The Lessons Robots Have Taught Banks

The General Data Protection Regulation (GDPR), which took effect at the end of May, brought a new regulatory headache—and a stiff workload—for banks across Europe. The potential increase in staff hours needed to deal with customer requests for data being held served as an added cost many would have dearly loved to avoid.  

One bank that found a solution to the conundrum was Nordea, which turned to robotic process automation (RPA). “It was quite difficult to estimate how many GDPR requests we were going to get,” says Ingrid Kristensen, the project manager for GDPR at Nordea. “I know the GDPR project [at Nordea] did some surveys with existing customers asking if they would be requesting access to their personal data. But it can be difficult for customers to answer that as well [knowing] that it will also be influenced by the media attention that this would get, and how other companies would handle it as well.” 

Nordea’s Robotics Centre of Excellence automated two aspects of the GDPR regulation: customers’ rights to access data, and their right to data portability. When a customer reaches out to the bank and requests an overview of their personal data that Nordea holds, that triggers a robot, of sorts, that finds the relevant applications where customer data is stored, collects the information, and then combines it into a template for an overview. Then a staff member in operations reviews and sends out that requested information to the customer.  

“In terms of time savings, when we were doing this manually before robotics was introduced, it took anywhere from one to three hours to process one customer. A robot can do this in a matter of minutes, so the actual manual labor that is left over here is just a couple of minutes,” says Kristensen. 

Teaching the Robot 

While RPA promises a lot of benefits for banks such as Nordea—including time savings and a clear audit trail—to make the most of the technology it is important to be aware of its limitations 

Nordea has been working with RPA since 2015 and has robotized almost 300 processes. One of the lessons the bank learned when planning for the GDPR project was that access rights are time-consuming and can cause delays. Kristensen says it is important to build a buffer for this into the schedule. 

Another lesson included getting the close involvement and commitment of key stakeholders and process experts early on. “To prevent scope creep, focus on developing a minimum viable product (MVP) first and then discuss additional nice-to-have features after the initial launch,” she says. 

Banks have started looking at RPA seriously over the last five years or so, with the last couple of years seeing a massive acceleration in development and deployment, according to Vinit Sahni, CEO of artificial intelligence (AI)-powered platform provider Arkera. Sahni, who was previously the head of the fixed income and foreign exchange (FX) sales divisions at Goldman Sachs, and had stints at Bank of America Merrill Lynch and Deutsche Bank, has been part of many committees over the years that have decided when people had to dismantle certain processes and bring in automation. 

“There has been a lot more seniority that has come on the technology side: CIOs, the whole digital strategy, incubators—if you notice all of this has led to bit of a cultural change to break down some of the barriers. I think RPA can only be brought about on a scale if you bring in the infrastructure and you create the right environment for it,” he says. 

Indeed, not all tales about RPA are quite that romantic. During the 2017 Sibos conference, held in Toronto, Matthew Davey, a managing director at Societe Generale Securities Services, spoke about being a “bit disillusioned” with the experience of RPA. 

“When we talk about AI, most of what we mean is machine learning, but we’ve also done a lot of work with robotics, with RPA recently,” he said at the event. “I have to say that we’ve been a bit disillusioned with that experience. When I talk to people internally, there’s been a lot of negative comment about RPA.”

Talking to Waters for this story, Davey says some of the early tools they used didn’t perform the way they thought they would. 

“There have certainly been some mixed experiences with it. And part of it is because it has the word ‘robotic’ in there—people picture a humanoid robot, and that is not helpful to RPA as that creates some unrealistic expectations from people that this robot will be incredibly clever and capable,” he tells Waters. 

SocGen uses RPA in operational processes such as reconciliation and generating reports. But as Davey notes, they are focused on relatively simple processes. “If you try and apply it to a complex process then that becomes very difficult,” he says. 

Davey says RPA may not be a good match in the bank’s valuation process for refunds, a very complex process with several hundred pathways. “When you look at how you might parameterize that with RPA then you end up spending a lot of time creating the parameters to try and capture that process,” he says. 

Then it is worth asking whether the process is too complex, says Davey, and whether there is an off-the-shelf package already available to do this from a vendor. For an invoicing process, for example, an invoicing package will probably get the job done without the need for investment in RPA. 

One of the obvious benefits for banks is the potential savings in terms of getting the RPA to do the work of human staff. However, there are also added costs in terms of staff who can understand the technology. At Nordea there is a team of controllers that works with monitoring the processes. 

“We see that as the number of processes we have over time is increasing, then the amount of development work that goes into maintaining these processes will also increase,” says Nordea’s Kristensen. “For instance, if an application is updated or changed or if we switch to a different application then every robot that would interact with that application has to be adjusted. There is definitely a maintenance cost but we try to factor that in when we do a business case analysis and a feasibility study for each new process that we want to robotize.”     

Another area to be careful about with RPA is that it’s a very specific tool. For example, at SocGen, part of one of its operational processes was defined in Italy, and when it was deployed in France it didn’t work properly. The reason for that was they were using an Italian keyboard in Italy and a French keyboard in France, and because the keys were in different places in the keyboard with different mappings, the RPA as parameterized didn’t work. 

