Banks Say 'Explainable Velocity' Backed by Real-Time Data is Key to Balancing Speed and Trust in the Age of AI

Most financial institutions are chasing the promise of near-instant AI-powered decisions. Today, companies like Revolut already process mortgages in a single business day. Yet as the drive for speed intensifies, so does the risk of opacity. Systems built for velocity often outpace understanding, leaving leaders unable to explain how key decisions are made. The new mantra across finance is clear: speed only matters when it can be trusted.

One expert pioneering this approach is Rohit Kela, an AI-driven fintech specialist and banking consultant. As an Industry Consulting Manager specializing in loan management systems at IT consultancy NTT DATA's Banking and Financial Services division, Kela’s authority comes from pairing deep expertise in loan origination with hands-on experience building real-world AI systems. His perspective bridges both sides of the equation, combining the technical precision behind automation with the business realities that demand explainable results.

"Speed is what draws customers in, but explainable velocity is what keeps them. AI that moves fast but can’t explain its reasoning is far too risky. Every decision needs to reveal its data sources, the logic behind it, and the exact path it followed. Otherwise, that kind of speed won’t last. Every calculation, every choice, should be logged and validated. Only then can you build speed that people can trust," says Kela. For him, explainable velocity isn’t just a technical framework. It’s a philosophy of responsible automation that balances the pace of innovation with the need for visibility, governance, and human oversight.

• The pillars of success: As major institutions like Citi embrace agentic AI, he says safe implementation depends on three architectural pillars: real-time architecture, robust agent orchestration, and a zero-trust foundation that demands an auditable trail. When orchestration fails, one misstep can trigger a cascade of errors. "If the first agent doesn’t hand off the right output, or the next one starts too early, the entire system can break," he explains.

For leaders looking to navigate the new reality, Kela lays out a pragmatic, three-phase path to production: start with data plumbing, establish governance, and then run small pilot agents on high-ROI cases in a sandbox environment. The phased plan also helps address the common anxiety around job replacement, establishing a new model for human-AI collaboration.

• Humans on the edge: Instead of replacing jobs, Kela believes agentic AI will redefine them. The new model delegates high-volume, low-risk tasks to machines, elevating human experts to strategic decision-makers for the most complex and critical scenarios. "The human-in-the-loop is reserved for the edge cases. A simple, low-risk loan with a good CIBIL score can be a straight-through process for an agent. But when a case is critical, the agent’s role changes. It prepares a summary for a human to analyze, saving them from having to check multiple systems."

• You can lead a horse to water: Building these systems is one thing, but getting people to use them is another. Kela points to a hard truth about AI adoption. "I attended a training session that began with twenty thousand people on day one," he recalls. "But by the end, only three thousand completed it." For him, the takeaway was clear. "You have to be self-motivated. You need to understand this is the future and adapt sooner rather than later." While he credits organizations for offering training, he believes they could go further by creating forums where employees can apply what they’ve learned and contribute to real AI projects.

While the ROI of speed is easy to measure, a much harder challenge lies in measuring the human side of the equation. Kela admits that quantifying adoption, confidence, and trust remains a key, unsolved problem for the industry.

• The ROI blind spot: "It’s easy to show a client how much time or money they could save with AI, but the real question is, will anyone actually use it?" Kela says. "That’s the missing metric. We can measure efficiency, cost reduction, even error rates, but adoption is harder to quantify. You can’t automate human trust. Until we figure out how to measure confidence and willingness to engage with these tools, the story is only half told."

Kela’s message to leaders is simple: build for real time, log every decision, and scale only what you can explain. Agentic AI is the next foundation of financial infrastructure, not a passing trend. But as innovation accelerates, Kela anticipates an industry split that could leave smaller players behind. The cost of advanced systems may widen the gap between global institutions and community banks that can’t afford to keep pace, he concludes. "Big banks can invest in these heavy, expensive tools. As they roll them out, they’ll take more of the market. The real question is what happens to the local banks. They risk being diminished, or disappearing from the market entirely."