From Pilots to Performance: The Real Work of AI in Banking

From Pilots to Performance: The Real Work of AI in Banking

Most banks will tell you they are “using AI.” Fewer can tell you what that means in practice: which processes have genuinely changed, which decisions are materially better, and what the organisation had to do differently to make any of it work. The gap between the announcement and the reality is, in most cases, considerable.

To understand what it truly takes to introduce artificial intelligence into the bloodstream of modern financial services, two distinct perspectives are invaluable: that of the banking leader navigating institutional scale, and that of the technology leader architecting the systems.

Krish Swamy, Chief Data & Analytics Officer at Citizens Bank, and Pradeep Gulipalli, Co-founder & CEO India of Tiger Analytics, bring complementary perspectives to this conversation: one from inside a large banking institution, the other from the frontlines of enterprise AI implementation. They sat down to talk about the reality of rewiring banking from the inside out—a journey where the lessons that emerged were as much about human readiness as about algorithmic prowess.

The Case for Reimagination

Banking has long been a vanguard of technological innovation, pioneering the foundational digital infrastructure—from global electronic payment networks to secure automated ledger systems—that modern commerce relies upon. The arrival of advanced AI is not a rescue mission for outdated systems, but rather the next logical frontier for an industry already accustomed to driving large-scale technological evolution. The potential is vast: optimizing risk assessment, personalizing client experiences at scale, and extracting insights from massive data ecosystems.

However, the deployment of these advanced systems faces a barrier unique to the financial sector. While consumer tech firms can rapidly iterate and deploy new models, a bank’s ability to utilize cutting-edge AI is strictly governed by unparalleled data privacy restrictions and rigid regulatory compliance mandates. Because banks handle the world's most sensitive financial and personal data, they cannot simply plug into unvetted, black-box models.

The Absolute Standard: In banking, technological capability means nothing without absolute data sovereignty and model explainability. A new AI model is only as good as its ability to pass a rigorous compliance audit.

Consequently, the institutional challenge is never about a lack of innovation or willingness to adapt. The true strategic hurdle is determining how to safely harness the power of next-generation AI within an uncompromisingly strict regulatory framework—transforming operational outcomes while ensuring that data privacy is never compromised.

As Krish Swamy puts it:

Systemic over siloed, outcome-driven over technology-driven serves as a vital guiding philosophy. It means resisting the temptation to celebrate isolated wins and instead asking the harder question: does this, in totality, make the bank meaningfully better?

Deciding Where AI Belongs

In a large bank with thousands of processes and a labyrinth of regulations, deciding where to deploy AI can be paralysing. To cut through that complexity, Krish relies on a simple three-question framework:

• Does it make customer interactions better?
• Does it simplify routine tasks?
• Does it achieve higher throughput without compromising on safety and privacy?

If the answer to any of these is yes, AI belongs in the conversation.

These questions are not a rigid methodology as much as a practical lens through which institutions can evaluate opportunities. Together, they reflect the three outcomes that matter most in banking: better experiences, greater efficiency, and trusted scale.

Of the three, Krish reserves particular emphasis for customer experience. In banking, the customer relationship determines retention, share of wallet, and ultimately long-term profitability. For too long, that relationship has been mediated by friction: long hold times, repeated form-filling, and generic offers.

AI changes that calculus, enabling banks to become responsive rather than reactive and present at the moments that matter.

Integrating AI Across Operations and Decision-Making

Weaving AI into the fabric of daily operations and decision-making requires both deep industry strategy and intense engineering muscle. Yet the banking sector’s enthusiasm for AI in general stands in sharp contrast to its caution around Generative AI and autonomous agents.

This is not timidity; it is institutional wisdom. A misclassified email is a nuisance. A misclassified loan application is a legal and ethical event. The stakes in banking demand that the guardrails be built before the race begins.

Even so, banking and financial services remain one of Tiger Analytics’s fastest-growing verticals. The appetite for transformation is evident. The challenge is ensuring that AI becomes embedded into how decisions are made, rather than remaining a collection of disconnected experiments.

The Bottlenecks That Need to Be Solved First

Beneath the optimism of any AI transformation lurk some stubborn bottlenecks that surface, almost without exception, across implementations.

The first is data quality. Banks are data-rich institutions, but that data is often messy, siloed, inconsistently labelled, and historically collected for recording transactions, not for revealing patterns.In institutions where digitization has occurred unevenly across various levels or geographies, the fragmentation runs even deeper. Add access restrictions imposed by cybersecurity policies and regulatory safeguards, and the picture becomes clear – the data exists but getting it into a form that a model can learn from is challenging.Feeding poor-quality data into a sophisticated model produces a confident wrong answer, which is arguably worse than no answer at all.

The second is the absence of AI-ready infrastructure. AI cannot be layered meaningfully on systems that were never built to support its scale, speed, or complexity. Legacy architectures, fragmented platforms, limited integration, weak governance layers, and inadequate compute environments can slow down even the most promising initiatives.

The third bottleneck is workforce readiness. AI tools, however well-designed, are only as effective as the people who work alongside them. Without proper upskilling, the technology sits unused, misused, or feared.

