In the second part of an interview for The Paypers, Kevin Levitt, Director of Global Industry Business Development for Financial Services at NVIDIA, further shares his thoughts on AI investments and the role of open-source tech in deploying AI-enabled apps.
Agentic AI is the hottest trend and topic right now. Everyone's talking about it. How can fintechs best differentiate marketing hype from applications that truly bring value?
Our survey results support the idea that agentic AI is one of the hottest topics in financial services. This was the first year we asked specifically about it in our survey, and 42% of respondents said agentic AI is a focus area for their firm. It already ranks in the top three of AI workloads in just one year because of the ROI it can generate. For example, if we look at the success the Royal Bank of Canada has had applying agentic AI utilising NVIDIA's full-stack accelerated computing platform – both the hardware and the software – to its capital markets business, what did they do? They deployed an agentic workflow with over a dozen agents to capture new data inputs, analyse that data, develop new research, and draft research notes for equity research analysts and others in the capital markets division to review and refine.
Using this new agentic AI-enabled workflow, what once took around 40 hours to generate (a new research report) has been reduced to about 15 minutes. That’s a roughly 60x productivity improvement for thousands of analysts and capital markets colleagues, who can now leverage agentic AI to deliver meaningful outcomes for the business.
When deciding how companies can be most effective in where and when to use agentic AI and in which applications, it’s important to aim it at the most meaningful projects and business opportunities. Don't point it at what's easiest, because that doesn’t deliver value or create competitive differentiation in the marketplace. Without impact, there is no ROI. We constantly advise customers to focus agentic AI on their most meaningful and impactful business challenge. For 2026, we’re seeing agentic AI expand beyond employee assistance, similar to what the Royal Bank of Canada is doing, into direct customer engagement. The accuracy of these models, along with techniques for deploying safer and more reliable ones, will keep improving.
Do you think customers will actually be willing to leverage agentic AI?
We're already seeing agentic AI used for customer engagement, where people interact with AI for help with their questions. PayPal is a great example of how AI enhances customer experience. bunq, another well-respected fintech, is also leveraging agentic AI to strengthen customer experience and engagement.
As consumers, our appetite for AI and our comfort with it will continue to grow, because we're encountering it not just in financial services but across almost every aspect of our daily lives. So, we continue to see our clients (banks, financial firms, and fintechs) investing in NVIDIA's full-stack accelerated computing platform for agentic AI use cases.
Beyond operational efficiency, which emerging AI use cases do you expect fintechs to invest in next, and how should they ensure a future-proofed AI strategy?
To build on what we were discussing earlier, around payments, foundation models, and fraud detection, there’s another area gaining momentum – transaction foundation models used for recommendation systems and ‘next best action’ suggestions. When you predict the next transaction, you’re not only helping fight fraud but learning about the consumer’s purchasing behaviours. That insight enables companies to make recommendations or issue incentives, such as coupons, to encourage the next purchase in a specific category or with a particular merchant. This will drive market share not just for that merchant but also for the brand or company-issued credit card. It’s about leveraging these recommendation systems to drive wallet share. This is exactly why we think of transaction foundation models not as single-purpose tools but as an intelligence layer at the heart of the bank — trained on proprietary transaction data, and then deployed across fraud, credit, personalisation, and more from a single shared backbone.
When it comes to future-proofing your AI strategy, there are two ways to think about it. First, consider scale. That means getting your data pipelines in order, because every AI workflow starts with data, moves through data preparation and model training, and finally deployment and inference. Preparing your data pipelines also means eliminating silos, which are common in banks. The more data you can feed into models and open source frameworks, the more accurate outcomes you will generate for both internal employees and external customers.
Second, transitioning from legacy CPU-based infrastructure to accelerated, computing-powered AI factories is crucial. AI factories are full-stack systems, encompassing everything: from GPUs and accelerated computing networking that connects servers at data centre scale to the appropriate software. Building the infrastructure is one thing, but maximising its utilisation is another. That is where NVIDIA’s platform software and NVIDIA AI Enterprise help: empowering data scientists to build and deploy AI-enabled applications efficiently.
That is really the key message: keep prioritising high-impact use cases and equip your most valuable people – machine learning engineers, data scientists, and developers – with the right tools to bring AI-enabled applications to market.
About the author
Kevin Levitt is the Senior Director of Global Industry Business Development for Financial Services at NVIDIA, part of NVIDIA's global financial services team. The team is responsible for setting NVIDIA's strategy in the ecosystem, focusing primarily on trading, banking, and payments, understanding the priority workloads and use cases for NVIDIA's accelerated computing platform, and helping both end customers and our partner ecosystem leverage it to build and deploy AI-enabled applications at scale.
