Will agentic AI overhaul trade finance operations?

At the ICC Banking Commission Annual Meeting in Paris, Trade Treasury Payments (TTP) spoke with Dominique Honoré, Global Head of Global Trade & Commodities at Crédit Agricole CIB about how artificial intelligence is being used across trade finance, and what it will take to scale its impact.

While artificial intelligence has been widely discussed across transaction banking, Honoré said that the clear starting point will be with small, targeted use cases that are embedded within existing workflows. “We already see practical cases where we are using AI to facilitate all the micro-processes,” she said.

These early applications are focused on areas long known for their manual effort, including data extraction, document checking, and the drafting of guarantees. By automating these tasks, banks are beginning to reduce processing times and improve consistency, without fundamentally changing the structure of operations.

From rules-based automation to agentic systems

Traditional tools such as robotic process automation (RPA) tend to operate within fixed parameters where the expectation for a tech system is to execute predefined rules without deviation. While effective, these remain limited in scope. “It is really based on rules, and the robot will execute the rules…[but] it will not go beyond,” Honoré said.

Agentic AI introduces a more flexible approach with systems that can extract information from documents, interact with internal systems, and ultimately adapt how it performs tasks within defined boundaries. 

This distinction is beginning to influence process design at number of banks, particularly in determining how these agents can be integrated across front-to-back workflows. Given its reliance on standardised rulebooks, trade finance is one specific banking discipline that may be particularly well suited to AI deployment.

In the case of guarantees, frameworks such as the ICC’s Uniform Rules for Demand Guarantees (URDG 758) provide a clear and structured foundation for training AI systems. “The fact that there is this framework of rules… makes guarantees a very good candidate for applying AI tools,” Honoré said.

By embedding both regulatory standards and internal policies into AI models, banks can automatically identify non-compliant clauses and inconsistencies between documents. A similar approach is being explored in documentary credits under UCP 600, where AI is being tested to identify discrepancies across large volumes of documentation. Here too, the technology supports analysis, but does not replace judgement. “We need, at the end of the day, a human expert to make the final decision,” Honoré said.

This change is liable to alter the work that trade finance professionals perform each day. Routine tasks such as manual document checks and data entry are expected to decline, meaning that more and more these roles “will focus on exception handling and risk analysis,”  Honoré said, adding that this means, “we will need people who have that risk culture and also analytical skills,”

In that sense, the adoption of AI is as much about organisational change as it is about technology.

Managing risk in an AI-driven environment

Unfourtnately, AI capabilities expand, so too do concerns around fraud and document manipulation. As tools become more powerful, bad actors can also use them to create fake documents that look very real. This makes it harder to spot fraud using old methods. But at the same time, AI can help banks detect patterns and flag anything that does not look right.

To manage this risk, banks are putting strict controls around how AI is developed and used. “We need to go through… validation from various departments of the bank, risk, IT, security, compliance, data protection,” Honoré explained. Each of these teams checks a different part of the system to make sure it is safe.

There are also technical controls built into the systems. For example, only certain people can access sensitive data, information may be encrypted so it cannot be read if stolen, and every change is recorded so it can be tracked later. This creates a clear record of what happened and who did what.

While these critical security steps can slow things down, they are nonetheless important. In trade finance, mistakes can be costly, and trust is critical. By putting these controls in place, banks aim to use AI safely, while reducing the risk of fraud and improving how they manage operations.

Scaling from use cases to infrastructure

So far, many banks have tested AI in small areas, like checking documents or extracting data, and the pilots show clear benefits. But using AI in one part of the process is very different from using it across the whole system. 

For Honoré, the main issue is not computing power, but data and readiness. “For us, it’s more how we can standardise processes… how we make sure that we have a quality of data and that the organisation is ready,” she said. 

At Crédit Agricole CIB, this meant starting with the basics. The bank built a trade data hub to bring all its data into one place and make sure that it is used across all teams as the unique reference. This took time, but it created a strong foundation, and without it, scaling AI would be both difficult and risky.

The goal now is to go further. Instead of using AI in small pockets, banks want to embed it across the full process, from front office to back office. “The real benefit will appear when we have designed all the architecture… and incorporated agentic AI on a really scalable manner,” Honoré explained.

For now, the industry is still in the middle of this shift. AI is already proving useful in specific tasks, but full transformation will take more time. It depends on getting the data right, aligning processes, and making sure the whole organisation is ready to work in a new way.