AI: the changing face of banking
As managing senior managing director at Accenture, Edwin van der Ouderaa's job takes him all over the world. After touching down in London, Singapore, New York or Tokyo, his day is dominated by meetings, lectures and consultations about the future of digital banking, and he's beginning to suffer from déjà vu.
"I often see the same people pop up in all places across the world," explains Van der Ouderaa. He calls them 'the expats': those people who are highly prized by banks for their skills in the digital realm, and - due to that expertise - as peripatetic as he is.
"There really is a shortage of these types of skills," he says. For the Accenture executive, however, this trend is to be taken less as a hindrance and more as symptomatic of the massive transformation the banking industry has undergone in the past decade or so. When Van der Ouderaa began in the job, he was busy lecturing bankers about how the digital revolution was just around the corner. Now, chief executives are telling him about their new fintech projects.
"The conversation has shifted so much," he says. "Every conversation involves the CEOs telling us, 'Everything is getting disrupted', or 'I'm going to start a big transformation', or 'I have started it, and these are my challenges. How do I solve them?'"
What's more, the conversation has shifted onto a markedly new terrain: how banks should properly harness the power of artificial intelligence (AI) - a term that, until recently, was the preserve of computer scientists and a handful of pulp sci-fi writers - has become the question of the hour.
The meaning and the machines
"First, we need to distinguish what that term really means," says Van der Ouderaa. "We have, on the one hand, the latest versions of what we call predictive analytics, which are really predictive mathematical formulas. Then we have machine learning, which is, essentially, neural networks."
Some people lean on this form of machine learning - a set of algorithms that interact with one another in a similar manner to that of the human brain's neurons - as a signifier of 'true' AI. Van der Ouderaa is not convinced. "I think they're just another class of mathematical algorithms that are very powerful," he explains. "They can capture a lot of use cases, where you can automate specific responses to specific situations."
In finance, they tend to thrive in situations that require a large dataset within which no pattern of information is immediately discernible to a human observer, or - in layman's terms - they find needles in very large haystacks. Detecting when money laundering is occurring is a prime example.
"You have a lot of fuzzy inputs that are kind of incomplete, or ill-defined, across lots of input parameters," says Van der Ouderaa. "But there's a pattern hidden in the set that you are getting, even if there's incomplete information, and then the machine learning solution will capture that, even when we, as humans, cannot."
They're followed by the chatbots. "They have what we call 'semantic engines', so that they can understand phrases and pieces of conversation, and can, without using external analysis, know how to respond to that. But they're still, in a sense, quite dumb," he says.
What they can do is provide the user with a simple response, which can work wonders in personal banking. Instructing a chatbot to send money to a certain account, for example, is a task that's fairly easy to program. "Most of what I do in standard dayto- day transaction banking as a retail customer can be done through chatbots," says Van der Ouderaa.
What the Accenture executive is most impressed by, though, are 'cognitive engines': AIs that showcase the ability to connect and reason from abstract concepts.
"They can just go through that thought process like we would, and then formulate an opinion on what's going on and respond with an action or a conclusion," he explains. "Those engines are what you need when you have complicated concepts to deal with, like a mortgage or life insurance agreement."
It is this version of AI that Van der Ouderaa is most excited by and sees the most potential for in the fintech space. "We can now create engines that will withstand the test of the regulators," he says. "Their Turing test, in a sense, is whether the engine can perform a sophisticated conversation and be emotionally intelligent, because one of the things that they need to do is to verify whether the user has understood what they are talking about. It needs to understand the user's true intent, whether they are going to understand all the implications of the product and if they are going to use it in an appropriate way."
For Van der Ouderaa, it opens up a wider range of possibilities for automation in banking than before. "You can imagine that with those types of engines, we could tackle a whole range of problems," he explains. "We can do quite complicated regulated advice, and I would say that, with the current type of engines, that would be the bottom half of the simple, regulated advice. But they're getting better. My guess is that in ten years' time they will be much better than human beings."
