Mary Lacity and Leslie Willcocks in conversation with Gabe Piccoli, MISQ Executive Editor
The Editor in Chief of MISQ Executive is Professor Gabe Piccoli. In December 2020 he asked Mary (Lacity) and Leslie (Willcocks) to discuss their six-year research program, with the goal of consolidating what we know about Robotic Process Automation, Cognitive Automation and AI, and identifying the remaining challenges for those organisations seeking business value from their automation investments. The conversation ranged widely, and the full transcript has been published in the MISQ Executive in June 2021, volume20, number 2.
Over the next three months we take highlights from the discussion. Part 2 will deal with business value, risks and jobs, and Part 3 with management and emerging challenges.
Gabe Piccoli: Mary and Leslie, besides your books, you published two case studies on service automation in MIS Quarterly Executive. While detailed case studies are useful for our readers, today we are here to reflect on your overall findings and most impactful lessons from your six-year research program. To help orient the audience, please start by explaining some of this terminology.
Mary Lacity: We research how enterprises automate services using a variety of automation technologies. When we began studying this space back in 2014, we encountered a dizzying array of automation products marketed as scripting tools, software robots, robotic process automation, artificial intelligence, desktop automation, cognitive computing, business process management automation, and machine learning, to name a few. The market was very confusing to practitioners and to us.
To make sense of the space, we looked at how these tools worked, the type of data used as input, how they processed data, and the type of results produced. From there, we conceived of automation tools as a continuum (see Figure 1). One end of the spectrum is the realm of Robotic Process Automation (RPA); the other, Cognitive Automation (CA), which people commonly call artificial intelligence (AI).
The realm of RPA consists of tools that automate tasks that have clearly defined rules to process structured data to produce deterministic outcomes. A ‘software robot’ is configured to process tasks the way humans do, by giving it a logon ID, password, and playbook for executing processes. RPA tools are ideally suited for automating those mindless ‘swivel chair’ chores performed by humans, like taking structured data from spreadsheets and applying some rules to update an ERP system. Leslie had the early insight that RPA tools ‘take the robot out of the human’, meaning that the tedious parts of a person’s job could be automated, leaving the human to do more interesting work that requires judgement and social skills. His insight resonated so much with practitioners, that we had to chase down more than one supplier and conference organiser that adopted the phrase as their marketing slogan without crediting Leslie. Automation Anywhere, Blue Prism, and UiPath are the top RPA providers by market share as of 2020, and likely in 2021 as well.
Leslie Willcocks: It is important to note, because of the consequences for ease of scaling, that these vendors do not all provide the same thing. There are variants of RPA ranging from desktop assisted RPA, through enterprise RPA, self-development packages to cloud-based services.
The realm of cognitive automation consists of more powerful software suites that automate or augment tasks that do not have clearly defined rules. We do not like to call such software ‘Artificial Intelligence’ because we believe the AI label aggrandises what these tools do. With CA technologies, inference-based algorithms process data to produce probabilistic outcomes. A variety of tools are in the realm of CA, such as tools that analyse data based on supervised machine learning, unsupervised machine learning, and deep learning algorithms, backed by powerful computing and memory. The input data is often unstructured, such as natural language, either written or spoken. For example, at Deakin University CA was used to answer natural language student inquiries. The input data can also be highly structured, such as the pixels in an image. Google’s Machine Learning Kit, IPsoft’s Amelia, IBM’s Watson suite and Expert Systems’ Cogito are examples of CA tools.
People refer to ‘strong’ AI as trying to use computers to do what human minds can do. The vast majority of organisations are a long way off that! However, ‘AI’ is widely and misleadingly used, especially by vendors, as an umbrella terms for RPA and CA, as well as much more advanced software that has not even made it out of the laboratory. The claims are not helped by the misleading metaphor of comparing the human brain with computing. We have in practice at the moment, in most of our businesses, what can be described as ‘weak, weak AI’—algorithms driven by massive computing power—we call it statistics on steroids.(1)
Although the realm of CA is vast, our research examined how the tools were used to automate or augment back-office processes and customer-facing services. Typical examples were processing medical claims, answering customer queries, and categorising user requests to route them to the humans who could help them.
