As financial services evolve, AI and advanced analytics are beginning to reshape how lenders assess and serve these segments, moving beyond static credit scores to more dynamic, data-driven decisioning.
Godrej Capital has built several features on its in-house AI platform, Saksham. As part of its AI roadmap, the company has also invested in multiple data initiatives—spanning both external and internal data sources.
Initial business outcomes have been encouraging, with the NBFC improving loan approval rates without compromising its risk thresholds for disbursements. Jyothirlatha B, CTO, Godrej Capital, speaks with FE FUTECH.
Edited Excerpts:
How do you measure the return on data initiatives (RoDI)? Specifically, how do you evaluate whether the returns from external data sources exceed the investment made in acquiring or subscribing to that data?
If you are trying to measure data initiatives in isolation, you are already looking at the problem too narrowly.
Given the interconnected nature of modern data ecosystems, we evaluate success through business-level outcomes rather than standalone technology metrics. Our key measures include uplift in conversion rates before and after integrating alternative data, improvement in approval rates without compromising portfolio quality, and reduction in customer acquisition costs.
We are currently focused on high-impact areas across our existing product portfolio. The effectiveness of external data is ultimately measured by its ability to improve targeting, enhance personalisation, and drive measurable improvements in conversion, efficiency, and portfolio performance.
What are the benefits derived from these data initiatives across functions and areas where they are currently being used?
Data is increasingly becoming a central driver of both growth and efficiency across the organisation.
For our mass-market offerings, data-backed capabilities are helping us deliver hyper-personalised offers, better timing of customer outreach, and more precise channel-level targeting to improve conversion probability.
In our cross-sell business, data is already delivering tangible impact through improved lead quality, stronger customer intent prediction, and higher conversion efficiency.
Beyond revenue outcomes, these initiatives are also strengthening risk management, improving operational efficiency, and enabling more proactive, lifecycle-based customer engagement.
How is Godrej Capital architecting its AI-driven underwriting stack to incorporate alternative data including cash flows, transaction patterns, GST, device data, etc. ?
Our underwriting stack is evolving into a modular, agent-based architecture designed for both scalability and precision.
We already have live agents in place, including the Unified CIBIL Intelligence Agent, multiple conversational agents, the Regulatory Insights Agent, the Customer Service Agent, and the KYC Agent.
In addition, we are currently deploying several specialised AI agents, such as the Financial Insights Engine, Intelligent Document Processing (IDP), and the Triangulation Agent.
Each agent addresses a specific component of the credit decisioning journey. In parallel, we are working closely with underwriting teams to codify complex underwriting scenarios into a structured decision map. This map will eventually power a centralised decisioning agent that orchestrates inputs from all specialised agents to deliver more contextual, consistent, and explainable credit decisions.
Alternative data—such as GST signals, banking cash flows, and digital behavioural insights—will be integrated through our broader data and customer intelligence ecosystem, ensuring all decisions are enriched with real-time, contextual intelligence.
How do you reconcile GenAI’s probabilistic outputs with the need for deterministic, explainable credit decisions—especially in regulated lending environments?
We manage the probabilistic nature of GenAI through a layered validation and governance framework.
Each model undergoes rigorous testing on both seen and unseen datasets, including the use of synthetic data generation techniques. This is supported by a combination of automated validation, manual review, and continuous feedback loops for iterative improvement.
All critical workflows and decisions are designed with a human-in-the-loop architecture, ensuring AI augments decision-making rather than operating autonomously in high-stakes scenarios. This ensures appropriate oversight, accountability, and responsible decision-making.
We also place significant emphasis on transparency by clearly communicating model limitations to users. This helps ensure informed adoption, reduces over-reliance on AI-generated outputs, and reinforces accountability in decision-making.
What changes were required in your data architecture involving governance, labelling, real-time pipelines to make it ‘GenAI-ready’?
Our approach to data architecture for AI readiness reflects a significant shift from traditional data strategies. While our existing stack already serves as a governed single source of truth, GenAI has required us to expand our capabilities to handle unstructured, real-time, and context-rich data. We are also evolving towards an intelligence-led architecture that enables a unified customer context across engagement channels.
To support this shift, we have strengthened our data governance and labelling frameworks to improve traceability, explainability, and auditability. We have also built near real-time data pipelines to support more responsive AI-led use cases, while designing a unified intelligence layer to ensure seamless data availability across systems.
While individual AI capabilities are currently monitored through dedicated dashboards, we are progressively moving towards a consolidated intelligence ecosystem that can drive cross-functional insights, orchestration, and decision-making.
How do you prevent model drift or mitigate hallucination risks in financial contexts?
In financial services, explainability, control, and reliability are fundamental to the success of AI.
To mitigate hallucination risks and ensure consistency, all AI capabilities are routed through Saksham, our enterprise AI platform that governs all external LLM interactions.
Through Saksham, we enforce system-level grounding prompts to ensure outputs remain anchored within Godrej Capital’s domain context.
We also maintain strict data governance policies, including zero leakage of personally identifiable information (PII) outside the ecosystem, along with standardised guardrails for compliance, tone, and responsible AI usage. In addition, we have continuous monitoring frameworks in place to detect model drift and periodically recalibrate models to maintain accuracy and alignment with evolving business, customer, and regulatory requirements.
Which alternative data points are used for credit decisioning?
We leverage a wide spectrum of alternative data to build a more comprehensive and dynamic view of customer risk, particularly for underserved or thin-file segments.
This includes GST data such as turnover patterns, filing consistency, and compliance behaviour; banking transaction data such as cash flow stability, income trends, and expense patterns; bureau-derived surrogate indicators that provide additional signals on the depth and quality of credit behaviour; digital engagement signals based on customer interaction patterns across journeys; and document-derived insights generated through IDP systems.
This allows us to move beyond traditional bureau-centric models and create a more inclusive, accurate, and context-rich approach to credit assessment.
