Robot technology is often associated with dangerous, dirty or dull work. Robotic process automation (RPA) takes that concept into the digital world.
As a software tool, RPA sets out to free human operators from repetitive tasks, and to boost efficiency and accuracy. It is especially useful where data needs to move between disconnected applications.
But robotic process automation is evolving.
Bringing artificial intelligence (AI) into RPA systems promises to make them more flexible, more resilient and smarter. An RPA system can extract totals from an invoice, and even send them on to a human supervisor if they fall within a certain range. AI can go further by reacting to unexpected changes or detecting trends. This helps to minimise the workload passed on to humans and detects anomalies or fraud.
Software suppliers foresee systems that can learn how to streamline or improve business processes on the fly or can adapt to outages. Ultimately, this could lead to enterprise resource planning (ERP) or customer relationship management (CRM) systems that operate autonomously.
What is RPA?
RPA first gained ground in industries such as financial services, telecoms, government and healthcare – businesses with rules-based processes that involve rekeying data. More recently, the technology has gained ground in manufacturing and logistics.
RPA also lends itself to tasks that are common across businesses – processing invoices, setting up suppliers, human resources and some areas of customer service. The ideal task for RPA is predictable and repetitive.
As a technology, RPA is not new – it shares its origins with rules engines and other basic business and workflow automation tools. However, RPA developed to solve a specific problem: systems that were not, and could not easily be, connected.
“RPA started over 10 years ago as a way to automate processes more easily via the user interface, rather than having to build a system-level integration or have a person rekey digitised data between systems,” says Cathy Tornbohm, a distinguished vice-president analyst at research firm Gartner. “RPA tools are the ‘Babel fish’ of the technology world as they can interact with any type of system.”
RPA in action
RPA began with a simple premise – to automate labour-intensive, clerical tasks. “RPA started out by mimicking computer users’ desktop activities, such as mouse clicks and keystrokes,” says Bernhard Schaffrik, a principal analyst at Forrester. “It has gone beyond simple mimicking, and is now focused on automating what computer users usually do manually.”
CIOs typically turn to RPA as an alternative to, or part of, business process automation (BPA), which relies on connections between applications. Where these connections are not possible, or are too expensive to create, RPA is an alternative to human data entry.
“The primary problem businesses are trying to solve when they invest in automation and AI is that they’re collecting data far faster than they can analyse it,” says Chris Huff, chief strategy officer at Kofax.
“Certain business processes within the ‘pool of data’ are more fit for automation than others. Primarily, business processes that include structured and standardised data and are driven by rules-based processes operating in a stable environment are the strongest candidates for RPA.”
RPA can also team up with optical character recognition or even speech-to-text systems to recognise data in scans or audio files, and use that data to move on to the next step. Used this way, RPA can speed up business processes or shorten time to market.
And, as a computer-based system, it is less prone to errors than human operators are.
“RPA has a lower error rate than a comparable human,” says Sebastian Schroetel, head of intelligent RPA at software supplier SAP. “We all have a bad day sometimes, when we get to work and make errors. That is reduced with machine-based automation. They do their thing and don’t have a bad day.”
Translation services and consulting firm RWS uses RPA internally, and on client projects. “We use it for certain translation projects for customers which are quite complex, such as where documents are using content from previous projects,” says George Bara, the company’s vice-president for partnerships.
“We build an RPA system to service all those interactions or to transform data and move it within multiple systems. One of the most common use cases in corporations that are using RPA is to take processes and systems and bind them together, only relying on human interaction when there is a need.”
Internally, RWS uses RPA in invoice processing. The system can extract data from an invoice for, say, its German branch, extract the relevant information and translate it into English – without having to translate the whole document.
“It’s about productivity,” says Bara. “It’s minimising time spent on non-productive repetitive tasks. Why would I spend time doing tasks that a robot can do on my behalf? Why have 100 people reading emails from customers or have a robot push them to humans?”
