AI AML/CFT Lending & Credit Risk

An AI innovator’s take on banking’s future: Amplifying human expertise

Sriram Tirunellayi

April 30, 2025

Read Time: 0 min

6 questions & answers on how generative AI is shifting the landscape

Abrigo's Director of Applied AI, Sriram Tirunellayi , shares advice and insight into AI, its role, and how it can power the future of banking.

Generative AI for safer, smarter growth

As a seasoned leader in data science and AI for financial services, Sriram Tirunellayi brings a practical perspective to the evolving role of artificial intelligence in banking. He holds multiple patents for innovative machine-learning and AI applications that address some of the industry’s most complex challenges. Today at Abrigo, he drives the strategy and roadmap for generative AI and applied AI/ML products, helping financial institutions harness emerging technologies to achieve safer, smarter growth. As Director of Applied AI, Tirunellayi is focused on bridging innovation with trust, finding new ways to power the future of banking through responsible AI.

In this Q&A, he shares how generative AI is shifting the landscape — not by replacing human expertise, but by amplifying it. He discusses where banks and credit unions can find the greatest opportunities to apply AI thoughtfully, why governance and explainability are critical, and how emerging technologies like agentic AI could reshape financial services. His insights offer a clear, balanced view for institutions looking to embrace AI as a long-term strategic capability.

Need help navigating AI? Access additional AI resources for bankers

get resources

What role does generative AI play in banking today?

Generative AI is one of the most exciting innovations in recent times that has the potential to make a profound impact across every aspect of banking and lending.

There are plenty of applications in banking that use traditional machine learning, a form of AI that is very good at finding patterns in structured numerical data. However, these applications are typically focused on solving specific tasks such as assessing credit risk, detecting fraud, or optimizing marketing spend.

Generative AI is unique in the sense that it works very well on unstructured data, leveraging foundational large language models (LLMs) that have been trained on web-scale data. Today's LLMs don’t just understand language; they follow instructions and respond with the fluency of a human expert. This has opened up a whole new set of applications in banking, where generative AI can be used to address challenges that were previously out of reach, such as:

Understanding free-form customer interactions
Distilling insights from complex documents
Generating personalized communications at scale
Reimagining how bankers interact with information.

Generative AI introduces a new layer of intelligence that enables automation and amplification of human expertise. For the first time, we have not just a tool but an intelligence amplifier that processes complexity, uncovers insights, and enhances human judgment in real time. This shift from prediction to understanding and interpretation is what makes this moment exciting and different.

Abrigo serves 2,500 financial institutions. What is your advice for banks and CU looking to adopt AI effectively?

First and foremost, it’s important to recognize that at its core, AI is advanced technology powered by algorithms and data — not a superintelligent entity. This framing matters because it sets realistic expectations for what AI can and cannot do, especially in complex industries like banking and lending. Rather than thinking of AI as a replacement for human expertise, it’s more helpful to view it as a powerful tool that can amplify human judgment, streamline time-intensive processes, and unlock new opportunities where traditional methods fall short.

The pace of innovation in AI is accelerating, and for banks and credit unions, keeping up is a strategic imperative. Financial institution leaders should prioritize building internal awareness, understanding how generative AI is evolving, where it’s being applied, and what it means for the future of financial services. As with adopting any emerging technology, it is crucial to understand both its potential and its boundaries. This ensures financial institutions deploy it in ways that are secure, compliant, and aligned with long-term business and regulatory objectives.

To identify meaningful opportunities, prioritize high-value areas marked by high repetition, time-intensive processes, and complex decision-making that depend on unstructured data and expert judgment, such as credit underwriting, loan servicing, compliance monitoring, and customer communications. These are domains where generative AI can amplify expertise, streamline operations, and unlock new efficiencies.

Equally important is establishing a strong governance framework from the outset, ensuring responsible use, managing risks, and maintaining regulatory alignment. Starting with targeted pilot programs in well-scoped use cases allows institutions to learn, adapt, and scale with confidence.

Over time, the institutions that succeed will be those that treat AI not as a one-off initiative, but as a long-term capability—embedded into strategy, operations, and culture. The key is to start early, learn continuously, and scale thoughtfully.

Credit risk modeling and underwriting have traditionally relied on historical data. Compared to traditional methods, what improvements have AI-driven models made in assessing borrower risk and improving the borrower experience?

If you peek inside the underwriting box, three major steps shape a credit decision.

First is the gathering of borrower data. One of the key risks in lending stems from information asymmetry — an imbalance of information between the lender and borrower. Traditionally, lenders have relied heavily on credit bureau data, which provides useful insights but can miss important, real-time changes in a borrower’s financial situation. Expanding the scope of data to include alternative sources — such as bank transactions, real-time updates on financial statements, income, and employment status — helps reduce this gap and gives lenders a more complete and timely understanding of borrower risk.

