What ChatGPT and AI Mean for Banking

Isaac Asimov considered his short story, The Last Question, as one of his best works. In it, he talks about a world where people present their questions to “Multivac,” a super-computer fed with decades of data that has a solution to most problems humankind encounters. Rings a bell? The current hype that exists around ChatGPT makes some people think that we’re getting closer to having our own Multivac. Cheaper and cheaper computing power is ushering in a democratization of sophisticated artificial intelligence (AI) tools. These are transforming banking to make them more secure, more efficient and … more loveable. 

Although some naysayers have described ChatGPT as no more than an “stochastic parrot” designed to create plausible sounding text without any accountability for truth, it has spawned a plethora of use cases, some more disruptive than others, from creating Seinfeld’s scripts to arguing cases before the US Supreme Court. And while applications of AI in financial services have been around for a while, they are now reaching maturity and leading to mass adoption.  

I remember watching this video in 2016 about a virtual personal assistant provided by a bank. Since then, I’ve often endured irrelevant responses from clunky chatbots on WhatsApp and other media. But I believe ChatGPT will change much, especially the GPT part. 

GPT stands for Generative Pre-training Transformer, a technology that could finally make natural language processing indistinguishable from human interaction, making it possible to deploy chat- or voice-bots with success in more contexts. This would quickly push our expectations around instantaneous customer service solutions around the clock.

Beyond customer service, preventing financial crime is another area providing obvious opportunities to deploy AI. Using it to prevent money laundering could save banks a staggering US$217 billion by 2030. Underwriting could account for another US$31 billion. These two applications in combination with customer service could generate almost US$500 billion in savings. Let’s explore these applications of AI and imagine where this might lead in the near future:

Customer Experience (Cx): Chatbots Turning Consumer Banking Into Private Banking

When it comes to our money, we want things to run seamlessly. If there is a problem, we want it solved immediately, 24/7. In fact, poor customer service is ranked as the No. 1 reason for customer churn in banks. Given that acquiring a new customer is anywhere from five to 30 times more expensive than retaining an existing one (depending on which churn study you believe), it makes business sense to invest in retaining customers through good customer service. Moreover, good customer service can increase referability, reducing customer acquisition costs further.

AI makes turning customer support into a source of competitive advantage easier than ever before. According to industry research, 95% of customer interactions will be powered by AI within the next five to 10 years and by 2023, the operational cost savings from using chatbots in banking will reach US$7.3 billion globally, the equivalent of 862 million agent working hours! This would free people up to focus on higher value interactions with customers. 

Machine Learning (ML) models can be used to predict which customers might be prone to churn, trigger retention activities and report back to the business what would need to be improved to prevent customers churning for similar reasons in the future. Retention actions and responses to customer queries could be handled to a growing extent entirely by machines. 

Financial Crime: Anti-Money Laundering (AML) and Transaction Monitoring

In 2020 alone, institutions globally spent an estimated US$274.1 billion on financial-crime compliance. 

Let’s start with AML. Until recently, banks had relied on rules-based AML systems to flag suspicious activity, whether it be a transaction over a certain threshold or money flowing into a specific high-risk country. Such methods of detection are hugely inefficient: over 95% of system-generated alerts are closed as “false positives” in the first phase of review, with 98% of alerts never culminating in a Suspicious Activity Report (SAR). This is where AI comes in, as ML models can analyze larger quantities of more granular data to build sophisticated algorithms that are much better equipped to accurately spot suspicious behavior while reducing false positives. According to a Mckinsey report, one leading financial institution improved suspicious activity identification by up to 40% and efficiency by 30% by using AI-based tools. But in order to be able to implement AI-powered AML systems, banks must first have the right data architecture in place, which remains a challenge for many. 

Moving on to fraud, the fight has become increasingly complex as banking goes digital. Fraudsters continually evolve to produce phishing, scams, malware infection and ghost websites, among other innovations. ML algorithms can, among other things, help banks correctly identify their customers, block fraudulent transactions and blacklist criminals through the analysis of biometrics, device information and other kinds of data. Using AI to correctly identify customers not only allows banks to reduce fraud, but to offer an entire line of identity services: through BankID, customers in Sweden use their banking credentials to file for taxes or access their medical records. 

Finally, complying with regulation has become increasingly complex. In many countries, an unwieldy amount of pieces of regulation are passed every year, making it humanly impossible to stay on top of everything. But AI can sort through all of the regulation in seconds and provide the right answer. 

Credit: Collections and Underwriting

Anyone can lend money, but not everyone can collect. Chatbots and other AI tools can be deployed to streamline collections. Not just the interactions with customers but even determining at what time a customer is more likely to answer their phone or what kind of language they’re more likely to respond positively to.  

Credit assessment is perhaps the area where AI has come under more public scrutiny, simultaneously causing hope and fear. Hope, because AI applied to credit analysis could improve financial inclusion for traditionally underserved borrowers, so-called thin-file credits, as it makes it possible to use more, and different kinds of data to offer better credit risk predictions. Fear, because this could lead to discrimination we may not feel comfortable with. For example, having a child with cancer or being hospitalized could increase the probability of bankruptcy. Should an algorithm be blind to this data? In many jurisdictions it may not even be legal to use many statistically significant variables to make credit decisions. 

Moreover, the transparency risk that arises from black-box models is a major concern for regulators across the world. To control this space, the EU AI Act is likely to become a blueprint for AI regulations around the world, like GDPR became the benchmark for data protection regulations. The Act includes the financial services sector in the list of sensitive industries and AI systems used to evaluate creditworthiness or establish credit scores would be subject to the mandatory requirements for high risk AI, including the need for human oversight. 

Zooming out from specific use cases, AI could be used to automate “money management” for the benefit of users. This could involve auto-savings, scheduled bill payments, investing according to a risk profile, nudges to help us make better purchasing decisions, avoiding interest payments whenever possible, and so forth. There are companies already out there offering several of these features individually. They could soon be orchestrated into a coherent offering, which brings us back to Multivac and virtual assistants. And I’m not sure banks are best placed to build the best virtual assistants.

Two quotes come to mind: Sir Ronald Cohen’s “Revolutions come from outsiders” and Bill Gates’ “Banking is necessary, banks are not.” It may be that the automation of banking that disrupts the market does not come from banks at all but from technology companies with advanced virtual assistants. One reason for this might be that these virtual assistants would be well placed to shop around different financial services providers to get the best deal for their customer without a conflict of interest. Another is that embedded finance can support more natural interactions through channels customers enjoy interacting with. We already have watches acting as our health coaches, why not also our wealth coaches? 

As Uncle Ben once put it: “With great power comes great responsibility.” Doubtless, AI is becoming very powerful indeed. Banks and firms with the wisdom to reap the benefits while averting perils such as deep-fakes and discriminatory black-box underwriting models, will leapfrog from being considered a necessary evil or a commoditized service provider to becoming an ally who is there 24/7 and truly speaks your language. 

We, at Revolut, are deploying state-of-the-art AI solutions across the business. If you are interested in being part of the revolution, check out revolut.com/careers. 

(In collaboration with Eduardo Haro)