Support Learning and Insight

It’s more important than ever to understand the challenges facing financial executives. Support the Financial Education & Research Foundation today.

Technology

The Strategic Case for RPA and Machine Learning in Finance, Part 1


by Chen Amit

Even after the machine-based computers came online, humans were still very much part of the equation, utilizing their skills to define the right theories and strategies.

Depending on your business, there are two types of automation/AI that a finance organization can start employing: machine learning and RPA (Robotic Process Automation). First, let’s get one thing straight—robots are not stealing our jobs. When NASA's Apollo program hired human computers to help decipher the math for the moon landing—so compellingly presented in the film "Hidden Figures"—today's machines were not readily available.

In fact, the math for the computational questions they needed to answer hadn't even been invented. Even after the machine-based computers came online, humans were still very much part of the equation, utilizing their skills to define the right theories and strategies. 

Software cannot assume total control if the process isn't defined. In both RPA and machine learning, the software requires a certain amount of predictability and rule-setting in order to “take over” a process. Only then, once the formula is defined, can applications be written to do the hard-computational work. 

Machines are infinitely better at doing the mundane, rigorous math and processing, which then frees humans to worry about the next strategic initiative. The definition of what is mundane is the real factor organizations need to wrestle with. Of course, you can get by without technology by hiring more people—that will always be true. The big question is, are those positions valuable to the business? Organizations that view headcount in a more optimal way don’t rely on hiring, especially if there is technology that can do the job better, faster, and cheaper.

Let’s look at a more apt example: piloting airplanes. Today’s airplanes are a sophisticated mix of measurement tools and corrective, automated systems that consistently manage the atmospheric and mechanical changes of flying an enormous aircraft. What’s the human-equivalent alternative? It’d be like having 20 full-time pilots on a flight instead of 2 to monitor and respond to each parameter or situation. The communication alone would be inefficient, never mind the resources required (e.g. paying and feeding 18 additional people). And let’s face it: the likelihood of error increases. The pilot’s job is to guide and respond to changes—it’s not worrying about every minute detail. It’s the same with finance technology. 

RPA is essentially code that has been written to act as the virtual user for existing software. In the good old days, they may have been called macros, but RPA adds more rule-based conditions and often runs on a schedule. The immediate benefit is that RPA improves often outdated legacy UI/UX situations that were not automated at their core, such as entrenched ERPs from a decade ago. In addition, RPA trains bots to traverse the software's application programmable interfaces (APIs). Ultimately, you still need software that is ready to be controlled. RPA in large-scale, legacy applications, under static processes and interfaces, can make up for the shortcomings of old designs and poorly executed processes. Time-consuming functions such as data cleansing that runs on the backend are ideal for RPA.

One RPA caveat is that if the interface changes—a parameter here, an option there—the “bot” needs to be reconfigured or redeveloped. Change management becomes critical at the upper and lower levels and more daunting as processes become more complex or entrenched. It’s not a problem if the process is stable, but if it’s still being flushed out, no bot will save you.

Machine learning is where much of that intelligence and logic is built into the software—for example, the ability to read a document (e.g. invoices, contracts, email request) and know what to do with it (digitize it, route it, archive it, etc.). More modern applications these days do this out of the box without having to employ RPA.

Machine learning is a must-have in more modern applications where the software continues to improve processes and can grow in its awareness and ability to serve. Machine learning should be able to pick up on workflows and, over time, surmise how to route information so that additional human intervention is not required. For example, an invoice from a consulting firm for the marketing department should always be routed to the CMO for approval and should mark that amount to the marketing budget line-item. The technical requirements for the software is to be able to read the invoice and understand the routing without an intermediary bookkeeper or AP clerk keying in and routing the invoice.

In both cases (RPA and machine learning), payback for automation and AI investments come with economies of scale. Because AI is tireless and precise, the larger the scope of tasks, the greater the value it is able to provide. 

Part 2 will look at how businesses can add AI into their real-world financial operations processes. 

Chen Amit is the CEO and Co-Founder of Tipalti.