AI plays a critical role by labeling and categorizing alternative data, such as bank transactions, to provide better visibility into cash flows and improve the accuracy of downstream ML models. Large language models (LLMs) are also being applied to read diverse formats of financial statements and automate the spreading of key financial information, significantly enriching underwriting data inputs with greater speed and precision.

Second is predicting the borrower's future payment behavior. AI models now leverage cash flow data and alternative datasets to more accurately predict credit risk, helping lenders better determine who to lend to and on what terms. By tapping into broader data sources, financial institutions are also expanding access to credit for underbanked and populations with thin credit files. Additionally, AI-powered analytics help assess the performance of automated decision models and identify potential biases, supporting the move toward fairer, more transparent lending practices.

Third is the final decisioning process. Credit analysts define the risk criteria, underwriting policies, and decision rules that govern how borrower data and risk scores are evaluated. AI assistants increasingly support this work by gathering relevant information from multiple sources and drafting the initial credit memo narrative, significantly reducing time and manual effort. AI also helps automate complex underwriting logic, define test cases, and identify potential gaps in decisioning frameworks — driving greater efficiency, consistency, and governance across credit operations.

Compliance teams often worry about AI creating rather than solving compliance challenges. What advice about compliance risk do you have for BSA Officers or risk managers evaluating AI solutions? Are there ways to balance automation and the human element?

It is important to recognize that while AI solutions can enhance compliance operations, they can also introduce new risks if governance is inadequate. My advice to BSA Officers and risk managers evaluating AI tools is to start with a comprehensive risk management framework that emphasizes the following:

model transparency
explainability
fairness testing
clear documentation of model design and validation

Institutions must ensure that human oversight remains embedded within all AI-supported compliance activities, particularly for high-risk functions such as transaction monitoring, customer due diligence, and sanctions screening. Automation can assist in prioritizing and surfacing potential issues, but final decision-making and accountability must always reside with qualified compliance professionals.

Finally, AI governance should be treated as an active, ongoing process — not a one-time certification exercise. Strong programs will rely on cross-functional teams that continuously audit, refine, and improve AI systems as regulatory expectations and risks evolve.

One of the biggest concerns for financial institutions is explainability—regulators and decision-makers need to understand AI-driven outcomes. How does your team ensure AI-powered solutions are transparent and interpretable?

Explainability is fundamental to the way we design and deploy AI in Abrigo's solutions. Our approach begins with prioritizing inherently interpretable models wherever possible, particularly in high-stakes areas such as credit decisioning. When more complex models are necessary, we implement dedicated explainability layers that clearly trace how specific inputs contribute to each outcome, ensuring transparency at every stage.

We also embed “human-in-the-loop” processes throughout the model lifecycle. Subject matter experts validate AI-generated outcomes against structured test scenarios, supported by AI evaluation tools (AI evals) that systematically measure outputs against defined benchmarks such as factual accuracy, completeness, relevance, and clarity. This combination of expert review and automated evaluation ensures a rigorous, unbiased assessment of model quality. Only models that meet or exceed our established thresholds are approved for production use.

Finally, we maintain a comprehensive governance framework that includes detailed model documentation, audit trails, independent validation, and ongoing monitoring and re-validation of AI systems. This approach ensures that our solutions consistently meet regulatory expectations, uphold ethical standards, and sustain stakeholder trust over time.

What emerging AI-driven banking applications excite you the most right now?

One area that excites me is breakthroughs in “mechanistic interpretability.” This is where researchers are beginning to map the "thought processes" of large language models. It's like building an MRI for AI: a way to look inside, map how these systems think, and understand the concepts they're using. As we learn to see more clearly into these models, we have a real opportunity to build AI that supports fairer credit decisions, smarter risk assessments, and more transparent banking experiences — not just faster ones.

I’m especially excited about the area of agentic AI. Unlike traditional generative AI applications, which rely heavily on human prompts, agentic AI applies sophisticated reasoning, planning, and self-correction to tackle complex, multi-step problems on its own. These systems are designed to work more like human teams — managing tasks independently, collaborating dynamically, reflecting on outcomes, and improving with each cycle.

A recent example of agentic AI involves Capital One’s “Chat Concierge” application. It moves beyond basic chatbot interactions by breaking down a complex task — like purchasing a car — into multiple coordinated steps handled by different AI agents. Instead of simply answering questions, the system can autonomously schedule appointments, estimate trade-in values, and manage other related tasks, working collaboratively to guide the user through a multi-step process with minimal human intervention. The goal? Make car-buying a less overwhelming and more enjoyable experience.

While agentic AI holds enormous promise, it is still early; many systems remain brittle, error-prone, and reliant on human oversight. Yet its true potential is clear: to move beyond automation and build intelligent systems — and financial services — that are not just faster and smarter but genuinely aligned with human needs.

By expanding access, promoting fairness, and rebuilding trust, agentic AI can help shape a future where technology serves people, not the other way around. That possibility is what excites me most.

Find out how Abrigo Fraud Detection uses AI to stop check fraud in its tracks.

fraud detection software

Blog