From human-in-the-loop to human-on-the-loop: How AI agents are redefining finance

  0 Be the first to comment

From human-in-the-loop to human-on-the-loop: How AI agents are redefining finance

Editorial

This content has been selected, created and edited by the Finextra editorial team based upon its relevance and interest to our community.

At Nacha’s Smarter Faster Payments 2025 conference in New Orleans, industry leaders from BNY and PayPal offered insights into how financial services are leveraging AI. One of the most striking shifts discussed was the move from human-in-the-loop systems, where humans are directly involved in decision-making, to human-on-the-loop models. In this new approach, humans maintain oversight, but AI agents handle most of the decision-making process, making the system more autonomous and ensuring human control and accountability remain intact.

For financial institutions, this shift offers immense potential for operational efficiency, enhanced fraud detection, and more personalised services. As AI continues to evolve, the role of humans is transitioning from active decision-makers to overseers of AI-driven processes. This will be a critical development that promises to redefine customer interactions, risk management, and even payment systems.

BNY using AI as a structural shift

Narayanan had a clear message: the integration of AI isn’t just a technical task — it’s a cultural transformation. “We’re in a race against time,” he said. “We’ve taken a platform-based approach. But it’s also a cultural change.” When leveraging AI, BNY is focused on efficiency, particularly within its internal operations. “Most use cases are in-house. First, we look at how we maximise efficiency within. We are partnering with organisations on AI, but not for specific use cases.”

Narayanan also described how payments infrastructure is becoming more complex. “Payment rails have blurred — there are so many. There’s a lot of fragmentation in business models — different agents look at different things.”

PayPal on AI’s evolution

Juvariwala placed today’s AI landscape in historical context. “Web 1.0 was the translation of physical copy to digital. Web 2.0 was interactivity with an online portal. The next evolution is not only interacting — it’s moving to prediction.”

He went on to offer: “We’re at a very interesting juncture where a lot of organisations have begun implementing AI. It started five to seven years ago with machine learning. It started getting used in different spheres — and is being used more and more. We’re on the cusp of a significant shift in how we do business.”

On changing consumer behaviour, he added: “Consumers are not going away, but the way they interact with us is going to change. How should we react to make sure we meet their expectations? How are we competing with our peers?”

Juvariwala also pointed to how AI is already impacting search. “The use of ChatGPT as the primary search engine is for fewer than 15% of those in the room. The actual numbers are starker. Google will be taken over — you’re no longer searching for the ‘what’. With ChatGPT and its future iterations, you’ll search the internet across retailers. It’s moving from specific search to how we think.”

AI in fraud, compliance, and automation

Juvariwala also gave insight into PayPal’s core focus areas. “PayPal started [using AI] in the last few months — things are changing every day. What I share here may be outdated next week. The primary [use case] is fraud protection and prevention. We started with ML, applied it pre-Covid. It’s been further enhanced with AI, now we predict and analyse fraud patterns.”

He noted the scale of focus within the company. “25% of PayPal is focused on fraud and compliance. PayPal is in the business of managing risk, and building trust. Fraud protection is a fundamental tenet of our business.”

On the consumer side, he said: “Our digital wallet is evolving into a commerce platform. It offers recommendations and insights.” Customer support is also evolving rapidly. “When you interact with Contact Us or Help, it’s first with an AI bot. It gives you standard responses and escalates if needed to a human.”

He closed by pointing to newer, more experimental areas, such as AI in cross-border payments and cryptocurrency. However, he also cautioned against complacency. “The more we invest in AI for risk and fraud, the more robust our engines are. But we do not have an upper hand — fraudsters have the same tools. It’s not like with AI, fraud has gone to zero.”

The future is now

Juvariwala highlighted the growing role of regulators in the AI landscape: “Regulators are increasingly active in this space. They’re communicating more with each other across countries. A bank is currently hosting a group of regulators for an annual discussion on how things are shifting.”

He also stressed the need for greater centralisation: “There’s a need, more than ever, for centralisation of asks from jurisdictions. Database mapping is evolving. It’s not static — India has initiated a list of sanctioned entities, China is imposing new rules. The world evolves, and AI tools are essential to keep up with these changes.”

Turning to the future of AI in finance, he spoke about agentic AI: “Agentic AI — online agents that will service tasks for us. There are different sorts of agents online, broadly forming a category of use cases — they accumulate tasks and service the final end solution for you.”

Juvariwala’s closing remark brought this evolving future into the present: “Amazon launched Buy For Me — it’s no longer the future, it is here right now.” He also discussed the shift from human in the loop to human on the loop. Instead of a human being directly involved in decision-making, we are moving towards a model where the human oversees AI-driven decisions, providing oversight but not direct intervention. This is the next phase in AI deployment where we will see more empowerment of the system to operate autonomously, while still keeping the human in a supervisory role.

Comments: (0)

/ai Long Reads

Lynn Housecroft

Lynn Housecroft Professional Support Lawyer at Squire Patton Boggs & SPP member

The quiet revolution of AI in pensions

/ai

Hamish Monk

Hamish Monk Senior Reporter at Finextra

What is deep learning?

/ai

Editorial

This content has been selected, created and edited by the Finextra editorial team based upon its relevance and interest to our community.