7 in 10 Indian banks have AI in production, but security and control hold back scale: Zeta

Indian banks have moved beyond experimenting with artificial intelligence (AI), with most institutions now deploying the technology in production, although scaling it across functions remains a challenge, according to Zeta’s 2026 CXO Survey on the state of AI in Indian banking. 

The survey, based on responses from 40 CXOs across 18 leading banks and non-banking financial companies (NBFCs), found that 70% of chief data officer (CDO) respondents place their institutions at the selective or scaled deployment stage, including 30% that have reached scaled deployment. 

AI adoption is strongest in structured and reviewable areas such as customer service, fraud and risk analytics, document processing and software testing. However, its integration into end-to-end workflows and consequential decision-making remains at an earlier stage. 

AI adoption is growing, but remains concentrated  

The survey highlights a growing gap between proving that AI works in production and deploying it repeatedly across an institution. 

About 88% of chief operating officers (COOs) said AI is delivering meaningful impact in retail lending, followed by customer service at 75% and current account and savings account (CASA) and back-office operations at 63% each. 

Adoption is strongest in high-volume, structured workflows where AI-generated outputs can be reviewed within existing controls. However, redesigning entire workflows around AI—including determining what the technology should execute, recommend or escalate—remains at an earlier stage. Around four in 10 CDOs said they could not yet identify a high-return use case at their institution. 

Despite growing adoption, most banks surveyed allocate less than 10% of new-project technology spending to AI. Zeta said this reflects caution around scaling rather than a lack of confidence in the technology. Lack of clarity on return on investment was the lowest-rated barrier while skills and security concerns ranked higher. 

Data availability is not the problem; usability is 

While 80% of CIOs and CTOs described their data environments as mostly ready for AI at scale, none considered them fully ready. 

The key constraints relate to making data usable by AI. About 61% cited insufficient labelled or training data, 53% pointed to privacy and consent issues, and 46% cited siloed data. 

Banks are increasingly using AI to address these gaps, with 67% already using or piloting AI to enhance or enrich their data. 

Technology infrastructure is also becoming more AI-ready. Real-time data platforms and API-first architectures have reached 79% adoption among respondents, while core modernisation and cloud adoption stood at 64%. Advanced analytics and machine learning operations (MLOps), which support the deployment and monitoring of AI workloads, stood at 43%. 

AI use in software engineering remains review-led 

AI has gained a strong foothold in software engineering, but adoption declines as the technology moves from generating outputs to executing tasks. 

About 80% of CIOs and CTOs reported using AI in testing and quality assurance, while 60% use it for code generation. Adoption falls to 40% for code review and 30% each for specifications and documentation, deployment and CI/CD, and incident detection. 

Security and data privacy emerged as the leading barrier, scoring 3.89 out of five, compared with 2.0 for lack of clarity on ROI. 

Banks prepare AI for higher-stakes decisions 

AI-led credit-risk models, predictive early-warning systems and real-time fraud decisioning are among the top priorities for at least 60% of chief risk officers (CROs) over the next 18-24 months. 

However, governance is still evolving. Around 60% of respondents said responsible AI frameworks are under development, while none reported organisation-wide implementation. Only 20% described their model-risk management frameworks as very mature. As AI moves closer to consequential decisions, banks are increasingly focusing on controls around AI identity, permissions, policy enforcement and auditability. 

The survey also found that AI is more likely to reshape jobs than reduce headcount in the near term. Half of operations leaders expect productivity gains to free up capacity for higher-value work, while none expect workforce reductions of more than 20%. 

Banks are building AI capabilities through specialist hiring and external partners faster than through internal development, which received the lowest capability score in the survey. “Indian banks have shown that AI creates value in production. The next challenge is making that success repeatable, and the survey is clear about what stands in the way: not conviction, but control,” said Sivaram Kowta, President, Zeta India. 

The survey identifies four stages of AI adoption: proving relevance, proving value and control, making deployments repeatable, and embedding the capability across the organisation.