Could AI be the Magic Bullet for Open Banking?

It could be argued that Open Banking hasn’t yet lived up to its potential. Although seven million people and businesses used Open Banking systems in January of this year, that is still a small percentage of the 156 million bank accounts in the UK. Companies like Bud, Plaid and Tink are successful on their own terms, but don’t match the rapid growth of companies like Klarna and Revolut. Amongst the general public, awareness of what Open Banking does is low. For this reason, some people are already asking whether we should stop trying to make Open Banking happen.

While this has been happening, 2023 has seen a surge in interest in the use of artificial intelligence (AI) in just about any and every industry you can name, and finance is no exception. The success of Large Language Models (LLMs) like ChatGPT have encouraged the proponents of Open Banking to suggest that maybe this is what Open Banking has needed. Perhaps by merging AI and Open Banking we can create the ‘killer app’ that will bring Open Banking into the mainstream – just as MP3 players existed for years before the iPod made them mainstream and in turn spurred the creation of smartphones.

With this in mind, let’s look at the current state of Open Banking and AI and ask the tough questions about whether there is potential here.

Open Banking in 2023

Since 2018, when the PSD2 directive required banks to open their APIs to authorised third parties, hundreds of companies have registered to use Open Banking protocols in the EU and UK. The latter has been particularly active in adopting the new system: as of 2022, 559 third-party providers were registered in Europe, and 221 of them were registered in the UK, reflecting the country’s strong standing in banking and technology.

Most of these registrations were either Account Information Service Providers or combined AISPs and Payment Initiation Service Providers. This seems to reflect that Open Banking is proving its worth more in the provision of innovative financial services rather than being an alternative to card payments at checkouts and eCommerce sites.

As mentioned above, Open Banking use is relatively low – there were 17 million users of Buy Now Pay Later services in the UK compared to the figure of seven million quoted above. However, it has been growing, and could go from being used by 15% of the public to 44% by 2027 at current rates of growth. That rise may not come from big name blockbuster companies like Klarna, but through Open Banking services being integrated into common processes: when a consumer signs up for mortgages or loans an Open Banking system could assess their bank account, for example. Most consumers wouldn’t even know that an ‘Open Banking’ transaction has taken place, it would just seamlessly integrate into processes that they already use (subscription payments being another example).

Enter AI

AI has been a part of payments and Open Banking for a long time now. It would be wrong to say that integrating machine learning and big data into financial services is a new phenomenon – a good part of those newly-registered AISPs will be using some form machine-learning, even if it is just a sorting algorithm, to make decisions based on customer account information. Even the most basic banking systems have elements of ‘if X then Y’ logic, such as the system that sends you a text if you are overdrawn. There are no humans making that decision, so it could be classed as a form of ‘artificial intelligence’.

In 2023, when people talk about AI, they generally aren’t referring to these relatively common systems that essentially function as flow charts, but a new generation of ‘generative AI’, pioneered by ChatGPT. This differs from machine learning and decision-making algorithms in that it is capable of creating ‘seemingly new, realistic content’. In practice, this means that you can ask LLMs to create ten taglines for a new product launch and it can do so, even producing entirely new content.

For example, we asked ChatGPT to write a paragraph about the adoption of AI in Open Banking applications and it produced the following:

“The adoption of Artificial Intelligence (AI) in Open Banking applications has witnessed remarkable growth, revolutionizing the financial industry. According to a recent study by McKinsey, AI applications in banking have the potential to create $250 billion to $350 billion in value annually. One of the key areas where AI is making an impact is in fraud detection and prevention. A report by Juniper Research predicts that AI-powered systems will save banks over $40 billion in annual fraud-related expenses by 2023. Moreover, AI is transforming customer experience in banking. A survey conducted by Accenture found that 79% of customers are more likely to trust AI-powered recommendations for their banking needs. These statistics highlight the significant role AI is playing in Open Banking, driving efficiency, enhancing security, and delivering personalized experiences to customers.”

This highlights the current generation of AI’s strengths and weaknesses. The paragraph it generated was grammatically and syntactically correct, but it also didn’t address the question, which was about Open Banking and not banking in general. There was also a much more significant problem: the three studies it cites don’t exist. This is because ChatGPT isn’t looking for evidence to back up its claims like a human would when creating an article based on that information, it is looking at patterns in existing content and concluding that McKinsey, Juniper Research and Accenture are the kinds of companies that are cited in articles about the banking industry. Even asking simple questions like ‘how many words are in this sentence?’ will produce wrong answers because it isn’t ‘thinking’ in the traditional sense (it says there are 6 words in this sentence, at the time of writing).

The kind of tasks that need to be done in Open Banking applications aren’t well-suited for this kind of technology. Being able to look at a bank account and see where savings can be made or whether a potential borrower can afford a loan is something that can be done with existing technology, and what LLMs can do can’t be applied to this. There are likely to be many advances made in AI, but, in short, the progress being made in LLMs just isn’t what Open Banking needs. Although it could have some application in customer service, where chatbots are already in use, until generative AI can sort fact from language that just feels fact-like it isn’t likely to be what Open Banking needs to push it into the mainstream.