Teaching software bots to take over repetitive manual tasks is the magical promise of Robotic process automation (RPA). Only it’s not magic.
The problem with RPA is that bots break, and if you start putting them in charge of mission-critical tasks, they could break your business.
“The value is when these bots are up and running, and they keep running,” says Mika Vainio-Mattila, partner and co-founder at Digital Workforce, an RPA services firm. “These solutions require, by nature, more maintenance than traditional IT solutions.”
[ Want a primer? Read also: How to explain Robotic Process Automation (RPA) in plain English. ]
That is not to say the promise of RPA labor savings and efficiencies is a mirage, but it does mean planning for bot maintenance as well as bot development.
RPA tools can record human interactions with applications and play them back, say with the goal of fetching data from one application and recording it in another. This is a way of achieving integration without APIs and the automation of repetitive manual tasks. For example, an RPA bot might look up data from two different reports, consolidate it into a single spreadsheet, and email it to one or more people. RPA bots can be created with little or no coding effort, meaning that bots can be created by business users with little or no involvement from IT.
“You can get results fairly quickly. Then you get into more complex automations, and you need to have more solid governance in place,” says Stephan Blasilli, who leads business transformation initiatives at a global renewable energy company.
“If somebody thinks you can just record a process and go home, it doesn’t quite work that way,” Vainio-Mattila says.
Here’s what you need to know about where things go wrong.
1. Why RPA bots break
Whether an RPA initiative is big or small, ambitious or modest, it should include a plan for when automations break, as they inevitably will, Blasilli says. That means having people available to do reactive maintenance (scrambling to fix a broken bot), but also trying to be proactive.
Blasilli points out that although “screen scraping” techniques for defining how a bot should interact with an application user interface are very popular – and relatively easy, particularly for nontechnical business users – their weakness is they map the layout of the screen. “If something changes, the bot will have a hard time identifying which of the fields to interact with,” he says.
Basic RPA bot creation techniques like screen scraping are also what is easiest for business users to accomplish without help from IT – one reason maybe IT shouldn’t be out of the picture, or at least should offer training on designing bots for resiliency.
RPA software vendors are working on making bot breakdowns less frequent, which is why there is so much talk about combining RPA with artificial intelligence. If bots can “understand” the tasks they are assigned, they should be less easily confused. But this is a work in progress.
2. RPA bot design makes a difference
“There are two sides to the solution – and number one is to build better bots,” says Vainio-Mattila. “It sounds obvious, but it’s not. Number two is to think through very carefully what is your model for operation and maintenance – and, I would add, improvement.”
One of the services Digital Workforce provides is Run Management – keeping bots running. This is an optional add-on to the firm’s cloud hosting of RPA platforms and can also be performed on-premises over a VPN connection. Clients can contract for the Service Level Agreement appropriate to the importance of their bots (is it okay for a bot that fails on Friday to be fixed by Monday, or does bot failure mean the client immediately starts losing business?).
One requirement: “We insist on setting the rules for what bots need to comply with, so that there is a quality gate at the front of the Run Management shop,” Vainio-Mattila says. “Part of that is also mentoring and guiding the development team on how to design good bots.”
Organizations managing their own bot maintenance would be wise to do the same thing.
You design better bots by anticipating the ways they might break. For example, rather than mapping an automation to the exact screen layout of an application, you can have the bot search for the field description to find the appropriate data entry blank or drop-down list, Blasilli says, or use application hotkeys (if available).
3. Well-designed RPA bots still break: How to prepare
Mike Tyson famously said, “Everyone has a plan until they get punched in the mouth.”
Planning and design can lower the odds of RPA bot breakage, but not to zero. The unexpected will happen. Maybe you designed your bot to search for a field label in a user interface, rather than depending on the screen layout, but the field label changed in the latest release. Or your bot will choke on some unanticipated data input.
Whatever. It’s broken, and now you must fix it. If your RPA program has gotten more ambitious, to the point where it’s managing business-critical processes, you must fix it fast.
“You need to have protocols or processes in place. Otherwise, you will have unsatisfied customers,” Blasilli says. Reactive maintenance, by definition, is about dealing with the unexpected, but you can plan to have staff available to deal with these issues and provide them with troubleshooting documents to guide them in finding and fixing problems.
Also needed: RPA expectations management. Maybe zero-maintenance, AI-powered RPA is just around the corner. Until then, a certain amount of expectations management is in order.
“It’s the responsibility of the departments who manage emerging digital technology to educate the organization about what RPA is capable of and not capable of, as well as the requirements and best practices to have in place,” Blasilli says. “On the other hand, it’s not a very invasive technology.” With an iterative approach, he adds, you can accomplish a lot.
[ Learn the do’s and don’ts of cloud migration: Get the free eBook, Hybrid Cloud for Dummies. ]
Great article. Good to watch for maintenance regarding bots. myinvenio is focussed on process mining before installing bots for RPA. Thanks.