KOHb - Getty Images

How RPA and machine learning work together in the enterprise

Use cases demonstrate how using RPA and machine learning with other AI techniques achieves 'intelligent automation,' but the best automation solution depends on a company's needs.

More enterprises have adopted RPA functions to automate rote, repetitive tasks, but sometimes they need more capabilities. Enter machine learning functions and the result is "intelligent automation" which, unlike RPA, can learn and adapt.

The choice between the two should depend on the use case, but in today's AI-crazed world, there's a misconception that intelligent automation must be better when, in fact, robotic process automation (RPA) may be a more elegant solution.

"We view AI/ML as knowing what to do, RPA is knowing how to do it," said Muthu Alagappan, chief medical officer at intelligent automation platform provider Notable Health. "For example, OCR can be used to extract information from insurance cards, photo IDs and clinical documents. From there, RPA [enters] the extracted data into existing systems of record."

How RPA and machine learning work together

RPA simply executes its programming, so if requirements change, it needs to be reprogrammed. Machine learning is more dynamic.

"Machine learning relies on large data sets to inform computer systems how to make decisions," said Tommy McEvoy, senior lead technologist in the AI practice at management and IT consulting firm Booz Allen Hamilton. "An exciting advancement in the automation space is the integration of these capabilities, where RPA becomes the engine that accelerates ML, NLP and AI capabilities with the ability to produce an output at scale."

By having RPA rapidly clean and feed data into a machine learning algorithm, an organization can achieve a fully automated solution. For example, Booz Allen developed fully automated service solutions that can capture a customer's refund requests over the phone, transcribe that information, classify the customer's intent and then translate all of that into an appropriate trigger for the automation.

"A true automation platform includes RPA and machine learning, as well as decision management frameworks and event architectures to trigger actions," said Bill Lobig, VP of product management at IBM. "RPA has driven a significant rise in document extraction technologies, systems integration and process mining. I think all of these things together are what you need for intelligent automation, but certainly RPA and machine learning are a big part of it."

Computer vision and RPA

Genpact, a global IT technology services company, uses computer vision to make RPA more effective and more applicable to a wider range of use cases. The company also pairs machine learning with computer vision to discover and mine existing business processes, as well as their deviations and variations. The company also uses machine learning to look at RPA engine log files to determine the root cause of issues that need to be resolved in RPA.

"We use the computer vision capability a lot because there's a lot of unstructured data sitting in PDFs and other things," said Sanjay Srivastava, chief digital officer at Genpact. "We use ML for three things: designing the [process automation] configuration rules, execution and eliminating systemic upstream issues that drive downstream problems."

Srivastava also underscored the need to build a data foundation, which some organizations overlook.

"I find people jumping into RPA without having thought through that. Data science professionals know they can't get a thought out of the garage unless they have a database structure set up, so I would not lose focus on that in the context of RPA," said Srivastava. "The true test of RPA is not around automating the stuff that you know, it's about the stuff you didn't know was happening. Process discovery and process mining are central to figuring out the footprint, which is a massive opportunity for data scientists."


RPA lacks intelligence. Intelligent automation adds machine learning and other AI techniques, such as computer vision and NLP, based on the use case. Might AI wholly replace RPA? That's unlikely because not everything that needs to be automated requires machine intelligence.

If the business process was badly designed to begin with, automation will just accelerate its execution.

"If you're pushing RPA into thousands of bots, it's a nightmare managing all of those bots and you don't know where things are breaking. It's time for a better way: intelligent automation," said Anand Rao, global artificial intelligence lead at multinational professional services firm PwC. "Where data scientists go wrong is in building a deep learning transformer model to show off intellectual superiority as opposed to building the thing a company really needs. Be cognizant of the right tool for the right kind of task."

Beware of automating business processes as-is

Today's organizations are heavily focused on cost control and improving efficiencies, both of which have driven RPA adoption. Yet, RPA is sometimes implemented without questioning whether the original process still makes sense. If the business process was badly designed to begin with, automation will just accelerate its execution. Alternatively, a business process may seem ripe for automation, but it may be that neither RPA nor intelligent automation is the best solution.

For example, an insurance company wanted to automate its paper-based claims filing process because it was slow, expensive and riddled with manual errors. Claimants mailed in paper forms, so the information contained in them needed to be retyped into a claims processing application. If the claims information could have been extracted digitally, the same process would have been done faster and with fewer errors. However, an even better solution would have been solving the data quality problem with a mobile app that allows claimants to enter and verify their information. In the end, the process wasn't automated, it was completely redesigned.

The lesson learned is to consider the business problem first, then consider technology options. 

Next Steps

Choosing between a rule-based vs. machine learning system

Dig Deeper on Enterprise applications of AI

Business Analytics
Data Management