AI in banking should be the co-pilot, not the sole pilot: Deepak Sharma

Even though the scope of Artificial Intelligence (AI) in financial services is massive, it is necessary that the final judgment still involves human wisdom, says Fintech veteran Deepak Sharma.

In a chat with The Federal, Sharma explains how AI is reshaping India’s financial services, ranging from multilingual inclusion to SME credit access, while shedding light on the path innovation is headed and what holds the sector back.

Excerpts:

Can you tell us a little bit about your background and current work?

I've been in financial services for over two decades, with a long stint at Kotak where I last served as president and chief digital officer. Post-Kotak, I decided to explore broader opportunities—still within financial services, but focusing more on technology and innovation. Today, I am on the boards of Experian India and Suryodaya Small Finance Bank as an independent director. I am also on the advisory board of Instabase, a global AI and Gen AI firm headquartered in the Bay Area. They work with leading institutions, and I help shape their Asia strategy.

Additionally, I mentor startups across diverse spaces—SME credit, multilingual tech, NRI investment platforms—bridging technology, security, SaaS, and AI.

I spend time helping growth-stage companies chart paths to profitability and expansion.

Could you talk more about the startups you’re involved with and what excites you about them?

I’m not a typical investor, but I do get involved when I see real-world problems being addressed meaningfully.

Take Devnagari, for instance—they’re building multilingual financial service platforms. India is largely multilingual, yet most digital banking experiences are still in English. That worked for the first 100–200 million users, but not the remaining 1.2 billion.

Then there’s Pais, a company working on unlocking credit for SMEs based on GST data. They help improve working capital and financial health — an area I find very relevant.

Another example is a platform focused on NRIs who want to invest in India. At Kotak, I helped build the NRI banking vertical, and we saw the limits of what banks could do. Now this new platform is trying to solve not just onboarding and remittances, but post-fund deployment—alternate assets, equities, private equity.

Even with Instabase—while not an investment—I contribute to solving how unstructured data can be converted into insights.

Every engagement is about solving tangible problems, either through capital or counsel.

You’ve mentioned AI co-pilots in banking. What does a co-pilot model look like in practice?

Think of an aircraft’s autopilot. It doesn’t replace the pilot but assists in navigation, reducing error and improving safety. Similarly, an AI co-pilot in financial services enhances human decision-making and eliminates mundane, repetitive tasks.

Take customer service: we’ve all had poor contact center experiences—long IVRs, uninformed agents, long hold times. Now imagine a bot that understands your context from past interactions—voice, text, transactions—and responds intelligently. If escalation is needed, the agent sees a full contextual dashboard, possible resolutions, and upsell suggestions.

For instance, if you call about waiving a credit card late fee, the agent instantly knows whether it can be waived and what products to offer next. That’s co-pilot mode.

Even in operations—say, preparing a credit proposal—AI can ingest data from 30+ sources (bank statements, balance sheets, tax filings), pre-fill templates, and suggest outcomes. It becomes an intelligent assistant to humans across roles: developers, credit officers, and analysts.

So, in a loan context, how much time does AI really save?

A manual review could take hours or days. An AI engine, with structured parameters and access to the right data, can complete the same task in under a minute.

Whether the data is text, image, handwritten, or from public sources, once the inputs and outputs are defined and an LLM is trained on them, the system generates a near-instant output. Co-pilots today might not deliver 100 per cent accuracy, but 90–95 per cent with high speed is a game-changer. And as models evolve, that number will only improve.

With fintechs moving beyond payments to credit, where will AI have the biggest underwriting impact?

Payments are the “how,” credit is the “what.” But increasingly, these lines are blurring.

Regulators now demand that banks understand the end use of credit. Did the borrower actually use the funds for the stated purpose? Many delinquencies happen because funds are diverted. The same is true for unsecured consumer loans—if people are borrowing just to repay other loans, that’s a red flag.

AI plays a critical role in analysing behaviour, verifying end-use, and helping embed credit within consumption. For example, you might buy a fridge and be offered an EMI option right there. Or take a cab ride and get fortnightly settlement credit.

Soon, credit and payments will be a single seamless interface.

But with such ease comes risk. How does AI help prevent fraud?

Great question. Convenience without control is a recipe for disaster.

In our real-time digital economy, money can vanish in seconds due to phishing, mule accounts, or identity theft. While the RBI’s initiative, like Mule Hunter, is a good start, we need more sophisticated AI models.

AI can help identify fraud patterns, trace fund flows, and flag suspicious activity in real-time. But we need to go further, like using cryptographic keys for user authentication. Think of a DNA-level signature that can’t be faked.

We also need AI audits. Just like IT and CA audits, algorithmic decisions must be documented, traceable, and governed. If a model makes a wrong call, the accountability still lies with humans. Governance frameworks must evolve as fast as the tech.

What’s your take on AI for investment advice and portfolio management?

AI is already helping financial markets by crunching data, identifying trends, and simulating scenarios. Pre-trained models allow us to test multiple outcomes.

Can AI help wealth managers and advisors give better, unbiased recommendations?

Absolutely. But it should still be a co-pilot, not the sole pilot. That final judgment should still involve human wisdom.

The content above has been generated using a fine-tuned AI model. To ensure accuracy, quality, and editorial integrity, we employ a Human-In-The-Loop (HITL) process. While AI assists in creating the initial draft, our experienced editorial team carefully reviews, edits, and refines the content before publication. At The Federal, we combine the efficiency of AI with the expertise of human editors to deliver reliable and insightful journalism.