Gabe Piccoli: What is the history of RPA/CA technology?
Mary Lacity: We’ll begin with a quick history of RPA. In 2012, Phil Fersht, founder of the outsourcing consulting firm Horses for Sources (HFS), used the term RPA in a provocative report entitled ‘Greetings from Robotistan, outsourcing’s cheapest new destination.'(2) It highlighted a UK-based start-up company called Blue Prism. Blue Prism was founded in 2001, but it did not become well-known until their Chief Marketing Officer, Patrick Geary, started calling the product ‘Robotic Process Automation’ sometime in 2012. That term really resonated with practitioners, so much so, that other automation companies started rebranding their tools to call them RPA. All of a sudden, there were over two dozen companies calling themselves RPA by 2016, with a claimed market size of $US 600 million.(3) RPA standards were desperately needed, so Lee Coulter, then CEO of Ascension Shared Services, started an initiative at IEEE. He became the chair of the IEEE Working Group on Standards in Intelligent Process Automation in December of that year.(4) The group published the first standard in 2017, which distinguished between enterprise RPA designed for an organisation and Robotic Desktop Automation (RDA) designed for a single desktop user. Blue Prism is an example of an RPA provider; Automation Anywhere began as an example of an RDA provider.
Gabe Piccoli: So where does RPA stand today, and what is the market value?
Leslie Willcocks: The RPA market was about $2 to $4 billion in 2020, depending on which consulting report you read.(5) RPA annual growth between 30-50 % is still predicted by nearly every source for the foreseeable future.(6) C-suite priorities for emerging technologies shifted rapidly because of COVID19. While the pandemic prompted many enterprises to postpone horizon technologies like edge computing and blockchains, enterprises became laser-focused on technologies that produce rapid return-on-investments (ROIs), and on the top of that list was process automation.(7)
Gabe Piccoli: Moving to cognitive automation, most MISQE readers are likely familiar with the long history of AI, beginning in the 1940s with the work of Warren McCulloch and Walter Pitts on neural nets, the Turing test, the Dartmouth Conference in 1956, and I am sure we all remember when IBM’s Deep Blue beat Gary Kasparov in chess in 1997.
Mary Lacity: Yes, and also the win over Brad Rutter and Ken Jennings in Jeopardy in 2011; Google DeepMind’s AlphaGo victory over Lee Se-dol in 2016.
Gabe Piccoli: MISQE published a special issue on AI in December 2020. In your work, how have these AI technologies…or if you prefer, cognitive automation technologies…been used to automate business services?
Mary Lacity: Organisations find it difficult to pivot CA tools designed for a specific context like chess, Jeopardy or Go to other contexts like processing health records, mortgage applications, and calls to help desks. The early enterprise adopters suffered through painful and expensive implementations, mostly due to the data challenges. Our case companies adopted CA tools with supervised machine learning, which needs thousands of labelled data examples to get the machine learning algorithms to proficiency. However, as much as 80 percent of an organisation’s data is dark, meaning that the data is un-locatable, untapped, or untagged. Enterprises first had to create new data and clean up dirty data that was missing, duplicate, incorrect, inconsistent or outdated. Enterprise adopters of CA tools also struggled with difficult data, which we define as accurate and valid data that is hard for a machine to read, like a fuzzy image, unexpected data types or sophisticated natural language text. In our CA cases, much of this work was done by tedious human review. But we did find examples of organisations that eventually got value from their CA adoptions, like Deakin University, Zurich Insurance and KPMG, and quite a lot more now, though these still tend to be islands of automation rather than enterprise-wide. As far as size of the CA market, it’s reported to be somewhere between $50 and $150 billion in 2020.(8)
Gabe Piccoli: So, CA is much bigger than RPA in terms of market size. You mentioned a convergence of RPA and CA. Would you explain that?
