Banks are learning that scaling AI requires more than models, with context, governance and business value emerging as the real differentiators.
Banks are aggressively experimenting with artificial intelligence, but many initiatives continue to struggle to move beyond pilot stages as organisations grapple with fragmented data environments, governance challenges and operational readiness gaps.
According to Kishan Sundar, CTO of Maveric Systems, the issue is not a lack of AI ambition. Instead, banks are discovering that scaling AI requires more than deploying models or launching proof-of-concept projects.
“The reality is that every bank is embarking on AI initiatives because AI has become a strategic priority,” Sundar told CRN India.
Different business units, technology teams and software vendors are pursuing AI-led initiatives, often in parallel. Yet a significant number of projects fail to progress into production environments.
“We know that a large percentage of AI projects never move beyond the proof-of-concept or pilot stage,” he said.
For Sundar, one factor increasingly determines whether an AI initiative succeeds at scale: context.
“Context is king,” he said. “If data is fragmented and context is not established, the system cannot understand why a query is coming or how reasoning should happen.”
Four foundations for AI at scale
According to Sundar, banks looking to operationalise AI at enterprise scale need to establish four foundational capabilities before focusing on models and applications.
The first is a formal AI governance structure.
“Before we engage with a bank, one of the first questions we ask is whether it has a well-defined AI governance council,” he said. Without governance, AI initiatives can become fragmented across business units and technology teams.
The second is cloud readiness.
While AI adoption does not necessarily require a cloud-first strategy, Sundar believes banks that have not embraced cloud infrastructure often struggle to scale AI workloads effectively.
The third pillar is data governance.
Many banks continue to operate with siloed data estates despite years of investment in data lakes, warehouses and marts. According to Sundar, this remains one of the primary reasons AI projects fail to move beyond pilot stages.
“Many AI pilots fail because the data being fed into the models is incomplete, fragmented, or lacks sufficient context,” he said.
The fourth requirement is the creation of a contextual knowledge layer.
While banks often possess multiple repositories of enterprise data, AI systems require relationships and business context to be established across those environments.
Sundar said organisations should think beyond traditional data architectures and build what Maveric describes as a knowledge layer capable of connecting fragmented information sources.
“If an organisation is unwilling to build that contextual knowledge layer, we often discourage it from moving ahead with large-scale AI adoption because the agents will struggle to reason effectively without context,” he said.
Governance becomes the gatekeeper
As banks push AI initiatives closer to production, governance teams are playing an increasingly influential role in deployment decisions.
Although India does not yet have a dedicated AI regulation in force, Sundar said financial institutions are proactively building governance frameworks around AI risk, compliance and accountability.
“The governance teams within banks are becoming increasingly stringent, and in many cases they are one of the reasons why AI pilots do not immediately move into production,” he said.
According to Sundar, many organisations are choosing to err on the side of caution as they navigate emerging risks around AI adoption.
“What we are seeing today is that governance teams are generally choosing to err on the side of caution. Rather than allowing potential risks to pass through, they are applying very strict controls and scrutiny before approving AI initiatives for production,” he said.
The shift is creating a new challenge for banks and technology providers alike: balancing innovation with regulatory confidence.
Production AI demands measurable outcomes
Production readiness is defined less by technology milestones and more by measurable business outcomes. In customer service environments, for example, AI-powered agent assist systems are assessed using metrics such as intent accuracy, next-best-action effectiveness and reductions in call resolution times.
According to Sundar, the objective is to ensure that customer intent is identified before the agent spends time diagnosing the issue, while recommendation engines consistently guide agents towards the correct resolution path.
Operational metrics such as call resolution time and hold time are also closely monitored to determine whether AI is delivering measurable business value.
The same philosophy applies to software engineering and technology modernisation projects.
Sundar said organisations adopting AI-assisted development must balance productivity gains with governance, traceability and cost management.
“For us, one of the most important metrics is whether development velocity improves without driving up costs,” he said.
At the same time, auditability, testing discipline and software quality controls cannot be compromised.
“Code should still pass peer reviews, functional testing and unit testing,” he said, adding that test coverage should increase rather than decrease as automation expands.
Building trust into agentic AI
As banks begin exploring agent-based AI systems, Maveric is placing increasing emphasis on explainability, auditability and human oversight.
Sundar said the company’s Safe AI by Design framework is built around three core principles: explainable AI, human-in-the-loop controls and governance guardrails.
“Every decision made by the system must be explainable and traceable,” he said.
Agents are also designed to escalate exceptions for human review rather than operate without oversight.
“The system is not designed to operate in a fully autonomous manner. Human oversight is always embedded into the process,” Sundar said.
The framework additionally incorporates continuous bias monitoring and audit logging aligned with industry standards and evolving regulatory requirements.
According to Sundar, banks increasingly expect AI systems to demonstrate the same levels of transparency, accountability and control that already exist across traditional technology environments.
Moving beyond experimentation
A recurring theme across banking AI deployments is the need to move from experimentation to operationalisation.
Sundar said the industry’s focus should shift away from adopting AI for its own sake and towards solving specific business problems.
The company has built several of its AI platforms and solutions alongside live banking customers and production environments rather than as standalone research projects.
As a result, the emphasis is on deploying systems that can withstand operational, governance and regulatory scrutiny rather than simply demonstrating technical capability.
“AI should never become a technology experiment,” Sundar said. “Every initiative must be tied to a clearly defined business value.”
For banks seeking to scale AI, the challenge may no longer be access to models or technology. Instead, success will depend on governance maturity, contextual data foundations and the operational discipline required to move from pilots to production.
FAQ
1. Why is AI in banking still largely stuck in pilot stages?
Banks are finding it difficult to scale AI due to foundational issues such as fragmented data, weak governance, and lack of enterprise-wide readiness.
2. What are the key underlying challenges slowing AI adoption?
The biggest barriers include poor data governance, siloed systems, and limited cloud maturity, all of which restrict the effectiveness of AI initiatives.
3. How does cloud readiness influence AI scalability?
Without adequate cloud infrastructure, banks lack the scalability, agility, and processing capability required to deploy AI solutions at scale.
4. Can AI play a role in solving these foundational gaps?
Yes, AI is increasingly being used to improve data quality, enhance system visibility, and strengthen cybersecurity—helping address some of the very challenges that hinder its adoption.
5. What should banks focus on to move from pilots to enterprise-scale AI?
Banks need to prioritize strong data governance, modern cloud infrastructure, and resilient, secure operations to successfully scale AI initiatives.
About the Author
As the Chief Technology Officer, Kishan Sundar helms the technology strategy for Maveric. His leadership in creating engagement and impact through customized technology solutions and emerging technologies will play a crucial role in accelerating Maveric’s revenue growth and fuelling its aspiration of becoming one of the top three Bank Tech companies.
Originally Published in CRN India
