Why HDFC Bank, Wipro, Palo Alto and Zepto Are Betting on Domain-Specific AI

Why Enterprise AI Is Shifting from Large Models to Domain-Specific Intelligence

Technology leaders from four major Indian and global firms told a summit audience that the real competitive edge in enterprise AI will come not from bigger foundation models but from domain-specific systems trained on proprietary data and governed by tight internal rules. The panel, held at Network18’s Rising Bharat Summit in Bengaluru on July 24, featured the chief information officer of HDFC Bank, the global chief technology officer of Wipro, the India managing director of Palo Alto Networks and the chief technology officer of quick-commerce player Zepto.

Ramesh Lakshminarayanan, group head of IT and CIO at HDFC Bank, said banks must “own their intelligence” because models built on in-house data deliver higher accuracy while satisfying regulatory demands. He revealed that the bank has already built an internal AI platform called ‘me’ with a team of 40 deep-tech engineers, using smaller, tightly governed models for banking use cases such as trade processing and credit card operations.

Wipro’s Sandhya Ramachandran urged enterprises to adopt contextual AI models designed for specific business problems, noting this approach improves accuracy and lowers latency and compute costs. Palo Alto Networks’ Swapna Bapat warned that the rapid emergence of autonomous AI agents creates a new cybersecurity threat surface, demanding zero-trust principles and continuous monitoring. Zepto’s Nikhil Mittal stressed a hybrid workforce vision where humans and AI agents collaborate rather than one replacing the other.

The consensus: as companies move from AI experimentation to production, differentiation will stem from proprietary intelligence, robust governance and secure deployment frameworks—not access to the largest language models alone.

What HDFC Bank, Wipro, Palo Alto and Zepto Reveal About Enterprise AI’s Next Phase

Why Proprietary Data Is the New Moat

The panel’s central argument marks a practical shift in enterprise AI thinking. Rather than chasing ever-larger foundational models, companies are recognizing that their own customer data, transaction logs and operational workflows provide a unique training corpus that no generic model can replicate. HDFC Bank’s ‘me’ platform exemplifies this: a small, focused model fine-tuned on the bank’s proprietary data can outperform a massive public model on tasks like fraud detection or loan underwriting, while also staying compliant with India’s data localisation and financial regulations. For other enterprises, the implication is clear: the AI arms race is moving from model size to data quality and governance.

The Cybersecurity Imperative for AI Agents

Swapna Bapat’s warning about AI agents highlights a tension that few enterprises have fully addressed. As autonomous software that can make decisions and pull data across multiple systems, these agents vastly expand the attack surface. A compromised agent could trigger unauthorised transactions, leak sensitive data or disrupt operations. Bapat’s call for zero-trust principles—where every action by an agent is continuously verified and governed—directly ties to the operational reality that enterprises will soon deploy thousands of agents across supply chains, customer service and IT systems. Without such guardrails, the productivity promise of AI agents could quickly turn into a security liability.

The Hybrid Workforce That Zepto and Wipro See

Both Nikhil Mittal of Zepto and Wipro’s Sandhya Ramachandran emphasised that AI adoption is not about wholesale replacement of workers but about augmenting them. Zepto’s example of automating repetitive tasks to free employees for higher-value work resonates with Wipro’s own Responsible AI Council, which ensures ethical deployment while retraining staff. However, that reskilling burden is real and large. As HDFC Bank’s Lakshminarayanan noted, the bank is already deploying AI in trade processing and credit card operations, but human oversight remains critical for high-stakes decisions. This hybrid approach demands a deliberate workforce strategy—one that many enterprises have yet to articulate in detail.

Three Concrete Steps Companies Can Take from the Bengaluru Summit

For business and technology leaders watching this shift, the summit offered several concrete, actionable insights rooted in the panel’s disclosures:

• Audit your proprietary data assets now. HDFC Bank’s ‘me’ platform shows that in-house models built on internal data can deliver better accuracy and compliance. Enterprises should map their unique datasets—transaction records, customer interactions, operational logs—and identify use cases where smaller, contextual models could outperform generic ones.
• Extend zero-trust to AI agents immediately. Swapna Bapat’s warning is a call to action: as autonomous agents multiply, security teams must implement continuous verification and strict governance for every agent’s actions. This requires new policies, identity management for non-human actors, and real-time monitoring tools.
• Design reskilling programmes for human-AI collaboration. Zepto’s approach of automating repetitive work and redeploying staff to higher-value tasks demands a structured upskilling plan. Organisations should inventory roles likely to be augmented, not eliminated, and invest in training that teaches employees to work alongside AI agents—a gap that Wipro’s Responsible AI Council framework aims to fill.