When it comes to making credit decisions, banks have tended to look backwards, not forwards. Increasingly, however, technology is allowing banks to rely less on balance sheets and more on up-to-the-minute transaction and invoice data. John Basquill speaks to industry insiders about what this change means for potential borrowers, and whether artificial intelligence (AI) heralds a new era for credit and risk management in trade finance.


Around two-thirds of SMEs in the US went into 2023 intending to apply for credit at some point during the year, according to a survey of nearly 500 businesses carried out by technology provider Codat. Many, however, are left “unable to access the basic capital they need”.

“21% of those we spoke to wanted credit in 2022 but couldn’t get it,” Codat says in a report published in May this year.

“Excluding those with a low credit score, 13.7% of surveyed businesses – the equivalent of nearly 4.5 million companies when scaled to the total US market – fell short due to limitations in the lending system, including overly complex applications, high costs and outdated decisioning processes.”

Much of that complexity stems from the applicant having to collate bank statements, annual financial information and other details, then provide that information to their prospective lender.

“Not all businesses are creditworthy,” Codat says. “But many of the most common reasons given for loans to be rejected, like a ‘thin credit file’ or ‘the business not being in operation for long enough’, indicate a lack of data rather than an inability or unwillingness to repay.”

Nearly three-quarters of SME respondents said they would be willing to share financial data directly with banks in order to simplify onboarding and even lower the cost of funding.

Doing so requires technology, however. That transaction-level data must be extracted and structured in a way that is useful for banks, and even then, further tools are required to carry out a thorough analysis of those findings. Across much of the financial sector, there is a growing sense that AI could bring that vision to life.


André Casterman is founder and managing director of Casterman Advisory, and has supported the growth of numerous technology and financial services start-ups, including trade finance technology provider Tradeteq and data management company Intix.


GTR: How are banks changing their approach to data collection and credit decisioning, and why is that important?

Casterman: There is a change from credit decisioning based on balance sheet information. The challenge around the world is that SMEs are the ones in need of funding, but in emerging markets, their balance sheet information is not always available or easily accessible. Also, by the time a balance sheet is published, it’s already old and that information might already be outdated.

More and more, it’s about using receivables data, payment data and statement data. If you can pump that information from a company’s accounting system, you can see how quickly invoices are paid, and how strong those relationships are between the buyer and supplier.

You can access the latest invoices, the latest payment flows, and really up-to-date information at a granular level, and use that for credit scoring.

One challenge is that doing that is clearly data-intensive, so it has to be automated, for example through APIs that can pull data in a structured way. You can then apply an algorithm to that and make decisions.


GTR: What role can AI play in this process?

Casterman: Where AI is great is that it can add insights from the unstructured world, bringing in context from the ecosystem in which that SME is operating. You can pull in information from all kinds of online sources about an SME and its ecosystem, but that’s going to be unstructured data. AI can interpret that, and add flavour to the analysis you already have from the balance sheet and transaction data. That’s where the magic is.

To give an example, AI might be able to point out that something negative is emerging in a particular sector – a relevant commodity price going through the roof, or some kind of political conflict – and this could impact that SME in a way the transaction data will not be able to tell you. AI is relevant when it comes to using new datasets, and getting a sentiment or a feeling of the macro environment as well as the financial situation.

In credit decisioning you do need external data to get to know as much about the client as possible, ideally beyond just what the transactions are giving you.


GTR: Are banks ready to start doing this? If not, what factors should they be considering?

Casterman: For a long time, I have been asking banks, ‘is your data house in order’? That’s not because of any commercial angle, but for compliance. Regulators are increasingly asking banks for details on specific transactions, delivered quickly, perhaps on payments a certain person has been making or on all payments above a certain amount.

And as a business, of course you want to be able to spot fraudulent activity. This means that in a lot of cases, that transaction data is already available.

However, it is still a challenge for banks. In many cases, that data could be spread out across multiple internal systems, and it’s very complex to pull that together in a linear way that is ready for processing by an AI engine. If all that data can be connected, in one context it could be used for anti-money laundering or compliance purposes, in another context for credit decisioning, and for other use cases too, such as liquidity management or forecasting.


GTR: When it comes to increasing financing availability for SMEs, are there still potential barriers that AI cannot address?

