Robotic Process Automation (RPA) and Artificial Intelligence (AI) are often used interchangeably. This bucketing is unfortunate, since they mean very different things.
RPA started as the application of technology to help employees replace their daily labors with those undertaken by robots. A clear example would be the appearance of a robotic arm in the General Motors assembly line in 1961. Humans were tiredly (and slowly) screwing side-view mirrors onto cars, and sagaciously saw that robots would do a better job. Machines were therefore designed to screw side-view mirrors onto cars again and again, without getting tired or complaining.
Note that RPA systems are “dumb.” Without precise directions from human designers about what to do, the robotic arms were just collections of metal. And once they received this explicit direction, they could only do one thing -- screw side-view mirrors onto cars -- and could never hope to learn how to do anything else.
Of course RPA evolution didn’t stop at the automotive assembly line; like many progressive technologies, it expanded into software. But the core remains: RPA software programs are similarly dumb, programmed by humans to do one task without any need or hope of learning beyond it.
For example, let’s say AppZen used RPA instead of AI. That would mean we programmed our software to have rules like “Flag any alcohol receipt greater than $50.” While of course that rule, enforced over and over again, does have some usefulness, it does not take into the account the unpredictability of human experience, and it certainly doesn’t have any desire to try.
An example of RPA used in the expense report space is Workfusion. This macro on the desktop allows users to set up rules to flag misconduct most often when porting data from one software system to another. So if the user is sharing data from the company’s T&E software to its reimbursement software, she can, using the example again, set rules that flag every alcohol receipt over $50.
The problem is that RPA is simply not “intelligent;” the system cannot get smarter over time. It’s set and forgotten; no smarter 2 years after the day it was designed. A Fortune 50 bank tried RPA for its expense report audit and struggled mightily. The system simply lacked the decision-making capabilities necessary for the complex, organic world of T&E.
AI’s history dates back to around the same time as RPA’s -- the late 1950’s. It was then that Marvin Minsky and John McCarthy described AI as any activity performed by a program or a machine that, if a human carried out the same task, we would say the human had to apply intelligence to accomplish it.
Of course AI has also grown into the software world, where it has retained the kernel of getting smarter over time.
In the example of expense reports, an AI system like AppZen’s can learn that, say, Nike’s market cap has grown by 35% in a given year therefore it could be permissible to expense more than $50 on alcohol to network with company executives. Or it could it learn that the British Pound is particularly strong against dollar in a given month and that $50 at the pub with the team after the conference ain’t what it used to be.
Clearly, both RPA and AI have trailblazing histories and vital present-day applications. RPA specializes in the black-and-white world of rules and repetition. AI, on the other hand, is designed to grow more intelligent over time, learning from real-world factors outside the proscribed world of the original task. While both types of systems of have their merits, we here at AppZen believe that AI provides the best service for expense report automation, and we’re committed to help make our platform more intelligent every day.