Artificial intelligence has long been hyped as revolutionary for supply chain finance. But despite some high-profile false dawns, practical applications of the technology are continuing to emerge and improve. John Basquill looks at the role of AI within the industry today – and where the next frontiers may lie.


In late 2018, the supply chain finance (SCF) industry was said to be on the brink of a technological revolution. A paper published in October that year touted “SCF 2.0” as a technology-driven platform approach to finance – albeit one still in its infancy – that would offer easier integration with buyers’ systems and smoother supplier onboarding.

But that paper promised “even greater change on the horizon”. SCF 3.0, it said, would apply artificial intelligence (AI) and machine learning to the vast amount of data stripped from invoices over several years, and use it to forecast the likelihood that a purchase order or invoice will be approved.

The author of that paper was Greensill Capital. In the two-and-a-half years following its publication, Greensill rapidly expanded its business, providing billions of dollars against invoices that had not yet been approved – a controversial product offering known as future receivables – before collapsing into insolvency in March 2021.

In that 2018 paper, the company argued that artificial intelligence “means the risk level of each invoice can be calculated instantly and accurately, so the buyer is no longer required to irrevocably approve the invoice first”.

“Greensill can make an early payment to the supplier quicker than ever before, upon purchase order issuance or invoice booking, because the data has calculated the risk,” it said. “The Greensill algorithm will also spot irregularities in invoice patterns, an indication of errors or fraud.”

With Greensill, along with key borrower Sanjeev Gupta, now the subject of an investigation by the UK’s Serious Fraud Office – as well as a probe by the Financial Conduct Authority and a criminal complaint from German federal prosecutors – that 2018 vision could not be further from reality.

Since Greensill’s demise, suggestions have emerged that its technological offering was not as advanced as that paper claimed. At the time of its insolvency, most of its risk assessment process was actually carried out using spreadsheets and basic programmes, people familiar with the company told Bloomberg in April.

In the days following its collapse, there had been hopes that Athene Holding Ltd, part-owned by private equity investor Apollo Global Management, would take control of the company’s IT systems, as well as its intellectual property and the bulk of its investment-grade SCF book.

However, that deal fell apart when it emerged Greensill had actually been reliant on Taulia’s SCF platform as the practical means of onboarding and paying suppliers, rather than its own technology.

That meant its SCF clients were actually considered customers of Taulia, a San Francisco-based working capital solutions provider, and with Taulia already having made the switch to a multi-funder model – meaning Greensill had little or no technology of value – Athene walked away.

Greensill, as it turned out, was not about to pioneer a new generation of SCF driven by artificial intelligence. Nevertheless, the technology remains a source of excitement for much of the industry. The key, experts suggest, is to look beyond the hype.


AI: hype versus reality

“Artificial intelligence is somewhat an overhyped term, and a very big container word,” says Marc Smith, founder and chief executive of trade finance-focused technology firm Conpend. “There’s a lot of confusion around AI, and I would be willing to bet that 80% of what’s labelled as AI, by very strict definition, is not.

“We have been working for years with artificial intelligence as a core part of our solutions. The benefits go well beyond the digitalisation of processes and enable a new way of handling data and knowledge.”

Smith explains there is a technical distinction between programming that will determine particular outcomes depending on the inputs, and a system that will attempt to make sense of something occurring outside predefined parameters.

“In many cases, you are just giving instructions – like a rule set – to a computer, and it just goes and runs them. With AI, rather than following a set of rules, you can give any input and it will try to do something with it that could be perceived as intelligent,” he tells GTR.

Colin Sharp, senior vice-president for Emea at supply chain finance provider C2FO, suggests that much of what is marketed as AI in the trade finance industry is more accurately described as data science.

“Some market participants talk about using AI in the supply chain, when really what they are doing is collecting and analysing data, and using that to make a decision themselves – such as the rate they offer a supplier,” he tells GTR.

“Really, data science would be the appropriate term for that. It doesn’t have cognitive reasoning, self-learning or anything like that.”

Sharp says that approach is of course legitimate – C2FO has its own data science team and collects reams of data from internal and external sources, he explains – but that it “doesn’t always give you the answers you need”.

“If you take the question of what rate to offer a supplier, for example, what if the right data isn’t available? What if you don’t know the cost of alternative financing for that supplier?” he asks.

“Our view is that the person seeking funding is in the best position to have a sense of what rate is suitable for them, based on their circumstances at that point in time, because that changes – and there are some changes that no artificial intelligence engine would be able to pick up.”

In C2FO’s case, that model means the company can offer different rates to different suppliers based on the circumstances they present, rather than based purely on the data collected and analysed.

“C2FO uses data science not to second-guess the right rate for a supplier – we get real intelligence on this directly from the supplier naming their appropriate rate,” Sharp says.