Casterman: SMEs might have that transaction data, but their funder doesn’t have it. What I hear in emerging markets is that SMEs aren’t always in a position to have reliable third parties come in and facilitate access to data for the funders. That means the data would have to be provided directly by those SMEs, so in effect, the funder has to rely on data that a potential future client has given them, and might not feel that is trustworthy enough to use in credit decisioning. Essentially, those markets may not be as organised, and an extension of that is a concern about access to data.


Gabby Macsweeney is head of policy and communications at technology provider Codat. The company provides an API that connects into small businesses’ accounting and payment systems, enabling users to extract and analyse detailed financial information on a continuous basis.


GTR: How is Codat’s API technology being used, and what kind of institutions do you work with?

Macsweeney: One of the main use cases is in lending, but we work with established banks, alternative lenders, as well as supply chain finance, invoice financing and B2B buy-now-pay-later clients. There are all kinds of indicators you can pick up from looking directly into small businesses’ financial platforms, and then by bringing multiple of those indicators together, you can add additional layers of insight.


GTR: How does AI fit into that? Do you have specific examples around AI being used for decision-making within a bank?

Macsweeney: AI is already being used in our products. If you can access open banking and accounting data digitally, you can start to cross-reference across different data types. For example, in invoice finance, you can look at invoice data from an accounting platform and cross-check that against banking information, to verify that payment has reached the bank account and that dates and amounts are accurate.

Another application is automating how data is processed, for instance from a bank statement or financial statement that might be in PDF form. Historically, some lenders have collected that kind of financial data and entered it line-by-line into their systems. The problem there is you need to be able to define transactions, putting them into specific categories. Doing that manually takes a long time and can result in errors.

Our platform can categorise spending automatically and place that data into a standardised format. Within our API, we have seen millions of transactions over time, and we have an AI engine that can look at all those different transactions to see how they have been labelled or categorised. You can apply that learning to future transactions or other data sets, and that can also be automatically filtered into a lender’s decisioning process.


GTR: What kind of role can AI play in risk management?

Macsweeney: Take the example of an invoice finance marketplace. In the past, you might see marketplace businesses that would let companies enter invoice data manually in order to seek financing. But increasingly companies are moving away from that. Now, marketplaces will connect directly into suppliers’ accounting systems and extract the information from there. That reduces the risk that invoice will be fraudulent, and also means you can access information about how customers have repaid those invoices over time.

Ultimately, you end up with a holistic view of invoice payments, and can more accurately calculate the risk associated with those transactions. So as well as benefiting from the efficiency that comes from automating manual processes, you can also have much better confidence in the data you’re using – which is more up to date – in order to reduce risk.


GTR: How is the uptake of AI progressing? Are there still barriers to adoption?

Macsweeney: Plenty of lenders are experimenting more widely with additional data sources and layers of intelligence they can apply to that, and especially among newer players, capturing that data digitally is widespread now. But for larger banks, anything that involves changing their risk structure is challenging. Those models are closely guarded and making changes is not always going to be an option.

Also, the level of governance within a bank means projects can be slow to put into action, and require a lot of tech resources and know-how. That all comes at a cost, and the cost is larger the bigger that organisation is.


GTR: Looking ahead, do you see banks relying on AI to make credit or risk decisions on a fully automated basis?

Macsweeney: There are some lenders that are aspiring for fully automated decisioning, offering pre-approved loans for smaller ticket sizes, where speed and convenience are what they’re leading with.

But that’s not really where most lenders want to play. What is more important is differentiating in terms of the relationship they have with their customers, so for them, automation is useful to remove some of the tedium of manual processes while giving more confidence and insight into the credit decisions they’re making.


The numbers: AI in financial services

In a survey of over 600 financial institution employees carried out by technology firm Quantexa, more than a third of respondents said their organisations were already using AI as of last year, or have pilots or proof of concepts already underway.

A further 25% reported that they were not yet using AI but intend to do so in the future, while around a third said they currently have no significant plans to use AI.

In terms of use cases, 19% of respondents said they use, or plan to use, AI for assisted decision-making, by building analytical models that can provide insight from data. A further 17% of respondents said they were seeking fully automated, data-driven decision-making using AI.

9% singled out credit risk decisioning as an application for the technology, while a further 20% saw a role for AI in risk management, including fraud detection and onboarding new customers.

The number of respondents who said their organisation would likely build AI capability in-house, compared to those planning to buy external solutions or use a system integrator, were roughly equal.