The Hard, Human Lessons of RPA

Asset managers and buy-side firms are exploring the potential of robotic processing automation (RPA) to slash costs, digitize functions and remove inefficiencies, but roadblocks to the technology’s adoption and implementation are teaching some hard lessons along the way. 

Firms should be wary of using RPA to solve complex problems because of the time, cost and effort required to build and implement the technology, according to Alanh Husson, chief transformation officer at Unigestion. During a panel discussion on RPA at The Summit for Asset Management (TSAM) in London on February 28, Husson said firms should rank processes based on their complexity before selecting use cases, adding that his experience suggests the secret is to gradually automate workflows, starting with the simplest tasks.

“The idea was not to tackle the most complex processes at the beginning; it was to try to do something light,” he said.

During the same panel discussion, Ulrik Modigh, head of product and operations at Nordea Asset Management, described his firm’s experiences with RPA as delivering “limited success.” The Nordic heavyweight applied the technology to a number of use cases but reaped little return compared to the cost of implementing the pilots and fixing technical issues.

“I think the gains that we saw were too small, and there was also a lot of investment that we had to make to keep the process up and running,” said Modigh.

The Learning Curve: Use Cases

Nordea Asset Management has used RPA principles and structures to automate manual processes around the management of equities. During the TSAM session, Modigh explained that the project took too long to implement and was shelved after a few months, adding that the firm later resolved efficiency issues using other, undisclosed methods.

“The lessons learned were [about] selecting the right processes. I also think it’s a matter of what kind of business challenges you are facing going forward,” he said.

Like most institutions, Nordea has spent the past decade looking at how to resolve legacy problems through the automation of manually led, low-touch processes. However, the asset manager has derived little benefit from its selected applications, partly as a result of the complexity involved in programming the technology to carry out intricate tasks and the requirement to constantly update the robot to keep pace with the evolving nature of the industry.

“There are daily changes in new products, new clients and new solutions. In this space, I have a hard time seeing robotics as a strong solution to support our processes in this type of [evolving] environment,” Modigh said.

More than midway through the panel session, moderator Glenn Murphy, COO at Stonehage Fleming Investment Management, asked members of the audience from the asset management community to share their experiences with the technology. One attendee explained how his firm’s first pilot was shelved due to the complexity of the function it set out to automate. In its second attempt, the firm selected a simpler use case—where the process could be easily mapped out—to program the technology, but later encountered a second major roadblock when deploying the robot. He outlined how he had scripted the code for the robot on a Qwerty keyboard in Germany, but when it was deployed in France the technology crashed because of the design of the French Azerty keyboard.

“It’s simple things like this that can trip you up if you are not careful,” he said.

Unigestion has applied RPA technology to its outsourced Simcorp client relationship management system to identify inflows and outflows of clients and investors. Though the application can involve some complex actions surrounding reference data, Husson explained that the technology is being introduced in phases and that to date the platform has reached 75% RPA. The Geneva-headquartered asset manager has learned that using a step-by-step approach is crucial to a successful RPA program, as overloading a system with robots can result in programming errors or cause the technology to crash.

All three panelists agreed that RPA isn’t a good solution when it comes to automating core business functions, which should be managed by specialists, though it can be valuable in supporting repetitive tasks, managing data not connected to primary business lines, communications, and less complex inefficiencies. Christian Kemper, head of operations for Europe and China at Deutsche Bank Wealth Management, said his firm has seen the consequences of stripping back on critical resources and has learned the importance of not wiping out the human factor and of retaining the knowledge within the firm to run RPA systems.

“One of the mistakes we made at the beginning was that we let too many people go, and we had to backpedal a little. That’s coming from me,” he said. “The two major points are: don’t use it to replace your core banking system, and don’t underestimate the fragility of robotics and the resources there to understand how they work.”

Kemper explained that robotics are programmed to mimic humans, and implementing drastic changes such as cutting down on staff can disrupt the technology. Basic RPA is programmed to function under a set of principles, which means altering the environment or source inputs can lead to errors or technical issues. Unigestion’s Husson echoed this point. He said a tactical approach is necessary in the adoption of robotics—that is, robots can be used to automate some functions but should not be used to replace the skills of an entire team responsible for the process.

