Risk management in India is crucial for expanding credit access to over 1.3 billion people, most of them either unbanked or underbanked. The need for advanced and accessible AI and ML technologies is essential to creating a unified and centralized system that maintains constant access to ever-growing volumes of data, validates these data from various sources, solves ID and KYC problems and maintains connectivity to end customers for collection purposes. A fast-paced industry with growing access and digitization requires a short time-frame for deployment, which makes transitioning from Excel-based systems to complex, code-heavy platforms an even costlier challenge.
In a recent webinar on “The Future of Risk Decisioning,” Economic Times editor Soushik Gosh, AMU Leasing CEO Nehal Gupta, Tartan Founder and CEO Pramey Jein, and Country Manager for India Varun Bhalla from Provenir discussed the latest trends to solve these challenges. Much like leveraging user-friendly, AI-powered software like ChatGPT, risk managers can now use automated machine learning tools that can build on robust volumes of data in real-time without the need for data scientists or tech experts to operate them. In this article, we explore key guidelines on how credit-risk managers can build AI and ML infrastructure to fit their needs and use diverse data to create a data-driven, competitive organization.
Diversify Sources: Alternative Data + Traditional
To make more informed decisions and explore new business avenues, it is crucial to diversify data sources beyond traditional ones to include and properly harness alternative data. AMU Lending primarily serves customers from rural and semi-urban areas who are new to credit, using alternative, mostly qualitative (unstructured) data harnessing tools to provide its own credit score. Therefore, qualitative metrics for fraud checks, such as field and address verifications, references and neighbor evaluations must be incorporated into an extensive and comprehensive scoring methodology. By combining these alternative data sources with more traditional and public data and applying specific signals, the company can create a comprehensive scoring methodology and provide credit scores to people who have never been banked before.
“Provenir plays a critical role for AMU as their platform integrates and validates data. Visibility only comes after integration and validation, and for a financial institution like ours, integrating Provenir’s validation system with our core loan origination and management system allows us to make sense of the data and make informed decisions, such as whether or not to serve a customer. This integration and validation of data points is how we can solve the issue of non-visibility of data.”
Nehal Gupta, AMU Leasing
To integrate systems and reach automated decisioning, start small
Due to disaggregation of data and lack of key primary identification in India, building a comprehensive customer view becomes harder, especially for credit new-comers. Integrating the data required to do this from multiple sources to make informed decisions can be a vexing exercise, made worse by legacy systems that fail to spot valid correlations to inform and streamline decision models. Furthermore, many risk managers have to rely on external vendors to run these outdated legacy systems. In looking to fix this, finding tech-savvy personnel in-house who understand the symbiotic relationship between code and data becomes another hurdle.
To navigate these issues, risk managers can choose to work with experienced partners and leverage ChatGPT-like platforms that can easily aggregate data via a single API and then apply machine learning capable of real-time data refinements, without the need for data scientists. The goal is to reach a point where the right data facilitates immediate model building, fueling the decision process. Sounds like a lot. Yet a feasible first step towards automation can be as simple as defining the risk tolerance threshold for a customer segment, then automating smaller loans or your standardized loan parameters for your key ticket size. Machine learning can then be deployed to streamline the credit scoring process. Finally, once fully functional, the technology can scale to other segments or product-types.
Keep Front-Ends Simple and Explainable through AI
Implementing machine learning requires understanding the automated decisioning process, not blindly trusting “black box” technologies. AI and ML systems need to be transparent and explainable to every user in the loan life cycle, and to regulators, as compliance requirements catch up with technology. The ideal platform should include an explainable dashboard, enabling detailed insights into each executed model for every customer. Simple visualizations should account for the impact of each variable for an individual user, making the system both trustworthy and accountable.
But it’s not just about simplicity for understanding decisioning models. Add to that the need to keep front-ends and interfaces in general as simple as possible for both corporate clients and end-users with little to no experience with tech products. The ideal platform should integrate seamlessly with both the lender and the client from the Loan Origination side, while allowing the flexibility to deploy friendly, customizable interfaces for the end-user.
Ensure Real-Time Access for Faster Time-to-Market
Real-time access to data sources is crucial for creating the right products for different customer segments such as gig workers and underbanked employees, in record time-to-market. While traditional methods of collecting data can be time-consuming and risky, the ideal platform should offer key income and employment data via a single API after the user’s consent, accelerating time-to-value per customer. For new-to-credit customers who have been unbanked in the past, credit scores may not be accessible and the ability to use both structured and unstructured data must function at optimal speed after proper integration, allowing the lender to aggregate references, payroll data, GST data, and utility payments, among others, in real-time. This enhanced workflow prevents customer dropouts, especially among those looking to acquire their first financial products.
Create Consent Architecture for Compliance
Lastly, to ensure compliance with rapidly changing regulations and protect user privacy, it is important to have proper consent architecture in place, as well as the right tech partner to maintain it. This means ensuring that data is collected only with user consent, and the systems used to store and process data are compliant with regulations. In India, Reserve Bank of India (RBI) guidelines require that financial institutions prioritize compliance. It’s essential to ensure that only necessary user information is stored and that the data collection process is non-intrusive. Following global gold standards in data regulation and compliance like the EU’s GDPR is important for both technology and practitioner standpoints.
“When it comes to security and compliance, it all comes down to the partnerships that you establish. It’s about trusting partners, seeing what the ethics of the companies and partners you work with are. Since we are a financing company and technology is not our core expertise, we have always worked in partnership with technical, industry experts we trust, like Provenir.”
Nehal Gupta, AMU Leasing
Building a future-proof, data-driven organization in the financial sector in India involves establishing a “culture of data”: putting in place the right infrastructure for storing and accessing data; adopting a curious, scientific (though not necessarily technical) mindset; implementing state-of the-art, error-free decision-making systems; and embracing technology adoption. By prioritizing these areas, financial institutions in India can build a strong foundation for success and navigate a rapidly changing landscape.