The Promise and Peril of Autonomous AI in Finance

By Pratik Shah and Rohit Pandharkar, EY India

The financial services sector has long relied on classical AI for tasks like loan default prediction and customer segmentation. The advent of Generative AI (GenAI) has further revolutionized the industry, enabling large language models (LLMs) to enhance customer engagement, operations, and risk assessment. Now, Agentic AI—a more advanced form of GenAI—is poised to transform the sector by autonomously managing complex tasks, making decisions, and solving problems.

What Is Agentic AI?

Agentic AI goes beyond traditional GenAI tools like ChatGPT. It acts as an intelligent genie within banking systems, orchestrating end-to-end processes that would typically require human expertise—but at greater speed, scale, and lower cost. For example:

• Customer Service: AI agents can dynamically converse with customers, authenticate identities, and retrieve real-time data (e.g., EMI amounts) from internal systems.
• Regulatory Compliance: AI can query vast knowledge bases (e.g., RBI circulars) to answer compliance questions, reducing delays caused by human bandwidth limitations.
• HR & Marketing: AI can convert training documents into engaging videos or tailor marketing content based on audience demographics.

India’s Leadership in GenAI

Indian startups are pioneering Indic LLMs that support multiple regional languages, making GenAI more accessible. The government is also innovating with initiatives like:

• Bhashini API (multilingual translations)
• Anuvadini (image generator with Indian training data)
• National AI Policy (including a public GPU cloud to promote GenAI adoption)

Challenges and Opportunities

While GenAI could add $80 billion to India’s financial services GVA by 2030, banks remain cautiously optimistic. Key challenges include:

• Rapid Obsolescence: New LLMs frequently outperform older benchmarks, requiring modular architectures and continuous monitoring.
• Implementation Costs: High costs and risks (e.g., AI-generated inaccuracies) demand careful planning.