Is RPA really an AI process or much less?

Does robotic process automation eliminate mundane tasks or just perpetuate a marginally productive process that should be redesigned in the first place?

robot gear automation

Although it’s been around for a few years, Robotic Process Automation (RPA) is the new “killer app” that companies are concentrating on deploying -- and an area that IT staff is enamored with.

[ Related: 4 steps to RPA success ]

It’s touted as a way to take mundane, standard workflow-related tasks, fully automate them, and relieve workers of the tedium of manual processes (e.g., expense reports, invoice creations, HR related tasks, call center operations).

It’s essentially a way for computers to monitor a repetitive task, and then learn how to do it autonomously. Many have pointed to this technology as an AI process, but is it? And is it really worth the investment in time and expense, when true machine learning based AI capability is only a short time away? Finally, are you actually perpetuating a marginally productive or even inefficient process that should be drastically redesigned in the first place by installing at RPA solution?

[ Related: How to deploy RPA successfully ]

Many of the major enterprise-class back office apps (e.g., SAP, Microfocus, Infosys, BluePrism) have RPA functions implemented into their product sets and offerings (some through acquisitions), in keeping with the new market buzz.

Some have even tacked on the “AI” label to their marketing pitches to make the products sound more modern and high powered. But if you are implementing an RPA solution, even one labeled as AI powered, will it ultimately provide the best possible long term corporate benefits?

The trouble with RPA

The problem with RPA is that it automates a fixed process already in place. It’s good at automating redundant manual tasks that users employ (e.g., call center look ups) or back room batch processes (e.g., HR workflows), and eliminates a number of necessary manual user interventions that hinder productivity. Deployment is relatively straight forward, and RPA can run on a local machine like a PC, On-Prem on an internal server in the data center, or in the cloud as SaaS. This diversity makes it relatively attractive to implement for many companies.

RPA is actually good at automating workflow processes. But what it’s bad at is improving those processes. It mirrors what people are doing, instead of studying how the task could be improved and made more efficient. A current task that takes hundreds of steps and has been in place for years is probably not the most efficient way to accomplish that work process.

RPA will automate those hundreds of steps and eliminate the need for manual processing at each step, but will it improve corporate efficiency and freeing up valuable resources to do other more productive tasks? It does so by eliminating the need for human labor, but not by streamlining a bulky process.

AI on the other hand is good as evaluating processes and suggesting ways they can be improved. Through machine learning algorithms, it not only records the steps but looks at optional ways that could either change the workflow or eliminate steps entirely. The analysis that AI provides enhances rather than just mimics tasks. It’s a much more involved undertaking to make it work, but ultimately a much more valuable endeavor as an efficiency booster.

Mimic vs. discover – which is the right path to follow?

RPA can be considered “low hanging fruit” in that it is relatively easy to implement and provides a streamlined workflow that increases efficiency, often at the expense of people’s jobs. But what it does is mimic an existing process by learning what the process consists of (often by a recording of human steps).

What RPA does not do is to discover what the key bottlenecks and inefficiencies are in that work process and offer suggested improvements that could make the entire operation more streamlined, faster to results or better integrated with corporate systems. On the other hand, AI done right, with its machine learning capability can provide a real discovery process that can tell organizations about their process bottlenecks and inefficiencies, and even suggest new and improved processes.

In the long term, this is a much more important process enabler than a simple mimicking of current processes, although it’s harder to implement as it needs true learning functions built on algorithms developed to learn about rather than mimic the corporate process. Over the next 2-3 years you can expect this learning process to become much easier to implement than it is today as rapid advances in AI tools are positively impacting the ability to use AI in many enterprises.

While RPA is not always a bad choice, it should not be a final destination. Rather, it should be seen as an intermediate step in the process of making corporate processes better, more effective and more efficient. Ultimately, the goal should be to design systems that evaluate processes to improve data flow, limit redundancies and increase processing time with minimal impact to corporate business flows. As a result, you should expect an implemented RPA system to need upgrading/replacing in the not too distant future.

The bottom line on RPA

 I expect RPA to be a short-term advantage for some companies as they try to make operations more efficient. But in the long term, true AI powered machine learning that provides for a discovery process is a better solution, although harder to implement. Nevertheless, I believe that companies should be focused on the benefits of an AI approach rather than choosing an interim solution that will be outmoded in a few years, or at least be aware that the implementation of an RPA solution is an interim step and not a final commitment.