Exclusive: ‘Moving beyond hype’ – Ren Zhang, BMO and Swapna Malekar, RBC and Kai Schrimpf, Credit Suisse in “The Paytech Magazine”
It seems that there’s a wealth of new use cases for AI within the payments landscape. We brought together three industry experts to consider successful applications, the hype cycle and how far AI can replace traditional banking practices The payments industry is moving to the forefront of pioneering use cases for artificial intelligence (AI) technology as it seeks to improve efficiency and boost security.
The plethora of digitised payment technologies now available has produced a huge amount of payments data, which advanced machine learning models
can use to generate accurate financial predictions and present customers with more personalised services. The growth in AI and payment technology is symbiotic. AI significantly improves payment ecosystems and can also help businesses using new platforms to fight fraud.
Having already reported an 80 per cent decline in card-based fraud since the introduction of smart payment cards under the Europay, Mastercard and Visa (EMV) standard, Visa in August 2020, announced the launch of Visa Smarter Stand-In Processing (Smarter STIP), which leverages AI to help Visa act as a backup to approve or decline transactions on an issuer’s behalf. Decisioning is made through real time, intelligent analysis, based on card numbers, card type, whether transactions are face-to-face, and a host of other data points.
Banks are deploying AI-based systems in record numbers. The newly-released Preventing Financial Crime Playbook, a collaboration between PYMNTS and NICE Actimize, highlights that more than $217billion has been spent on AI applications for middle-office use cases such as fraud prevention and risk assessment. Some 80 per cent of experts report that AI reduces payments fraud and almost 64 per cent say it is vital to stopping fraud in action.
AI is also being deployed in the back office. In September 2019 at Sibos in London, for example, SmartStream unveiled SmartStream AIR, a Cloud-native, AI-based reconciliations platform that it created in its Vienna Innovation Lab. AIR makes reconciliations as simple as possible, drawing on AI and machine learning technologies to ensure the user just has to upload the right files to get an instant, accurate result.
So, with AI now seemingly touching every part of the payments journey, has it finally moved beyond the hype? We asked three experts for their views: Ren Zhang, chief data scientist for BMO; Swapna Malekar, AI product lead at RBC; and Kai Schrimpf, head of global transactions and monitoring with Credit Suisse.
The Payments Magazine: Can we start with each of you describing your day-to-day relationship with AI?
Ren Zhang (RZ): I’m enterprise chief data scientist in charge of BMO’s AI Centre of Excellence, and enterprise chief data and analytics officer. For us, AI is a day-to-day function now, as we explore ways to add value to the business and develop reusable AI capabilities to scale adoption.
Swapna Malekar (SM): I’m product manager at RBC. We started really small, with some statistical analysis, then transitioned to AI as we enhanced our capabilities. I now develop products and solutions using machine learning, AI and any other technology that seems pertinent!
Kai Schrimpf (KS): I consider myself an AI consumer rather than an AI creator in my role trying to solve the global financial market problems of transaction monitoring, money laundering and terrorist financing. I talk a lot to the data scientist community about solutions that will hopefully make things better. It’s a problem that’s now 50 years old and we still have big issues. I’m presented with new AI solutions, some of which work very well, some of which don’t, on a daily basis.
TPM: What do you think has been the most successful application of AI in the financial space – something that has had a major impact on, say, customer experience or fraud prevention?
RZ: On the customer experience side, it’s personalisation and recommendations. Now, customers feel that banks are talking to them in a much more relevant way, knowing them better, based on their historic transactions and feedback, so the customer experience is more streamlined and relevant.
The other is AI’s impact on the cyber and fraud control area. We use the mass of data associated with a customer’s digital footprint to detect early signs of a potential cyberattack and fraud.
As a result, for the customer, there will be less disruption to their experience, and less financial impact, or need to raise a dispute.
KS: For compliance, the technology is now there to process massive amounts of data, learn from it and apply logic that we couldn’t five years ago. In my area, the transaction monitoring space, the biggest impact is in predicting alert outcomes. If you have an alert, your machine decides there is money-laundering risk in this pattern of behaviour, and the analyst takes a look and checks it out. If it’s risky, we tag this specific pattern of behaviour, the inflows and outflows, the products used, the industries used, all of the stuff that is inherent to the transaction.
SM: What’s really worked in all the products and services I’ve led, or seen in the industry, is data-driven decision-making. You have a humongous amount of information in your hands, demographic information about your clients and transaction and financial information. Merging all of that together into one contextual data set is important, as is bringing out the insights to make accurate and relevant decisions.
This has been very successful in banks and fintechs because AI can understand the creditworthiness of a client, based on contextual information, to decide whether they are eligible for a loan or other service. Or, in the age of COVID-19, you can provide relief for certain clients by understanding their needs.
Even though the algorithm might say one thing, if you merge all that data and the human insight together, you can provide a better experience and decision-making for customers.
