (Neural net model achieves 90% accuracy in predictions)
Fig 1.3 (Prediction accuracy achieved for various income buckets)
CreditVidya’s Income X solution can thus enable seamless digital lending, by providing an accurate source of Income for the customer’s flowing through the lender’s loan sourcing app.
1. For a digital lending app, among the users who have shown intent by verifying their mobile number Income X Solution was able to provide income for 700% more users compared to the existing solution using bank statement scrapping tools
2. Income X has enabled lenders to get customers’ affluence within 1 minute time compared to 24 Hours+ TAT for physical statement collections OR document uploads.
1. A leading NBFC has increased their current approvals rates by 2X for their Consumer Durable applicants using Income X as the major source of information.
2. Another NBFC has started Cross Sell/ Up Sell programs basis the affluence levels estimated using Alternate data and pre-approving customers for higher ticket size credits.
3. One of the pioneers in digital lending has introduced Income X in their flow before calling Bank Statement Scrapper APIs to improve customer experience and control drop off rates.