RPA – limitations and prospects
So, RPA is efficient and effective – but it does have its limitations.
“RPA is essential for BPA, but it doesn’t consider how IT systems are connected through data flows and how these IT systems compute and deliver results behind those scenes, which are managed and triggered by users through data entries or mouse clicks,” says Forrester’s Schaffrik.
This is leading users of RPA, and suppliers, to try to make systems smarter. One advantage of RPA is that it is data- and application-independent. This allows it to work hand-in-hand with machine learning and AI tools, as well as human operators.
“Because an RPA tool can take data from any feed, you can use other ‘sister’ tools to be a complex rules engine, or read a piece of paper and structure it or structure an email so the RPA tool can complete the prescribed next best actions,” says Gartner’s Tornbohm.
Jack Watts, NetApp
But as a rules-based, deterministic system, RPA can’t make decisions on its own. For that, it needs to team up with machine learning or artificial intelligence. It also relies on structured data. AI offers the potential to replace human interpretation of unstructured data with an automated process.
“Once a machine is stood up, it is deterministic – it does the same automation time and time again,” says SAP’s Schroetel.
But in the future, RPA will go “broader and deeper” into business processes, he adds. This could lead to intelligent, or even autonomous enterprise systems, such as ERP. “We may see models that can take intelligent decisions and, step by step, move towards robots that don’t just build cars, but more intelligent ones that run financial processes,” says Schroetel.
This will draw on developments in both AI and supporting technologies. “Across industry, RPA is being leveraged to become more intelligent,” says Jack Watts, head of AI at NetApp. “Computer vision, for example, is being leveraged in a greater way.
“In the past, this would have been used to recognise characters and numbers on a cheque being paid into a bank. Today, entire invoices, purchase orders and even forms of identity are being ingested into ERP systems.”
The whole invoice-to-payment process can be automated, with limited human oversight. And improvements in computer vision are extending RPA into other fields, ranging from medicine to law enforcement.
But perhaps the greatest potential for smarter RPA lies in systems that can learn and improve as they work. More prosaically, systems that also use their intelligence to make it easier for humans to automate processes in the first place.
“RPA application vendors have introduced discovery tools as well as process mining solutions for process streamlining and re-engineering,” says Jukka Virkkunen, co-founder of Digital Workforce. “Better functioning and business-supporting processes will continue to operate autonomously and will not require management like the current ones.”
In short, the next generation of RPA will find, and fix, problems.
“RPA is being integrated as part of larger intelligent automation systems,” says Kalyan Kumar, chief technology officer at HCL Technologies. “You can have a chatbot or cognitive virtual assistant at the front end and RPA in the back end to execute activities or as part of an orchestration process. Then you really start to get value from RPA.”
Chris Huff, Kofax
RPA, although effective, can be time-consuming to set up because a human has to capture the processes and write the rules. Over the last few years, RPA suppliers have moved towards low-code or no-code environments to make it easier for business units to create their own RPA systems or bots. With tools such as Microsoft’s PowerAutomate, bots can even be created by individual users on their desktops.
“Over the past 10 years, RPA has upskilled an entire generation of non-technical citizen developers,” says Kofax’s Huff. “Now ‘automation builders’ can use AI for intelligent content processing, to allow better collaboration between people and machines.”
But whether RPA, even with AI, will deliver its potential will depend on how it is implemented. “There is always a risk that you just automate an inefficient process,” says Gartner’s Tornbohm. “There is also a risk that you hold on to a bad applications suite when it should be rethought – and that you lose some opportunity to justify the legacy replacement as it is supported by RPA.”
CIOs need to be sure that RPA is not simply patching up systems that are past their sell-by date.
And, as RWS’s Bara says, the best way is to start projects small and then grow. “Projects are not successful when they try to tackle large problems early on,” he says. “The way to do it is to break it down so you solve big problems by solving small problems. That is where RPA needs to go.”