Outlook: Where Is RPA Headed?

This article series began with examining what robotic process automation (RPA) is. It then made the case for why you should implement RPA in your organization. Following that was an examination of when to use RPA. Most recently, we covered how to scale RPA. This final article examines where RPA is headed.

The most powerful trend in RPA is its connection with artificial intelligence (AI) to vastly expand the universe of automation possibilities. As you might have gathered from previous articles, RPA is a powerful tool that is limited in scope. It excels at handling rules-based, simple, repetitive processes. However, RPA is less helpful in processes requiring judgment or cognition.

That’s where AI comes in. AI is defined as “the simulation of human intelligence processes by machines, especially computer systems. These processes include learning (the acquisition of information and rules for using the information), reasoning (using the rules to reach approximate or definite conclusions) and self-correction.” AI is the brain to RPA’s brawn.

Many of the largest consulting firms believe that AI is a natural outgrowth of RPA. Deloitte notes that use of cognitive automation—of which AI is part—is highly correlated with successful scaling of RPA, suggesting that it is a natural next step in the automation process. PwC notes that “RPA is shaping up to be the precursor for the broader use of artificial intelligence, or what is known as intelligent process automation (IPA).” RPA can be seen as the first step towards becoming an AI-enabled enterprise.

The table below presents some of the most salient differences between RPA and AI. We’ll focus on two key ways AI expands the possibilities for RPA—access to unstructured data and self-learning. We’ll then conclude with a brief case study of RPA and AI implemented in concert.

Must be taught Self-learning
Uses structured data Uses semi- or unstructured data
Can’t make judgments Can make judgments
Automates repetitive, well-defined processes Automates cognitive, ill-defined processes
Best for: handling the easy work Best for: handling the exceptions
Top applications:


    • Opening spreadsheets
    • Copying data between programs
    • Comparing entries
Top applications:


    • Chatbot language processor
    • Facial recognition image classifier
    • Text extraction

Access to Unstructured Data

The first way that AI expands the possibilities for RPA is by using different technologies to provide structure to unstructured data. As noted earlier, one key limitation of RPA is that it can only use structured data such as spreadsheets and tables. Email body text, transcripts, and other unstructured data are a bridge too far for RPA, yet are estimated to make up 80% of all data generated. New AI technologies such as natural language processing (NLP) and machine vision (MV) provide a way around these limitations. We’ll examine each in turn.

Natural Language Processing

Natural language processing is defined by SAS as “a branch of artificial intelligence that helps computers understand, interpret and manipulate human language.” It works by breaking down sentences of text—spoken or typed—into their smallest elements and then building them back up using techniques like part-of-speech tagging that you might have used in school. While Apple iOS’s Siri and Amazon’s Alexa are the most notable use cases of natural language processing, banks like Royal Bank of Scotland have been utilizing NLP on customer feedback forms to analyze trends in customer satisfaction.

Connecting NLP to RPA might involve automation such as using NLP to process customer support emails to determine what the problem is and then use RPA to take steps to solve it or direct the issue to a human in the appropriate department.

Machine Vision

Machine vision is defined by Cognex as “all industrial and non-industrial applications in which a combination of hardware and software provide operational guidance to devices in the execution of their functions based on the capture and processing of images.” Think of this as natural language processing (NLP) for hand-keyed PDFs. It has already seen some adoption in financial services. CapGemini notes that “[a]t JPMorgan, lawyers spent thousands of hours studying financial deals. Now, an AI system is doing the challenging job of interpreting commercial loan agreements, taking on a task that has swallowed 360,000 hours of work by lawyers and loan officers. The AI system reviews documents in seconds and is less prone to error. The system has cut down on loan-servicing mistakes, many of which resulted from human error, in interpreting 12,000 new wholesale contracts per year.”


One of the key elements of AI is its ability to learn, defined by KPMG as the ability to “gain knowledge from data as ‘experience’ and generalize what is learned in upcoming situations.” An example of this in financial services can be seen in fraud detection. Deloitte notes that “presented with a database of information about credit card transactions, such as date, time, merchant, merchant location, price, and whether the transaction was legitimate or fraudulent, a machine learning system learns patterns that predict fraud. The more transaction data it processes, the better its predictions are expected to become, to the point where it can predict situations just before they actually happen.”

RPA and AI: A Case Study

RPA and AI were used in concert to help Hollard, South Africa’s largest private insurer, from drowning in its backlog of customer support email requests. Each of the 1.5 million yearly emails needed to be “interpreted and classified” to follow SLA and regulatory requirements.

Using an implementation partner, Hollard configured a robot to perform the following steps:

    1. Access the email
    2. Interpret the content
    3. Classify and file documentation
    4. Extract relevant data
    5. Update the system
    6. Connect to the relevant human to carry out task
    7. Send a confirmation email to the requestor upon completion.

You’ll notice that some of these steps sound like typical RPA processes—updating system logs and sending confirmation emails. However, tasks like “interpret,” “classify,” and “connect to the relevant human” are all tasks requiring judgment and thought and would not be possible without AI.

What were the results? The backlog risks were removed by automation and the robots have become almost fully automated in real time, performing the tasks 6 times faster, 91% more cheaply, nearly error free, and have saved almost 2,000 hours—over 11 FTE equivalents—per month.

Component Improvement
Automate it 98% automated
Do it faster 600% faster
Free up FTEs ~2,000 hrs/month (> 11 FTE equivalents) saved
Do it more cheaply Cost/transaction down 91%
Do it more accurately “quality of information has improved drastically and is virtually error free”

Interested in learning more about what RPA can do for you? Read 2 of our case studies, Piloting RPA: Leading a Powerful First Journey Into Automation and Using RPA for Validation: Taking Process Execution to a Whole New Level.