Down-to-earth offshoots of artificial intelligence are increasingly accessible for digital transformation work. These projects tapped into machine learning – with existing talent.
Why Robotic Process Automation (RPA) projects fail: 4 factors
No one wants to see their Robotic Process Automation project fail. Check out when and where RPA can go wrong – and learn from common mistakes
Robotic process automation (RPA) can be a great fit for tedious and repetitive processes, but it won’t fix a process you don’t fully understand or that is otherwise fundamentally broken. That’s a basic – and frequent – misstep that commonly leads to an RPA project not achieving its intended goals.
[ Need a primer? Read also: How to explain Robotic Process Automation (RPA) in plain English. ]
“People are trying to apply RPA before they really know how their processes work,” says Antony Edwards, COO at Eggplant. “That tends to fail as they are constantly discovering new exceptions and variants.”
Indeed, “exceptions and variants” are not harbingers of RPA success. When evaluating candidate processes for RPA, most experts will recommend looking for predictable, rules-based processes that repeat regularly. Processes that change or otherwise don’t follow a predictable path, on the other hand, tend to be a bad match for RPA.
[ Want a more agile culture? Get the digital transformation eBook: Teaching an elephant to dance. ]
Why RPA projects fail
“RPA tends to fail in two scenarios: Either the process being automated is not as robotic as initially thought, or the resulting automation is run in an environment that is much more dynamic than previously identified,” says Scott Totman, VP of engineering at DivvyCloud. “In either case, the tooling requires much more maintenance and ongoing development.”
More maintenance and ongoing development work is not the kind of metric you’d boast about. In fact, it’s sort of antithetical to why you’d consider RPA in the first place: RPA aims to make processes more efficient and reduce the potential for human error.
When and where RPA fails
Let’s take a closer look at four fundamental reasons why RPA can fail.
1. The process is more dynamic than you realize
This is probably the biggest pitfall: It essentially means you’ve automated the wrong process.
“If the environment in which the automation runs is more dynamic than expected, then the RPA tooling must incur additional complexity to make sure it can continue to operate in an ever-changing environment and still produce the right outcomes,” Totman says.
Again, that kind of defeats the purpose. If a process requires decision-making on a case-by-case basis, you still want humans closely involved.
“Tasks involving creative thinking, brainstorming, interacting with the physical world (like pulling papers from a filing cabinet) are better handled by people,” says Aaron Bultman, director of product at Nintex. He notes that this doesn’t mean you can’t automate any pieces of these processes – a workflow automation tool can help handle the repetitive steps of a process that also requires human decision-making and skill. And the workflow automation tool and RPA can work in concert.
“The idea is to only leave the human work for the human to perform,” Bultman says. “Nobody likes working on low-value, thoughtless, or repetitive work.”
[ Is that process a good fit for RPA? Learn from the experts: How to identify Robotic Process Automation (RPA) opportunities. ]
2. The target UI changes, but your RPA bot doesn’t get the memo
RPA is good at following instructions; it’s not good at learning on its own or responding to unexpected events, a distinction some people don’t make since they lump RPA together with AI. (As it stands today, RPA is typically “dumb,” though we’re likely to see the growth of use cases where it’s deployed in concert with cognitive technologies.)
“We’ve seen an RPA bot break when it encounters scenarios which it was not trained [for] or instructed to manage,” says Vishnu KC, senior software analyst lead at ClaySys Technologies. “One of the most common examples is changes in the target user interface. To put this into perspective, if an RPA bot has been set up to go to a web page and click on the top-left corner to reach a sign-up page, it can do that as long as the bot is able to find that particular button in the top-left corner of the page.”
A problem arises if you move that sign-up button to the middle of the page – or anywhere other than the top-left corner, for that matter. The RPA sequence will halt because the bot will keep looking for it in the top left. KC notes an irony here: Technically speaking, we’d be wrong to call this an RPA failure; it’s really a human error.
“The bot is working exactly the way it was told to,” KC says. “It just didn’t anticipate such a roadblock.”
This tends to be a bigger issue than some people realize when they first deploy RPA.
“The typical problem with RPA is the rigidity of the process and the dependency [and] sensitivity of the applications or systems that are being automated,” says Muddu Sudhakar, CEO at Aisera. “The reason behind that is the fact that RPA is typically leveraging screen-scraping technologies, with the obvious problems [that occur] when the UI screens change.”