Leslie Willcocks: RPA and CA have different histories that are now converging to what practitioners are calling ‘intelligent automation’. The idea of intelligent automation is to institutionalise a well-designed automation program using a platform for pluggable tools that are best-in-class. In our case studies, business value was not derived from the selection of one technology or service vendor, but through the ability to identify and connect different technologies that maximise the full potential of modern automation technologies. To facilitate this, vendors are at different stages in developing automation platforms that can harness both RPA and CA technologies. Some enterprises also use their own platforms to buy best-in-class tools and integrate them internally.
Gabe Piccoli: Would you give an example of an integration of RPA and CA?
Mary Lacity: IBM is a good example…at least they drink their own Kool-Aid. IBM’s global IT outsourcing services business combined Blue Prism RPA software with components of IBM’s Watson technology. In one service area, namely, Global Technology Services Technology, Innovation & Automation, IBM has well over one thousand customers, and a range of automation tools, including 200 RPA software licenses run out of London and Amsterdam.
Email ticket triage was one process that was automated by integrating technologies. IBM’s customers have email task IDs with which they can send a support request to IBM. The requests are typically things like “Hey, my printer is not working” or “I’ve forgotten my password.” Any email request needs to be logged into the ticketing system and routed to the correct resolver group. Before this was done by humans. From 2017, the RPA software robot uses its logon ID to log into the ticketing system. It retrieves the email and logs it into the ticketing system and engages the CA tool, which categorises the issue: “this is a network support request”, or “this is a telephony request”. The CA tool typically has a high degree of certainty, having been trained on thousands of historical tickets. When confidence is low, the ticket is escalated to a human for routing and the label fed back to the CA tool for learning. The RPA tool closes the loop by identifying the appropriate team and routing the ticket to the correct resolver group.
Leslie Willcocks: The example is just one of many. Increasingly we see CA tools feed into RPA software which acts as the execution engine, especially in banking, insurance and financial services organisations. A bank, for example, will have an interactive chat bot at the front end in dialogue with customers, but it will draw upon RPA to get the information it needs to be able to have a more accurate conversation with the end user, for example, about a stolen credit card. Going forward we will see even better integration of RPA and CA software moving towards automation exchange platforms that are increasingly cloud-based. And these will integrate with much bigger digital platforms operating as the foundations of the mature digital businesses that will be very much with us by the late 2020s.
Next Month … Part 2: Business Value, Risks and Jobs
(1) See Willcocks, L. (2020) Why misleading metaphors are fooling managers about the use of AI. Forbes April 23rd, 1-5.
(2)Fersht, P. (2012). Greetings from Robotistan, outsourcing’s cheapest new destination, HfS Research, https://www.horsesforsources.com/robotistan_011112
(3)Fersht, P., Snowdon, J. (2018). RPA will reach $2.3bn next year and $4.3bn by 2022... as we revise our forecast upwards. https://www.horsesforsources.com/RPA-forecast-2016-2022_120118
(4)IEEE 2755-2017 - IEEE Guide for Terms and Concepts in Intelligent Process Automation https://standards.ieee.org/standard/2755-2017.html
(5)Robotic process automation (RPA) market revenues worldwide from 2017 to 2023: https://www.statista.com/statistics/740440/worldwide-robotic-process-automation-market-size/
(6)Fersht, P., Snowdon, J. (2018). RPA will reach $2.3bn next year and $4.3bn by 2022... as we revise our forecast upwards. https://www.horsesforsources.com/RPA-forecast-2016-2022_120118
(7)Allied Market Research, reporting on 13th October 2020, estimated the RPA market at at end of 2027 as $US19.53 billion at an annual compound growth rate of 36.4%.
(8)KPMG and HFS (2020). Enterprise Reboot: Scale Digital Technologies to grow and thrive in the new reality. https://home.kpmg/xx/en/home/insights/2020/08/enterprise-reboot.html
(9)Grand View Research (2020). Artificial Intelligence Market Size, Share & Trends Analysis Report, https://www.grandviewresearch.com/industry-analysis/artificial-intelligence-ai-market
IDC (2020). IDC Forecasts Strong 12.3% Growth for AI Market in 2020 Amidst Challenging Circumstances, https://www.idc.com/getdoc.jsp?containerId=prUS46757920