“I would recommend keeping the knowledge in the company. I think that is the key. The knowledge of the detail and the knowledge of the process [are critical], because if the robot goes down you still need to run the shop. If you don’t have the knowledge, it’s game over,” Husson said.

Initial Drivers

The main drivers of adopting RPA are the removal of inefficiencies and cost-cutting, but there is no such thing as one-size-fits-all in the world of fintech. The panelists explained that RPA implementation should always depend on an individual firm’s IT infrastructure and business processes.

Husson explained that robotics can often target productivity issues, such as siloed systems or the speed of processing tasks, where a machine can compute large sums of data faster than a human. “[RPA] has the ability to get rid of the key macro global problem that exists [relating to Microsoft Excel], and a robot will be able to fill an Excel spreadsheet with 100,000 lines in a second, without mistakes,” he said. Such manually led processes are at higher risk of operational errors or compliance failures.

Deutsche’s Kemper outlined how his firm analyzed the potential of RPA to cut costs by decreasing the number of repetitive tasks, reducing operational risk and automating functions unrelated to its core business lines. The bank’s RPA pilot targeted processes that used an external data source or imported feeds, including digitized data formats such as Excel sheets. Using a robot, the external data is reconciled for Deutsche’s core internal functions. Kemper explained how other departments in the firm are now leveraging the robot to automate tasks across human resources, the front office and finance. He recalled that when the IT team originally built the robot no-one was interested in using it, but over the years the technology has gradually changed how teams operate and approach tasks. Deutsche Bank has built its own proprietary robot and has also worked with third-party provider Blue Prism on certain projects.

When building and implementing an RPA system, Deutsche’s development teams are forced to document processes step by step, information which is then used to program the technology. The immediate benefit is that firms can identify inefficiencies and examine ways of speeding up and improving the workflow.

Nordea Asset Management’s core driver was curiosity. Modigh explained that the firm decided to experiment with RPA with the objective of testing its ability to cut out inefficiencies. But after a few months with little to show for it, Nordea ended all running projects and has no immediate intention to revisit the technology any time soon. Modigh explained how the asset manager leverages an all-to-all platform in which all of its processes operate in one system. He said introducing RPA on top of the consolidated platform proved overly complex, especially in areas requiring flexibility to scale and change over time.

“Our fear at the time was that if we create a surface of robotics on top of our platform, the day [we] want to make changes to the business or to the processes, we would have to go back to all the robotics [systems] and make sure they are all aligned and able to communicate and perform the right way. I see how it can help a fragmented infrastructure, or how it can help an [IT] infrastructure. It is a fantastic tool to get short-term gains, but it was curiosity about the technology that was the main driver [for us],” he said.

The Future of RPA

Today, basic RPA systems operate based on a set of rules or principles. However, as technology advances, more and more robots are becoming increasingly intelligent—blurring the lines between subsets of artificial intelligence such as machine learning and natural language processing.

“I think the next big change—and I don’t know how we are going to manage that in the future—is that currently the advantage of RPA is that it can be built in your garage. You don’t need a lot of skilled resources and it is pretty straightforward. It’s logical thinking [computer programming]. Once you go into artificial intelligence, you will need engineers, and, again, we will go back to developing the software. So it is going to be interesting to see how that moves from one to another,” said Deutsche’s Kemper.

One of the biggest challenges is acquiring the skillset or budget to throw at building a single robot, let alone a batch of smart robots. The panel discussed the potential of buying ready-made robots with AI capabilities from third-party providers, especially for use cases such as document management. Kemper said it is only a matter of time before the market moves in that direction and realizes the emerging demand. The introduction of ready-made intelligent robots could potentially also lower the barrier for entry for smaller investment firms unable to build the technology internally, shifting the competitive edge in their favor.

For institutional firms looking to automate processes, another developing concern is that the industry is becoming increasingly dependent on robotics and black-box solutions. As all technology is subject to glitches and error, the panel’s consensus was that effective governance controls and audits need to be put in place now in order to mitigate risks. Unigestion’s Husson explained that reliable governance should begin with the same strategies for effective implementation—ensuring RPA is used for non-critical processes and maintaining a core layer of human expertise for managing bottom-line operations.

“We need to build the expertise internally, this is really key. People knowing the business in one hand and the tech in another hand, that’s what’s important,” he added.