Ask most people where artificial intelligence (AI) is showing up in banking, and they will point to chatbots and personalised financial recommendations. While these are easy to demonstrate and increasingly familiar, they are not where AI will have its biggest impact.
AI in banking should be thought of as operating at three layers. The first talks to the customer. The second makes decisions about transactions and customers through fraud detection and credit. The third helps run the institution itself, from liquidity and compliance to everyday operations. While public attention falls away as one moves inward, the importance of the layer does not.
The first layer is where AI has moved fastest. In major digital banks, AI already handles routine questions and helps frontline staff find information. But as simpler interactions are automated, what remains is disproportionately complex. Disputes, exceptions, and cases that require reassurance are harder to automate, and incremental gains are more difficult to achieve.
Fraud and credit sit one level deeper. Traditional fraud systems relied heavily on fixed rules. Newer models can examine behaviour across accounts, map relationships between entities and process thousands of signals at once. JPMorgan says AI has enabled its transaction-screening operation to review more than twice the volume while cutting manual checks in half. However, this is an arms race, and fraudsters have access to the same tools, using them to automate phishing and build more convincing synthetic identities and deepfakes. As fraud detection improves, so do the attacks.
Credit decisioning is another major use case at this layer. Lenders can enhance a static bureau score with a current view of income, cash flows, spending and other signals, which can help them assess borrowers with thin credit histories. However, richer data does not automatically produce fairer decisions. Opaque models can make decisions harder to understand and exclude some borrowers more aggressively.
The third layer is the least visible, but it may also create the most durable institutional advantage.
Let’s take treasury as an example. Banks continuously decide how much liquidity to hold, how deposits may behave and where to deploy capital. AI does not need to run the balance sheet to matter. It can support deposit modelling, liquidity forecasting and cash-flow analysis, and a marginal improvement in a forecasting process can change how much money a bank needs to keep idle and how much it can deploy elsewhere.
The same logic applies to compliance and operations, where AI can review know-your-customer documents, prioritise alerts and assemble complex cases for human reviewers. These are not eye-catching use cases, but their value can compound across thousands of decisions.
In my experience working on AI rollouts, the harder part is rarely the technology or the model. These functions often run on judgement or institutional knowledge that is passed on but never written down. An experienced analyst knows which exception is worth escalating and which is not, but may struggle to explain why. Automating the documented process can start delivering results, but capturing that undocumented institutional knowledge is what makes these systems work at scale.
A survey by BCG found that more than 80% of the largest global banks in its sample were using AI in some form within treasury, compared with around half across the sample as a whole. While the impact so far is modest, institutions with more data, specialist talent and the capacity to embed AI deeper into their own workflows may pull further ahead as these applications mature.
This matters because India has built some of the world’s most powerful shared financial infrastructure through Aadhaar, UPI and Account Aggregator, creating a strong foundation for the second layer of AI. However, that advantage will not automatically carry over to the third layer. Treasury, risk, compliance and operations AI must largely be built within individual institutions, using their own data, workflows and accumulated judgement. Shared computing capacity, testing environments and sector-specific models can lower the entry cost, particularly for smaller banks and NBFCs, but they cannot replace that institution-specific work.
India’s next financial advantage will depend on whether its institutions can move beyond generic tools and embed intelligence into their core decisions. The next phase of AI in banking will not be won by the best chatbot, but by the institutions that detect risk earlier, allocate capital better and operate more intelligently behind the scenes.
Manish Agrawal is a strategy lead at Chime, the largest neobank in the US, where he works on scaling capabilities and operational infrastructure for digital banking.
Catch all the Industry News, Banking News and Updates on Live Mint. Download The Mint News App to get Daily Market Updates.
Log in to our website to save your bookmarks. It'll just take a moment.