IBM Leader Diagnoses Core Problem Holding Back AI in Banking

For all the talk of AI-powered transformation, the digital banking experience is often stagnant. Financial institutions have largely failed to deliver meaningful value, usually stemming from insufficient architecture. Banks are layering flashy applications on top of broken, fragmented systems, creating an illusion of progress that masks a dysfunctional reality. While the apps look new, the core systems needed for actual intelligence remain stuck in the past.

It's a problem that Prashant Parida*, the AI Strategy, Transformation & Portfolio Management Leader at IBM, knows all too well. In his role, he helps reshape a $20B global services business and directs multi-million-dollar AI investments. As financial institutions grow, he explains, their focus often narrows from customer innovation to managing internal complexity. It's a problem his own company faced after hundreds of acquisitions. The common result is a collection of disparate legacy systems that don't talk to each other, stifling real progress. "Most institutions are building the application layer without solving the grassroots problems, and that's where it's breaking. A consumer can ask any question, but the system is incapable of answering it. This is the distortion customers feel. The promise they are shown never actually comes through," he says.

• Net worth nonsense: The gap between promise and reality isn't just a vague feeling for consumers, but can be a diagnostic tool for institutions. Parida points to a frustratingly simple interaction with his own banking app to pinpoint the source of the failure. "My banking app promises insights, but it fails at a basic level. When I ask it to show my spending, the view breaks because it can't consolidate multiple accounts. When I ask a more fundamental question about my net worth, it has no answer. The bank doesn't understand me as a whole person because its business model remains focused on generating transaction fees instead of delivering what I, the customer, am actually looking for."

• Get your data right: In Parida's view, a bank's inability to answer such a fundamental question is a clear symptom of a broken foundation. The only way for AI to deliver on its promise is to honor a disciplined, "ground-up" sequence. It's a lesson he compares to the dot-com bubble, where initial excitement gave way to the painful realization that success requires solid fundamentals. "For any business to be functional, you need layers of capability. First is infrastructure, then the data foundation layer. But if you have acquisitions, you will have data in so many disparate systems that don't talk to each other. You must get that data layer right before you can do machine learning. Only then is AI going to work."

A core mistake, Parida explains, is that many institutions are building from the top down. That top-down approach can lead to what Parida calls "AI theater," a phenomenon where the flashy interface masks stagnant intelligence.

For him, following a ground-up blueprint is essential for achieving "contextual banking," where a financial institution can see a customer's entire financial life in one adaptive view. He points to a few bright spots in a sector grappling with these exact challenges. JPMorgan Chase's award-winning LLM Suite is proof that a massive incumbent can adopt the right philosophy. "JPMorgan Chase is an example of an incumbent trying to do something meaningful. The LLM they're building is being done 'ground up.' They are investing in that core LLM, and I think that's where it has to be," Parida agrees.

• David's blueprint: In parallel, startups are already demonstrating what's possible. While banks often possess much of the necessary customer data, their legacy architectures prevent them from unifying it into such an adaptive view. Outside of JPMorgan Chase, he predicts that innovation in this space will likely come from startups acquired by larger institutions. "An app like Monarch shows the way. It's not a bank, but it aggregates all of a customer's financial information, including cards, savings, and investments, into a single view. This is what banks should be doing. Evolving beyond a simple transactional capability to build from the ground up with a customer-centric view," Parida outlines.

• More than a metric: Ultimately, he connects this technological argument to its real impact on customers: trust. Parida's hierarchy of trust is fundamentally enabled by a well-built technology stack. Within his framework, safety is fundamentally compromised without a secure data layer, and true partnership is difficult to achieve without contextual AI. He sees the end goal as democratizing services previously reserved for the wealthy to enable a new level of partnership. "The highest level of trust is achieved when a bank can move beyond a simple metric. Helping someone with a high credit score is easy. True partnership is demonstrated when the bank proactively helps a customer fund their child's education, understanding that their credit score may be temporarily low. If a bank’s policy is to just deny anyone below an 800 score, how can trust be built? The customer is just another metric," he cautions.

Looking ahead, the pressure on banks to solve these foundational issues will only intensify. Parida forecasts that the long-term competitive landscape could be reshaped entirely by new technologies challenging the role of traditional banking intermediaries. "If quantum computing reduces the cost of compute, crypto is going to come back and reduce all the intermediaries in the banking system. That will not be good for big banks, because they want to control the system and crypto will take that control away."

—

* Prashant Parida's opinions expressed in this article are solely his own, and do not necessarily reflect the views of IBM.