“RPA is a highly specific tool,” Davey says. “And you need to think about those kinds of details to make sure that you define things down to a keystroke level because it is literally replacing what a human being will do on a machine.” 

It’s not all problems, of course—the use of RPA can also expose bad practices on occasion. Davey cites one bank that deployed RPA and upon looking at the statistics after a month or two of operation, found that it wasn’t working as they expected. When it looked into the details, it found that the people who were previously doing the job were not doing it completely correctly, whereas the RPA was. 

Davey says he believes that RPA has great potential going forward, especially when combined with machine-learning processes. He predicts a big increase in RPA adoption in the next few years, and thinks RPA will be at the front of the queue because of the cost-reduction potential, the return-on-investment (ROI) and being able to get a full audit trail of everything that is happening. 

Legacy Integration

Looking back over the past three decades, many of the same platforms Davey was working with years ago are still in existence today, he says. The challenge of legacy technology is probably one of the main things that hasn’t changed because of the volume of business that is processed on those platforms, he notes. 

Deploying RPA onto legacy systems can produce fairly immediate benefits, he says, but it can also make it harder to make changes to the underlying systems. “There is a bit of a double edge to it that makes it harder to make those changes. Because we talked about how they are so specific, if you start to change systems, then your RPA will fall over, so that is an important dynamic that people need to think about,” he says. 

Arkera’s Sahni mergers and acquisitions (M&A) present another challenge. As the M&A environment heats up, and even as banks try to break down siloes and bring together business units or acquire new lines of business, complexity is added into the system and processes can get entangled. 

However, Nordea’s Kristensen takes a bit of a different view. She maintains that the issue of legacy systems is conquerable. “We have had a very good experience working with the legacy systems in Nordea. Especially if they are an old mainframe system, then that is quite straight forward to combine with RPA,” she says. 

Much of the success—or failure—of projects also depends on the task they’re trying to complete, which may require radically different understandings of what RPA is and how it applies to a specific business process, rather than being a one-size-fits-all template.

According to Edward Sander, president of Arachnys, which provides RPA for know-your-customer (KYC) processes, understanding the linkage between RPA and machine learning is important from a financial-crime perspective. He says that what most RPA technology does is simply enable an organization to automate data collection and transposition into a data field, and that is usually within some type of template or data schema within an application. 

“That is really all they do,” says Sander. “It is a very mechanical activity. It does provide operational benefits and standardizations and consistencies. But it is a commodity benefit. I would venture to say that you are going to see something like a hype-cycle curve where eventually the benefits that RPA technology can offer a financial institution will reach a plateau. What is far more important and why machine-learning investments today are going to reap significant benefits is because they help to automate the actual investigative process, not just information collection and information assembly.” 

It is important to draw a distinction between RPA and machine learning. RPA is a form of AI, but it is vastly different from the discipline of what would be considered true machine learning. The processes that are best suited to RPA are those that involve structured data with the user looking for a single answer. On the other hand, machine learning is typically using large amounts of unstructured data and it is more about inference-based assessment, producing an answer with probability rather than concrete results. 

But he also says they work very well together when combined. “They are very powerful complementary technologies. And that is probably one of the biggest changes since my comments at Sibos is that we are seeing more of them working together—RPA tools with AI components in them,” says Davey. 

He gives the example of RPA helping with data collection for AI, in terms of collecting the data and getting it ready for the algorithm to process. Conversely, AI can also create data input for the RPA. It could be an AI image recognition for a birth certificate, a KYC check, or using natural language processing from a customer-service chatbot that picks up input for the RPA.

The challenge is to make sure they are properly integrated into an internal ecosystem so the technologies play well together. “Because often you are dealing with deploying multiple tools, it becomes a slightly more complicated deployment but the benefits can be a lot greater if you can get them working together,” he says. 

Nordea Life & Pensions Norway has built a robot using a combination of machine learning and traditional RPA to handle disability insurance claims. “They can get the handling time of these claims down from 75 days to a matter of minutes,” says Nordea’s Kristensen.  

The March of the Robots 

Interestingly, according to a recent survey of financial services professionals by Broadridge, 80 percent of respondents are at least assessing the value of AI, machine learning or RPA initiatives. 

“How do you create the right ecosystem in the bank that can absorb these technologies from the outside? You can’t create all these inside. It is going to be legacy systems, and there is a 15- or 20-year-old culture there. So the only way you can do it successfully is if you create the right ecosystem that encourages outside technologies,” says Arkera’s Sahni. 

Kristensen echoes this point, noting that it’s easy to get lost in siloes, hence why it’s so important to be holistic when developing a plan for the entire bank.

“We still have a long way to go and it has definitely been a learning journey,” says Kristensen. “I think what we have learned is that it doesn’t matter how good you are in a robotics team within an organization, or how sophisticated you are within that little team; you still need to spend a lot of effort in educating the rest of the organization. Because as one team within the organization we are not able to identify every opportunity in a bank, so we need the whole bank to work together here. That has taken some time and it is still an ongoing process to educate the bank on this.”