TPM: Looking at it from another perspective, what AI hype have you yet to see delivered in the industry?
KS: In my area, the biggest hype is around false positives. The large banks are running a 95-99 per cent false positive rate and there are a lot of solutions out there that promise to bring that down to 40 per cent. Now, I’m quite certain that good AI could do that, but in large corporations, you’re running up against legacy data quality issues that will prevent that AI from functioning the way it should. And, when I talk to the really smart data scientists in my firm, with the triple PhDs, they tell me ‘I could build something that is better, but it’s going to be hard for you to understand how it works’. I always have to say ‘then I can’t use it, because I’m going to have to explain it to a regulator’.
SM: In the last few years, I’ve started hearing about the ethical use of AI, which is very pertinent to what we’re trying to do and the kinds of solutions we’re trying to put out into the market. But the whole ‘explainability’ challenge or even, for example, bias in AI, all those challenges are being talked about a lot and they haven’t been solved yet.
TPM: When it comes to developing AI solutions, what’s your attitude and approach to working with third parties?
SM: For a few products and services that we want to develop, we partner with third-party vendors, but, at the same time, we have an in-house research centre. I’ve seen this hybrid approach in most companies; lots of organisations are developing these AI, machine learning and data science capabilities in-house. Banks are trying to be technology companies, and companies like Facebook are trying to be banks. It’s an interesting time.
RZ: Definitely a hybrid approach. But I would say you have to be careful, in terms of purpose, when working with a third party. There are a couple of obvious areas where it’s very effective. A specialised area that they are very good at, for example, or something brand new that we are trying out. Or when the third party provides a certain capability that allows you to improve your process by doing things more effectively. But its important to also have an in-house capability to ensure you don’t lose your competitive edge.
TPM: What are the typical barriers to adoption you’re facing?
RZ: Unfortunately, there are many. On the technology side, the main barrier is infrastructure; the basic infrastructure that’s needed to enable AI is not there. As a result, many applications we’re talking about become very costly to implement. That becomes a showstopper. On the people side, it depends on the bank. And, even within the same bank, it depends on the group. Some people are ready, they know what they want to do, and some people are sceptical. In some groups, adoption is low due to fear of failure.
So, how do we address these issues? On the infrastructure side, you have to have some enterprise investment, to have the foundational data problem addressed, making it easily available, making it cheaper for individual teams to apply AI, rather than investing in the entire infrastructure to adopt that. For people, there has to be very good education. There has to be an AI and machine learning team to sit with them and address a business problem, working together.
SM: First and foremost, data. Data is such an issue, especially in a large organisation, because you have so many business groups and so many points of contact with customers. A customer can be a wealth management client, a retail client or a mortgage client, and, usually, there is no single source of truth about a customer. So, just to acquire and integrate that data, and merge it into one single source, normalise it and clean it, is a massive operation.
On top of that, there is stakeholder expectation, because AI is this buzzword, and people expect to solve so many different things with it. My job is to educate my stakeholders, and help them to understand the genuine use cases.
TPM: A final question – how far will AI go in terms of replacing traditional banking practices and personnel? Are we all going to be replaced by robot overlords?
KS: Will we soon be in a world of Blade Runner or Terminator? I don’t think so. It’ll be a long time before Skynet comes! But the technology is advancing rapidly.
In New York, 10 years ago, there were toll booths every couple of miles. You had to stop and pay a toll booth assistant $2 to drive on the road. Nowadays, a picture is taken of my car and the money gets automatically deducted from my account. That’s a development where a specific job description has been replaced by technology. Will all jobs go that way? I don’t think so.
Ultimately, human beings are herd animals and there will always be people who will pay a premium for human interaction. Maybe a couple of generations from now, things will be different, but I don’t see the replacement of human intuition, human ethics or the human mind on the horizon yet. You can enhance technology, as vastly improved computing power has done, with AI the cherry on top. But I don’t currently don’t see it replacing a vast majority of the jobs that humans do.
RZ: Will it affect the existing workforce? Absolutely. It will impact hugely. But does that mean people are going to lose their jobs? I think what will emerge is different types of skillsets, with people getting to do more value-added work. For example, on the fraud investigation side before, you’d have to manually review so many cases that you probably had lots you would never get a chance to check. Now, with AI doing the first screenings, your capacity is going to be bigger and the outcome is going to be much better.
- Chris Baker Named Chief Strategic Officer at POSaBIT Read more
- Fintech Job Cuts: An Analysis of the Impact on the Industry Read more
- Wolters Kluwer’s CCH® Tagetik expert solution recognized by Nucleus Research Read more
- Court of Appeal Rules in First Case Over Developers’ Duty of Care to Crypto Asset Owners Read more
- Vancity’s Visa credit cards are counting carbon footprints with ecolytiq Read more