Ben Arber, CEO of Complidata, a provider of artificial intelligence automation and compliance solutions, shares his views on the transformative impact of AI in the trade finance industry.

 

Let’s get this out of the way up front: ChatGPT did not write this article about the use of artificial intelligence (AI) in trade finance. As the incoming CEO of a fast-moving, dynamic, technology-focused scale-up, however, generative AI has been a significant help. Now halfway through the clichéd ‘first 100 days’ I have already used it to help produce job descriptions, take minutes of board meetings, generate estimates of target addressable market (TAM) for strategic decisions, and several other tasks that would previously have taken more time or expense to complete. And would probably have been of lower quality if I had done them on my own.

What are the sources used by generative AI? What would a regular internet search yield versus the large learning model (LLM)? What are the risks of using it for everyday tasks? Some of this seems obvious to me. Anything I produce in my name, no matter how trivial, I check for quality, tone, relevance and ethics – and of course acknowledge any external sources. And Big Tech is relatively open about how its LLMs assess/intake data and the associated risk. But there are bigger questions at hand regarding the power of AI in general, and generative AI specifically, and its potential to influence human development. The big step forward is a combination of the ability to ask very specifically for what one actually wants, and to have the level of specificity, detail and (probably) accuracy one expects.

Regulation in the EU, a letter from influential billionaires calling for the cessation of research and development, and mild hysteria in the media: there is plenty to consume on the potential dangers of AI. I happen to be from the Michio Kaku school of thinking, which has long held the view that machines will be dramatically transformative for the human race but will never acquire common sense. The rise of Skynet is hopefully not imminent or probable.

Here are five potential ways the trade finance industry may be impacted in the medium term by the advance of AI:

  1. Document processing

Technologies like imaging, optical character recognition (OCR) and cognitive processing have long promised major advancements in the extraction of data from bills of lading, invoices and other trade documents. But they have not delivered. In fact, OCR has been so underwhelming in trade finance that its use across banks is extremely limited despite its availability for over a decade. New Intelligent Document Processing models built using natural language processing (NLP) specifically for individual document types and trained on thousands – sometimes millions – of documents, therefore truly empowering machine learning, is rapidly changing this current over-promise and under-achieve situation.

  1. Document checking

Whether it is a terms check of documents under UCP 600 against a letter of credit (LC), a workability check for an LC issuance, a documentary collection or a bank guarantee, AI is making a difference. Long believed to be one of the processes untouchable by technology due to the level of learning required to spot a discrepancy, import/export document checking is suddenly being automated. Ironically, traditional rules-based engines are powering the disruption, albeit supported by advanced technologies like NLP and automated discrepancy language generation.

  1. Compliance

Handling large amounts of usually irrelevant sanctions hits has long been the bane of trade operations folks. In addition, manual red flag checks under AML policies introduced over the last decade have added significant responsibility and time to transaction processing. Reducing hits in sanctions and automating red flag identification have both suddenly and recently become major potential wins for commercial banks handling trade finance transactions. In addition, the holy grail of improved counterparty risk assessment and entity resolution may be revolutionised by generative AI. The ability to better detect shell companies, ultimate beneficial owners and nature of business within banks’ trade finance clients’ buyer/seller base at a transaction-by-transaction level – without using an investigator or a human research-powered analytical tool – could be very high.

  1. Automated decision-making

End-to-end processing or delivering on the age-old promise of ‘straight-through-processing’ is no longer the unobtainable dream. The machines do not just generate language describing a discrepancy, or a sanctions hit, but can decide whether to progress or interdict. Of course, the human decision-making on top of the AI – whether it is a four-eye process with the first two eyes being a robot for example, or a threshold above which every decision becomes a recommendation for a human – means that there are always levers and dials that can be adjusted up and/or down. But the automation of decisions, both commercial and compliance, has been a reality for some time now in the most advanced trade finance organisations.

  1. A 360-degree view

Trade finance is a bit of a paradox inasmuch as it is inherently global and yet any individual bank only really sees its own flows. AI flips this on its head, allowing any bank to benefit from comparing client transactions with industry-wide data, anonymised or destroyed via confidential cloud computing for example, to ensure compliance with regulation, data privacy and client fiduciary considerations. This immediately opens up a world of opportunity, both in the ability to make decisions on any specific transaction and also to assess risk on a non-client counterparty, for example, understanding what many peer lenders think (shell company? good credit?) in addition to their singular view.

 

Where is generative AI in all of this, I hear you ask? I’ll tell you: it supercharges document identification and data extraction. Looking across millions of publicly available trade documents means that decision-making is a lot more accurate and explainable (yes, to auditors and regulators) than you might expect. And compliance processes in trade finance benefit significantly from generative AI experience in non-trade transactions such as payments and global markets products.

Trade finance, together with receivables and supply chain finance, is changing rapidly. Of course, the same was said by many people of technology available in the 80s and 90s, and yet paper bills of lading are still in circulation. But as we probably said back then as well: this time it feels different.