Van der Ouderaa predicts that, by then, it will be humans being tasked with confirming the simple advice and computers the complicated, rather than the inverse situation as it currently stands. "When you then look back deeper into the back office, you can see there's an analogy there with all the more sophisticated back-office work that needs to be done," he says. "We already have robotic process automation that can take out between 30-60% of repetitive back-office work."
Already, he is fielding questions from Accenture's clients on introducing AI to handle these kinds of problems sooner rather than later. "They say their own people can't deal with the complex aspects of regulatory or internal compliance aspects, the sheer complexity of the products themselves or the vastness of the hundreds of different variations of the products," he says. "It's all too much for a human being, so they are looking for solutions along the lines of what we call 'augmented employees', whereby that person is still doing the advice or the handling, but he or she is accompanied by an engine that is advising them."
Whether they will or not depends entirely on the grace and favour of banking regulators. Luckily, it seems that they are on board, too.
"Most regulators we talk to are very supportive," says Van der Ouderaa. "They all understand that things are going to change, and that it's better to embrace the change and control it."
The advent of AI in banking also, potentially, makes their own job a lot easier. "The promise is, they hope, that these new solutions will give customers more stable advice," he explains. It could also radically improve the traceability of certain financial transactions, a prospect anticipated by forthcoming EU legislation that requires actions taken by machines to be explicable to regulators. The billions in euros, pounds and dollars collected in fines over the past decade have proved to be a powerful incentive.
"A lot of tier-one capital that could have been used to shore up the balance sheets of the banks has been used to pay noncompliance fees," says Van der Ouderaa. "In that respect, they want to be able to help the banks."
What they definitely do not want, however, are black-box solutions. It is notoriously difficult, and sometimes outright impossible, to track the reasoning behind why neural networks make certain decisions. Van der Ouderaa is firmly against the spread of these solutions throughout the banking industry, notwithstanding its reputation for divining previously unknown patterns and efficiencies.
"I know that machine learning can be very powerful for the specific classes of solutions that I've been talking about, but the problem is that, especially with neural networks, it's very hard to understand why they do what they do," he explains. "That's why I personally prefer cognitive engines - I believe that there you can trace the reasoning, as if you could see inside the brain of a human being. With machine learning, that's not really the case."
Of all the national regulators he has talked to, Van der Ouderaa has found that the UK's Financial Conduct Authority is the most receptive towards reviewing and approving AI-based solutions. The Accenture executive goes as far as to say that they're at the forefront worldwide, having set a key precedent in advocating for the 'sandboxing' of new financial technologies before their wider release.
"Lots of regulators that I talk to are trying to copy them, basically," he says. "They're having positive conversations with banks on what can and cannot be done."
It makes a change to how business was traditionally pursued. "When you went to regulators in the past, it was like going in front of a firing squad," Van der Ouderaa recalls. "A bank would say, 'I want to do this and I want to do that', and then they would look at you, say 'No', and tell you not to come back more than twice. Now, it's different. There's a fintech mentality among regulators."
Whether this momentum is sustainable remains to be seen. Even if banking executives and regulators can agree on the benefits that AI could potentially bring to the financial sector, it still requires labour, namely the coders and managers capable of crafting and integrating these kinds of analytical engines into a bank's systems. Increasingly, the kind of people HR departments want to hire are proving hard to find.
"There is a shortage of talent at the level of executing change," explains Van der Ouderaa. "All these people have to be new, and so banks are frantically looking for those types of people, because there aren't that many people that have already done that kind of work before."
Despite this, he is firmly convinced that the horse has bolted when it comes to AI's role in banking. Customers are beginning to see it in the way banks are rebranding themselves. "I see two models emerging," says Van der Ouderaa. "There will be banks that go purely digital, with no human beings, and then there will be the ones that still advocate a personal relationship. A lot of challenger banks will try and advocate that personal relationship, that 'quality-time' aspect."
Ironically, that proposition will rely on the progress digitisation is making in doing away with human input in repetitive back-office tasks. "It's all about getting the friction out of the system," he explains. "Getting rid of all the paperwork, of all the complex processes, all of that." And in the end, that will probably free up enough staff to permit what most customers wanted all along: to solve their problems by talking, in the end, to a human being.