"It's a perspective grounded in experience. Tiger Analytics has worked with banks and financial institutions on some of the sector's most persistent pain points – from credit risk monitoring and loss forecasting to fraud detection and customer acquisition. For one of our clients, an enhanced fraud detection model helped a leading credit card issuer cut losses by 15%. In another, our value-driven scorecard approach helped reduce customer acquisition costs by 50%. AI can exactly address these consistent patterns of issues plaguing the industry, but only once the underlying data infrastructure and governance are sound. The capability exists. The question is whether the institution is ready to use it." Pradeep explains.

Measuring What Moves and Mapping the Market

How does an organization know if the technology is working?

This is the question that separates genuine transformation from the theatre of transformation.

Krish Swamy is clear: the answer lies in before-and-after studies grounded in metrics that actually reflect customer and business outcomes.

Take customer satisfaction scores. What were they before a particular AI-driven intervention? What are they after? The delta is real. It is auditable. And it is far more honest than the number of models deployed, or the volume of data processed.

When looking at the wider banking sector, a clear line is emerging between leaders and laggards.

A firm that has deployed a chatbot in customer service and considers itself an AI company is not in the same league as one that has embedded AI deeply into credit risk modelling, fraud detection, personalisation engines, operations, and compliance.

Pradeep Gulipalli is equally clear about the cost of inaction:

"Not every institution needs to move at the same speed and in a sector as heavily regulated and operationally complex as banking, that is not a weakness. A regional cooperative bank and a global transaction processor face fundamentally different constraints, risk appetites, and customer expectations. The gap between these two is not primarily a technology gap. It is a gap in data readiness, organisational will, and the ability to trust a model's output enough to act on it. The pace of AI adoption, within reason, is a strategic choice.

The distinction between piloting and embedding will become increasingly difficult to ignore, not because of competitive pressure alone, but because customers will begin to feel it. The institution that knows you, anticipates you, and serves you without friction is not offering a better product. It is operating on a fundamentally different informational foundation. That gap, once it opens, tends to widen."

Governance: Building the Circuit Breakers

Speed is AI’s greatest gift and its most significant risk.

A system that processes millions of decisions in the time a human analyst reads one file can also propagate errors at a velocity that humans cannot track.

Krish Swamy is precise about what governance requires in this context.

"Agents can propagate errors at speed. And so, when that happens, we need the right circuit breakers."

Circuit breakers combine technical controls, process design, and organisational culture—determining exactly when a human must override the model.

Pradeep aligns closely with this necessity, adding that establishing guardrails and context-setting is critical for any AI system.

In banking, governance cannot be an afterthought layered onto AI after deployment. The consequences of getting this wrong are not abstract: a miscalibrated credit model can systematically disadvantage entire borrower segments; a fraud detection algorithm with poor oversight can freeze legitimate accounts at scale; and a running pricing model left without review can drift into regulatory risk before anyone inside the institution notices. This is why governance must be designed directly into the workflow from the start: escalation thresholds that define when a human must review before action is taken; audit trails that allow any decision to be reconstructed and explained after the fact; human override points that preserve institutional accountability; and clear boundaries on model autonomy.

That last point matters more than it might appear. An AI system with no defined ceiling on its autonomy creates a governance grey zone that tends to widen over time. The institutions that manage this well share a common orientation: they treat AI governance not as a compliance obligation to be satisfied, but as a design discipline to be practised. The difference shows in audit outcomes, in regulatory relationships, and in the long-term resilience of the systems themselves.

AI and Me: Personal Philosophies

It is one thing to deploy AI at scale; it is another to reflect on how one’s own thinking and decision-making is being shaped by it.

Both Krish and Pradeep are candid about their personal relationships with the technology they champion.

Pradeep views AI as an intellectual equalizer. While he willingly challenges and overrides AI in areas where he is already an expert, he finds immense value in using it to navigate unfamiliar domains and engage with new disciplines more effectively.

Krish speaks of balancing an “intentional use of AI” alongside what he calls “intentional non-use of AI.” The former is about leveraging technology where it genuinely adds value. The latter is a deliberate choice to preserve certain kinds of thinking, judgment, and presence that he does not want to outsource—like playing the violin.

It is, in miniature, the same framework he applies to the institution: AI where it earns its place, and restraint where it does not.

Looking Ahead

There is a version of the future that the financial sector has been promising for years: one where the customer is known before she speaks, where credit decisions are objective, where fraud is caught in a blink, and where the endless friction of paperwork disappears.

The blueprints for getting there are increasingly clear. Yet success will depend not on the sophistication of individual models, but on whether institutions can align technology, infrastructure, governance, and people around shared outcomes.

The F1 car is ready. The remaining question for leadership is whether the infrastructure, the fuel, and the drivers are truly ready to keep up.

This article is written by Bhavana Pandey. Pandey is the founder and CEO of Techquity India. Her work sits at the intersection of technology, business, and finance.

Note to the Reader: This article has been produced on behalf of the brand by HT Brand Studio and does not have journalistic/editorial involvement of Mint.