“The smart way that we use data science is operationally, in terms of how we service hundreds of thousands of suppliers simultaneously to deliver a good service to them.”


Practical applications

Hype aside, there remains real excitement about the potential benefits that artificial intelligence could bring to the SCF industry. But unlike Greensill’s promise of a technology-driven revolution, the reality is that existing processes are streamlined, sped up or made less prone to human error.

As Conpend’s Smith says, there are “real practical applications of AI within the trade finance and supply chain space” already in existence.

“One of the obvious examples would be around compliance, by picking up unusual activity for example. Artificial intelligence is very good at detecting anomalies in documents and transactions,” he suggests.

“Another would be extracting data from documents. Across supply chains, there could be millions of different kinds of invoices that need to be understood, as well as packing lists, bills of lading, certificates, and so on. Artificial intelligence can learn to interpret information that it hasn’t seen before, and present it in a useful way.”

Major financial institutions that use Conpend’s technology include ING Bank, UniCredit and Commerzbank, as well as a host of non-bank providers such as Finastra and Surecomp Marketplace.

Other banks have turned to AI for optical character recognition, to make paper documents machine-readable. For instance, Standard Chartered partnered with computing giant IBM in 2019 to digitise shipping documents, with the system using continuous machine learning to redefine and reclassify data inputs over time, gradually improving its accuracy.

When it comes to supply chain finance specifically, any technology that can streamline invoice handling will be welcome – whether driven by data science, AI or a combination of both.

Bob Glotfelty, vice-president for growth at San Francisco-based SCF provider Taulia, describes invoicing as “really complicated”. “It sounds like it shouldn’t be, but it is,” he tells GTR. “That’s because invoices move in the opposite direction to all the other document flows.”

In other words, while purchase orders and payments flow from the buyer to the supplier, invoices are sent the other way – so if a company has tens or hundreds of thousands of suppliers, it is integrating vast numbers of invoices from different sources into a single enterprise resource planning (ERP) system.

“The bigger suppliers might be able to set up API integration within that system, and automatically transmit data that way,” Glotfelty says. “But the mid-segment suppliers that might be sending a few dozen or a hundred invoices a month might not feel that kind of integration is fully justified.”

In that scenario, he says companies could use AI-driven parsing technologies to automate the receipt, sorting and processing of invoices.

In Taulia’s case, the company has partnered with Google to extract data from invoices, identifying different data fields and ultimately shortening the amount of time it takes to gain approval.

Another potential application is around ESG. Oliver Belin, founder of financial supply chain analytics platform Calculum, tells GTR that corporates are increasingly demanding their suppliers meet specific criteria.

“The world is changing. Companies cannot just look at their own goals, and optimise their payment terms based on that,” says Belin, a supply chain finance industry veteran who has held senior roles at TradeIX, PrimeRevenue and SMBC.

“Besides looking at suppliers’ financial metrics, country rating and industry-specific information, corporate buyers also need to look at environmental and sustainability factors, and incentivise minority-owned suppliers.

“The question is: How can you support your decision-making process based on that data? AI-driven systems can integrate all of that, calculating the best terms based on those various supplier characteristics and peer group data.”


The next frontier

For Torben Sauer, managing director of Conpend, financial institutions should not be seeing AI as a way of replacing staff with technology. “It’s about freeing up capacity to focus more on value-added tasks and working smarter,” he says.

“For instance, there could be the possibility to let the system handle 80% of the transaction handling – the easy ones – and leave the remaining 20% that require human thinking and decisions. Banks still want to have a human touch; not everything can be predefined.”

That means technology – even systems that automatically improve as greater quantities of data are collected – does not have to be perfect. “It just has to be better than a human being,” he says. “A system might not extract 100% of the right data, all of the time, but it could still be an improvement.”

Looking ahead, however, there are emerging technologies that could operate much more like a human brain.

“A lot of the more popular algorithms used by artificial intelligence were developed in the 1980s,” Conpend’s Smith explains. “A newer technology is deep learning. Instead of creating something that’s already quite specialised – like natural language processing, for example – deep learning is trying to go to a higher abstraction level.”

With invoice processing, for example, Smith says AI models are pre-defining the outcomes that banks, buyers and suppliers seek. “With deep learning, you can actually just say, ‘these are a bunch of documents, tell me what they contain’.”

Whether deep learning has a role in trade finance is an open question, however. Smith points out companies would need powerful machinery and vast amounts of data, more akin to the technology behind self-driving cars than behind banking and finance.

“At the moment, I don’t know if there are justifications to do that business-wise,” he says, “I think it makes more sense to digitise everything at the source than it does to develop a custom-built deep learning CPU, simply for processing documents.”