India’s financial services industry has spent the past three to four years investing in AI. By now, banks and NBFCs have had enough experience to demonstrate that the technology can improve individual tasks. The harder question is whether those gains can translate into fundamental changes in how financial institutions operate.

Radhika Saigal, Financial Services Consulting Leader at EY India, describes the industry as being at an “interesting inflection point”. The returns from AI are real, she says, but they remain concentrated largely in productivity and efficiency rather than in fundamental business-model transformation.

That distinction is becoming increasingly important. Making an existing task faster does not necessarily change the economics of the process surrounding it.

“Financial institutions have become very good at proving that AI can work,” Saigal says. “The harder question is can you redesign an entire process around AI and actually change the economics of that process?”

From proving AI works to proving its value

Lending offers a useful example. An AI system that summarises a credit document might save an employee 20 minutes. That is a measurable efficiency gain, but it does not fundamentally change how the lending operation functions.

The larger opport Indian Financial Services Has Moved Beyond the AI Pilot. The Harder Test Is Value

unity would be to apply AI across document ingestion, analysis, underwriting, decision support and ongoing monitoring. In that model, AI is not simply inserted into an existing workflow. The workflow itself is redesigned around the technology.

Saigal describes this as the “pilot-to-production-to-value” gap.

Financial institutions have become increasingly proficient at demonstrating that an AI use case works. The harder task is taking that use case into production and showing that it has materially improved cost, speed, capacity, risk or another meaningful business outcome.

“The next phase of ROI will not come from having 500 AI pilots,” she says. “It will come from taking 5–10 high-value domains and rewiring them end-to-end around AI.”

The numbers are not intended as a measure of current industry adoption, but as Saigal’s prescription for where institutions should focus. Her broader argument is that AI investment needs to move away from accumulating use cases and towards redesigning selected business processes.

The central question is therefore changing from “What AI use cases can we deploy?” to “What business outcomes can we fundamentally change with AI?”

The real challenge starts inside the live banking environment

Indian banks and NBFCs offer a particularly demanding environment for AI deployment. They operate at enormous transaction volumes, serve increasingly digital customers and have sophisticated technology organisations. But those advantages also bring complexity.

AI has to work alongside legacy technology, fragmented data environments, regulatory obligations, cybersecurity requirements and model-risk controls. Building the model is therefore only one part of the deployment challenge.

“The issue is not whether an Indian bank can build an AI model,” Saigal says. “The issue is whether that model can safely participate in a live banking workflow.”

That requires the model to work with the right data and permissions while meeting requirements around explainability, auditability and human oversight. It also has to perform reliably when exposed to the scale and operational complexity of millions of transactions and customer interactions.

The organisational structure of traditional technology delivery adds another layer of difficulty. Technology teams have typically built platforms while business teams have defined processes. AI increasingly cuts across that separation because the technology and the underlying business problem often need to evolve together.

The challenge, therefore, is not simply acquiring or building better models. It is connecting those models to the systems, controls, workflows and decisions through which financial institutions actually operate.

Why the engineer is moving closer to the business

This is where the forward-deployed engineer, or FDE, is becoming increasingly relevant.

The role puts an engineer much closer to the business problem instead of several organisational layers away from it. Rather than moving a requirement through the traditional chain of business, product, architecture, engineering and operations, the FDE works directly with the business, understands the workflow, experiments with AI, builds the solution, takes it into production and iterates based on what happens in the real world.

Its importance lies in addressing the last mile of enterprise AI.

“The model itself may be relatively easy to access,” Saigal says. “The difficult part is connecting it to enterprise data, legacy systems, APIs, controls, workflows and human decision-making.”

That makes the FDE less a variation on a conventional engineering role and more a response to the organisational distance that can separate technology development from business transformation.

For financial services, where AI must operate within tightly controlled and regulated environments, that distance can be particularly consequential.

GCCs face an ownership test

India’s financial services GCCs could have an advantage in this transition. Many already combine three capabilities central to the FDE model: domain expertise, engineering talent and proximity to global products and platforms.

But having those capabilities is different from having ownership.

The traditional GCC model has largely been built around scale, efficiency and execution. Saigal sees an opportunity for Indian GCCs to evolve into AI product and engineering engines for their global organisations.

That would mean moving from “We execute what headquarters gives us” to “We own the problem, build the solution and scale it globally,” she says.

The distinction is important. If an India-based team builds an AI pilot that is subsequently handed to headquarters for deployment, the underlying operating model has changed little. A more meaningful shift would occur when India-based teams own the AI product lifecycle — from defining the problem to building, deploying and scaling the solution.

Financial services adds another dimension. An engineer who understands AI alongside banking risk, compliance, data privacy, controls and legacy platforms operates at the intersection of technology and regulated business processes.

But that opportunity depends on genuine ownership being transferred to GCC teams, rather than simply adding more AI responsibilities to existing roles.

For CIOs, the AI decision is really a process decision

For a mid-sized private bank, Saigal’s approach is to begin with business outcomes rather than the number of AI experiments.

She suggests focusing on high-value domains such as lending, customer service, operations, fraud, compliance and software engineering, and defining specific objectives around cost-to-serve, turnaround time, revenue, risk reduction, productivity or customer experience.

The teams pursuing those outcomes also need to be structured differently. Engineers, product specialists, domain experts, data specialists and risk and control professionals need to work together around the same business problem.

The bigger shift, however, is redesigning the workflow rather than simply adding AI to it.

A co-pilot layered onto an inefficient process may make that process somewhat faster. It does not necessarily remove the inefficiencies built into the process itself.

“A co-pilot that sits on top of an inefficient process may make that process slightly faster,” Saigal says. “An AI-native process can fundamentally change how work gets done.”

Governance needs to be designed in the same way. In financial services, responsible AI cannot be reduced to a review at the end of development. Security, privacy, model risk, explainability, audit trails and human oversight need to be built into the architecture from the beginning.

The race is now from prototype to measurable value

The ingredients required for this transition are not necessarily scarce in India. Saigal points to the country’s engineering talent, data resources and financial-services scale.

What needs to change is how effectively those capabilities are brought together around business problems.

“If I had to identify just one thing, it would be the speed at which we move from experimentation to ownership and deployment,” she says.

That makes the next phase of AI adoption as much a test of organisational execution as of technological capability.

The institutions that emerge as significant AI adopters may not simply be those with the biggest AI budgets. They may be the ones that can move an idea through the entire chain — from business problem to prototype, production and measurable value — with the least friction.

“The next generation of AI leaders will not necessarily be the institutions with the largest AI budgets,” Saigal says. “They will be the institutions that can take an idea from a business problem to prototype to production to measurable value faster than everyone else.”

For Indian financial services, the competitive question is therefore becoming narrower and more demanding.

It is no longer simply whether a bank or NBFC can experiment with AI, or even whether it can put an AI application into production.

The harder test is whether it can redesign critical parts of the business around the technology — and make those changes work reliably at scale.

“The competitive advantage will not be who experiments with AI first,” Saigal concludes. “It will be who industrialises it first.”