Dutch bank ABN Amro has partnered with AI automation company Complidata to streamline trade finance operations. Announced on May 30, Complidata’s TradeSpeed solution will be brought on to help the bank’s staff analyse the multitude of physical documents in trade finance transactions, including letters of credit (LCs) and documentary collections.

Complidata’s TradeSpeed solution leverages various AI techniques, such as large language models (LLMs), a way of teaching an AI to understand human language, to parse document information and increase the level of digitisation possible, allowing it to extract key data and check documentation for compliance.

TradeSpeed was launched in 2022, and has been used by over 20 financial institutions, Complidata tells GTR. It competes with similar products such as Conpend’s Trade AI and Cleareye’s Cleartrade software, but is one of the first to use LLMs. According to the company, LLM usage allows for “the accurate identification of discrepancies with minimal false positives and expert-like automated narrative generation”.

Complidata’s CEO Ben Arber explains to GTR: “We built AI models for data extraction, rules for document checking against LCs and AI models for compliance processes, with deep natural language processing (NLP) and machine learning capabilities. We started using GenAI and LLM last year, which moved the dial in a few ways, such as Swift message analysis.”

Burak Aslan, lead product owner in trade finance at ABN Amro says of the deal: “TradeSpeed will empower our operations team members to focus on making decisions rather than completing time-consuming tasks. This will ensure increased capacity for value-added service towards our clients. Building on that foundation we envision exposing our capabilities to clients in the long term, so they can benefit from our capabilities as well.”

Arber explains that the software is meant to augment existing workflows rather than operate as an independent process, helping to lessen the impact of any errors: “Our clients use TradeSpeed as a co-pilot, helping empower human decision-makers use their time more efficiently and making decisions rather than running processes. The AI models and rules within TradeSpeed and other Complidata solutions are supervised with a high level of focus on model governance and monitoring to ensure against drift and to optimise results.”

He says that Complidata’s models were trained on trade finance documents, achieving high performance with as few as 10-15 documents for new models, a significant reduction from the over 100 documents initially required.

Complidata may not have a first-mover advantage for long, however. Conpend’s CEO Marc Smith tells GTR that it will bring LLM integration into its Trade AI product in Q4, despite some questions over the value of the technology: “One of our conclusions [from internal testing] is that LLMs should only be applied as a last resort as they are slow and expensive and require extensive tuning to reliably produce consistent outcomes.

“For many applications within the trade finance domain, traditional NLP is still more accurate and significantly faster. However, they can clearly add value in the interpretation of larger bodies of unstructured text, such as the unstructured Swift fields and paragraphs of text within documents.

“These are, therefore, the applications that will be included in our Q4 release.”

ABN Amro’s trade finance business has faced challenges in recent years. Once a giant of commodity trade finance, ABN Amro dramatically pared back its exposure to the sector in 2020 following a string of frauds in Singapore, the US and Germany.

“The bank has since then focused on growth, customer service and innovation in trade finance,” says a spokesperson for the bank.