How TrueBalance Is Rebuilding India’s Credit Infrastructure With AI

In India, millions of salaried and self-employed individuals earn steadily but remain outside the formal credit system. Traditional bureau-led underwriting often fails to reflect their real financial behaviour. TrueBalance, operated by Balancehero India, is attempting to address this structural gap using an AI-led alternative credit infrastructure.

In a conversation with CiOL, Charlie Lee, CEO, TrueBalance, outlined how AI credit infrastructure is evolving in India. The discussion explored alternative credit scoring, explainable AI, regulatory guardrails and the future of responsible lending.

Interview Excerpts:

Credit decisioning in India still relies heavily on incomplete bureau data. From a systems perspective, is the real risk today bad borrowers or bad credit infrastructure that fails to capture repayment reality?

From a systems perspective, the challenge is less about “bad” borrowers and more about the inability to underwrite them properly. A large segment of India’s middle class and the next billion consumers, earning roughly ₹20,000 to ₹1 lakh per month, remain outside the legacy financial ecosystem because bureau data does not fully capture their financial behaviour.

Our approach has been to build an Alternative Credit Scoring (ACS) model based on alternative data collected from smartphones. Using AI technologies such as LLM, NLP and named entity recognition, we underwrite users who are otherwise excluded. Despite high application volumes, our final default rates are around 4 per cent, which suggests that better infrastructure and better signal reconstruction can improve risk assessment without lowering standards.

Many lenders use AI for speed, but regulators increasingly worry about opacity. How does moving from black-box models to mathematically explainable AI change accountability in lending outcomes?

We collect and process more than 5,000 datasets from Android devices. These include SMS data, bank statement information and behavioural signals. Using LLM-based SMS parsing, CharBERT NLP to decode unstandardised bank statements, and named entity technologies, we extract financial events from unstructured data.

The key difference is explainability. We can explain credit decisions based on alternative data, for example, patterns in app usage, such as engagement with banking or credit services versus non-financial apps, as well as communication behaviour. Explainable AI is not just important for regulators; it is equally important for users. It enables transparent and auditable decision-making and aligns with compliance expectations under RBI’s Digital Lending Guidelines.

Generative AI is often positioned as an interface layer. In your work, how does GenAI materially improve risk modelling and default prediction rather than just customer interaction?

Generative AI plays a core role in our underwriting engine, not just in customer interaction. Once a user provides consent to access data, our model processes alternative signals and generates four critical scores within minutes for pre-approval:

• Probability of Default (POD)

• Credit Limit Optimisation Score

• Risk-Based Pricing Index

• Collection Score

Our Alternative Credit Scoring system relies heavily on generative AI technology. We have filed a patent for a generative AI-based probability of default model that applies token-level log probabilities in consumer lending. This strengthens risk prediction and supports underwriting, pricing and collection strategies in a unified framework.

Given that AI-led credit decisioning increasingly depends on reconstructing incomplete data, how far does the Union Budget go in strengthening India’s credit and data infrastructure so that fintech lenders can deploy explainable, compliant AI models?

India has built advanced financial infrastructure through initiatives such as eKYC, UPI and the Account Aggregator framework. These provide organised guardrails for fintech players. Clear rules and established technology standards reduce uncertainty, which is critical for deploying AI models responsibly.

When digital identity systems and consent-based data-sharing frameworks are strong, it becomes easier to implement explainable and compliant AI systems. Regulatory clarity and structured infrastructure are essential for sustainable innovation in credit.

Financial inclusion is frequently equated with higher risk tolerance. How can AI-led reconstruction of missing credit signals expand access without increasing systemic credit risk?

Without AI-based alternative credit scoring, it would be difficult for our target users to access credit easily. Our system analyses over 90,000 alternative data sources, including SMS, bank statements and behavioural signals, to reconstruct creditworthiness where bureau data is incomplete.

This is about smarter assessment, not taking extra risk. Because underwriting is based on multidimensional behavioural insights, we can extend access while maintaining disciplined risk management. The default performance we see reflects that inclusion does not necessarily require higher tolerance for losses.

As regulators globally push for responsible AI and auditability in financial systems, do you believe explainable, research-driven AI will become a regulatory requirement rather than a competitive differentiator in lending?

Explainable AI is increasingly becoming the norm. Regulators are moving toward stronger expectations around transparency and auditability. We focus on demonstrating how our AI model works through patent filings, technical research engagement and measurable performance outcomes.

As oversight frameworks mature, explainability will likely shift from being a differentiator to being a baseline requirement in digital lending.

TrueBalance recently announced its transformation into an AI-powered financial platform designed to expand access to formal financial products. The company states that its AI-as-a-Service lead generation model has delivered an average month-on-month growth of 43 per cent over the past two quarters. It has filed six patents, including one covering its generative AI application in credit risk prediction.

With partnerships such as Prefr and Upswing, the platform reports reaching over 90 million customers nationwide. Operating with NBFC and PPI licences, Balancehero India says its long-term objective is to become an AI hub for financial inclusion by integrating more banking and NBFC partners into its ecosystem.

It reflects a broader shift in digital lending, from speed-led growth to infrastructure-led accountability. As AI systems become central to credit decisioning, explainability, regulatory alignment and data integrity are emerging as defining pillars of responsible expansion.