For banks, the technology could transform the cost base, deliver previously unimaginable levels of personalised customer service and reinvent a banking career, whether working on the frontlines in the contact centre or analysing datasets for investment clients. Amid all the hype, however, it’s important to remember what generative AI is and what it isn’t.

“It’s a very large general purpose neural network, trained on trillions of words with billions of parameters,” said Edward Challis, Head of AI Strategy at UiPath. “It can generate completely new data, often in response to a natural language prompt, and because it can see so much data, and read so much of the Internet, it has what we might call common sense and some kind of inferential, deductive like capability.”

But while ChatGPT can pass business school, law and legal exams, its tendencies to “hallucinate” mean its outputs still need to be fact-checked and it still can’t reason and learn as effectively as humans³. It is advancing fast, however, with ChatGPT3.5 already outgunned by ChatGPT4.

What does this mean in action? For banking, almost everything. It’s hard to think of any job in financial services that could not be enhanced by the power of generative AI. “Whether it’s customer service or compliance or capital markets, so much of banking involves listening, reading and writing,” noted Challis. “Almost all of the day-to-day functions could be augmented and improved by using generative AI.”

Obvious candidates are frontline customer services, with generative AI able to handle huge volumes of queries quickly and accurately. One study⁴ found that customer support agents using a Gen AI tool saw a nearly 14% increase in their productivity, largely because the agents could participate in multiple chats at once and spent about 9% less time per chat. Customer satisfaction metrics were unchanged, which suggests the improved productivity didn’t have a negative impact on interaction quality – though nor was it improved. It also acted as an accelerant when it comes to competence, with agents with just two months’ tenure who used the AI tool performing as well as an agent with six months’ tenure working without the tool.

“It completely changes the experience of banking for the banker,” said James Longstaff, Vice President – Innovation Network at Deutsche Bank. “Our people spend so much time gathering data that it stops them from being able to read the room, focus on the client and add real value.”

Indeed, Christian Fjestad, Head of Innovation at SpareBank1 Østlandet, said one of the most exciting things is the potential to offer high net worth banking experience to the mass market as generative AI does the legwork of serving up context and accessing product and account information. “We can use it to hyper-personalise products and services, which is something customers increasingly expect” he said.

Because the technology is best suited for working with recurring patterns, and spotting anomalies within a data set, it’s a potent tool for assessing credit risk and flagging potential fraud. On the wholesale side it can support the research department by analysing huge volumes of data and surfacing what’s relevant in a standardised format. Any high-volume process or repetitive task is a candidate for deploying the technology.

Triaging for AI

Given the wide scope of generative AI’s talents, banks will need to think carefully about which areas of operation to prioritise for its transformative touch. Most are expected to start with internal operations first in order to reduce the risks as they build comfort with the technology. 

“We need to be 100% sure of the technology before we start letting our clients have conversations with it directly,” said Christian Fjestad of SpareBank1 Østlandet.

Banks will need to assess which areas to prioritise for ChatGPT deployment based on three criteria. “First, technically, is there a model we can get to work, and how will it integrate with our systems,” said James Longstaff of Deutsche Bank. “Second, the business case: does the technology make something better, faster, lower risk or make something economic that wasn’t before? And third, is it responsible? This last one is essential for a bank.”

This questioning approach is essential for an industry that lives or dies by its reputation. Banks that have earned the right to act as custodians for their customers’ money and personal and financial data through centuries of prudent management cannot afford to risk their reputations amid the hype of AI. “What is the problem you are trying to solve with this technology and is this technology really the best solution compared to other tools?” asked Christian Fjestad of SpareBank1 Østlandet. “This is the problem when something is very hyped – you can have a cool solution looking for a problem and that means you don’t see the risks clearly.”

Humans: more important than ever

Against this backdrop of cautious enthusiasm, a phrase keeps cropping up. “The human in the loop.” While many fear about the impact on jobs once generative AI goes mainstream, those tasked with rolling-out this technology stress that the human factor has never been more important.

“The human factor will be critical to success,” said Christian Fjestad, stressing humans are essential to maintain the checks and balances of responsible AI. That is why, employees should be upskilled to leverage its potential and democratise expertise.

Edward Challis of UiPath said ChatGPT can be used to turbocharge the productivity of nearly all workers in a bank. “Writing an SQL query, for example, is beyond most people, and so they make decisions without answering these queries of a large database,” said Challis. “ChatGPT can upskill almost every employee to do this. It’s an incredibly powerful technology.”

The risks of this “levelling up” can be managed by using existing protocols. “Even with Excel there are lots of sheets proliferated that are slightly wrong and banks already have quite sophisticated ways of managing this,” said Challis. “Banks are great at managing risk and human error and from this perspective the industry is best-placed to adopt this technology safely and at scale.”

Cloud: the foundation step

Integrating disruptive technologies with all their potential and risks is nothing new given that banks have been navigating serial waves of technological disruption for the last 20 years. Now, however, carefully mapped transformation plans must accommodate a technology that heralds a tsunami of change across all areas of the enterprise.

For Edward Challis of UiPath, much of this work should already have been done to adopt cloud technology. “Generative AI is predominantly a cloud technology and unless banks have laid the foundation for cloud and cloud services this is going to be very hard,” he said.

This was echoed by James Longstaff of Deutsche Bank. “Our AI activities are so closely linked to the cloud journey we had already started,” he said. “That journey has given us a head start with our AI ambitions and we’re certainly an early adopter and collaborator on AI.”

Technology disruption in banking usually comes up against the buffers of legacy issues yet, counter-intuitively, these decades-old systems with their messy data silos and sticking plaster fixes, are prime territory for generative AI to add real value. “It’s a technology that can handle quite unstructured data that was generated by humas for humans,” Challis explained. “This complex legacy IT estate of the traditional banks means generative AI actually adds much more value to them than to a digital first business.”

Risk management: best in class capabilities

There are, however, risks attached to this new technology. There have already been high profile headlines about the fallibility of ChatGPT and its tendency to generate fabrications, regurgitate falsehoods and amplify bias.

Edward Challis of UiPath is confident, however, that banks are well placed to manage these risks. “Banks were some of the early adopters of using algorithms to make decisions on a client’s behalf,” he said. “Banking is already ahead of many other industries to mitigate and manage this risk without bringing all innovation to a halt.”

James Longstaff of Deutsche Bank agreed. “When it comes to our risk appetite, we do not have a free pass for this technology,” he said. “We have to get this right and think across the whole life cycle of a generative AI model, making sure we’re continuously testing and monitoring the data going in, the transparency of its outcomes, and checking for model drift and bias over time.”

Some of generative AI’s capabilities are a double-edged sword: it can be fantastic at spotting fraud signals but can also produce deep fake documents and audio-visuals that could circumvent security controls. The experts interviewed for this piece agreed, however, that banks are well-equipped to rise to this challenge.

“As an industry, we’re best in class when it comes to security and compliance,” said Christian Fjestad of SpareBank1 Østlandet.

From hype to game-changer

It’s clear this technology is a true game-changer. “It’s letting us do things that we could not do before because they were previously uneconomic,” said James Longstaff of Deutsche Bank. “Things that were buried are now possible. This technology has such potential it would be crazy not to use it.”

While once the industry feared the potential competitive threat posed by Big Tech, it seems many now think it is AI that will define future success.

“Banks that don’t use AI will get left behind so quickly,” warned Christian Fjestad of SpareBank1 Østlandet. “The speed is so rapid, you need to jump onboard now. There’s a learning curve and there are no short cuts – the organisation needs to learn and mature alongside the technology. You do not have the time to wait and see.”

The AI arms race has just become more competitive.