This article is by Sandeep Khuperkar Founder & CEO, Data Science Wizards (DSW)
India’s financial services sector is at an inflection point that has little to do with technology adoption and everything to do with trust.
For the better part of a decade, banks, insurers, and non-banking financial companies have used artificial intelligence to assist human decision-makers flagging a suspicious transaction, recommending a product, speeding up a routine approval. The human stayed in charge. The AI stayed in the background.
That arrangement is changing quickly. A new generation of AI systems commonly described as “agentic AI” does not merely recommend. It acts. It can read a document, verify it against policy, check it for fraud indicators, decide on a course of action, and execute that decision, often without a human touching the case until something goes wrong. For a retailer recommending a product, the consequence of error is mild inconvenience. For a financial institution approving credit, settling a claim, or flagging a fraud case, the consequence of error is someone’s money, someone’s eligibility, or someone’s legal standing.
This is the distinction Indian financial services leadership must internalise quickly: agentic AI is not a faster version of the automation we already trust. It is a fundamentally different category of system, and it demands a fundamentally different category of governance.
Why “High-Integrity” Is Not a Buzzword Here
Every industry likes to claim its operations are mission-critical. Financial services has a stronger claim than most, for a simple reason: money and trust are largely the same thing in this business. A bank’s core asset is not its balance sheet it is the public’s confidence that the balance sheet is accurate, that decisions are fair, and that someone is accountable when something goes wrong.
Agentic AI, deployed carelessly, threatens exactly that confidence. An autonomous system that approves a loan it shouldn’t, denies a legitimate insurance claim, or flags an honest customer as fraudulent does not just create an operational error. It creates a trust event the kind that regulators investigate, customers remember, and competitors exploit.
This is why the conversation among serious financial institutions is no longer “should we adopt agentic AI.” It is “how do we adopt it without compromising the integrity our entire business is built on.”
The Regulator Has Already Noticed
This is not a hypothetical concern for the future. The Reserve Bank of India’s recently released draft Guidance on Regulatory Principles for Model Risk Management makes the direction of travel explicit. The draft applies across the entire regulated financial ecosystem banks, NBFCs, co-operative institutions, asset reconstruction companies and asks regulated entities to maintain the ability to instantly override or deactivate any AI model, to build in documented human oversight, to disclose to customers when AI has driven a decision, and to assign clear, board-level accountability for AI governance.
Read carefully, this draft is not really about restricting AI. It is about insisting that AI be operated the way every other critical financial system already is: with a named owner, a defined kill switch, an audit trail, and a human who can answer for the outcome. Financial institutions that treat this as a compliance checkbox to satisfy after the fact will find themselves perpetually behind. Institutions that treat it as a design principle built in from day one will find that good governance and good customer experience are, in this case, the same thing.
Why Most AI Initiatives Still Stall
Despite years of investment, a striking number of AI initiatives in financial services never progress meaningfully beyond a pilot. This is rarely a failure of the underlying technology. It is a failure of operating model.
A proof of concept can succeed in isolation one model, one use case, one carefully curated dataset. Production is different. The moment an AI system has to coexist with dozens of other systems, respond to a regulatory change overnight, route an edge case to a human reviewer, and remain explainable to an auditor eighteen months later, the challenge stops being about model accuracy and becomes about institutional design.
Put simply: an enterprise can build excellent AI and still fail to operate it responsibly. The two capabilities are not the same, and conflating them is the single most common reason ambitious AI programmes in this sector quietly stall.
Governance as a Source of Competitive Advantage
There is a tendency, particularly in regulated industries, to treat governance as a constraint on innovation a necessary cost that slows things down. In the agentic AI era, this framing is not just outdated, it is backwards.
Consider what a customer actually wants when an algorithm influences a decision about their money. They want to know that a human can be reached if something feels wrong. They want a sense that the decision was reasoned, not arbitrary. They want confidence that the institution remains accountable, even when a machine acted on its behalf. Every one of those expectations is, in effect, a governance requirement—and an institution that builds for it deliberately is not slowing down. It is building the only kind of AI deployment that customers, regulators, and boards will actually trust at scale.
The institutions that will lead India’s next phase of financial digitisation will not be the ones that deployed the most AI agents the fastest. They will be the ones whose customers never had reason to ask “who is accountable for what just happened to my application.”
What This Means for Leadership, Not Just Technology Teams
The temptation inside large financial institutions is to treat agentic AI as a technology programme, owned by the CTO or the data science function, with periodic updates to the board. This is a mistake that will become more costly as these systems take on more autonomous responsibility.
Agentic AI governance is a leadership responsibility for three reasons that have nothing to do with engineering.
First, accountability cannot be outsourced. If an autonomous system makes a flawed credit decision, the institution not its technology vendor, and not the algorithm is answerable to the customer and the regulator. That responsibility sits with the board, not the IT department.
Second, the operating model has to change before the technology can be trusted. Approval workflows, escalation paths, and customer disclosure practices all need to be redesigned around the assumption that some decisions will now be made, in the first instance, by a machine. That is an organisational design question, not a software question.
Third, trust, once damaged, is disproportionately expensive to rebuild in financial services. A single high-profile failure of an autonomous system—a wrongly denied claim that makes headlines, a discriminatory lending pattern uncovered by a journalist will cost the institution far more in reputation than the AI programme saved in efficiency. Leadership that views governance as the slow lane will eventually discover it was the only lane that led somewhere durable.
The Path Forward
None of this is an argument against agentic AI in financial services. The opportunity is real and substantial: faster underwriting, more consistent claims processing, dramatically improved fraud detection, and customer service available at a scale human teams alone cannot match.
But the institutions that capture this opportunity responsibly will be the ones that redesign their operating model alongside their technology building human oversight, auditability, and clear accountability into how these systems function from the outset, rather than retrofitting governance after a failure forces the question.
India’s financial sector has an opportunity that few global markets have had: to build its agentic AI capability at the same time regulatory expectations are still being defined, rather than years after the fact. Institutions that use this moment to build trust into the architecture of their AI not as an afterthought, but as the foundation will not just stay compliant. They will set the standard the rest of the industry eventually has to follow.
The question every financial services leader should be asking their teams this year is not how quickly they can deploy agentic AI. It is whether, a year from now, they will still be able to explain clearly, confidently, and to anyone who asks exactly how every one of those AI-driven decisions was made.
Be the first to comment