Robotic process automation has helped companies overcome their legacy straightjackets to cut costs and accelerate business processes. But it has often been applied in a siloed fashion, creating new management overheads as RPA systems proliferate. Now, vendors and analysts are describing the next wave of automation, which applies AI in order to automate more complex processes and even decision making. Will this ‘intelligent automation’ help address the management headaches of RPA, or will it amplify them?
Robotic process automation has been one of the stand-out success stories in enterprise IT in the past decade. It has spawned a crop of unicorns – among them UiPath, Appian, Automation Anywhere, BluePrism and Celonis – and a market worth $1.6bn in 2020, according to Gartner. Last year, the analyst company predicted that “90% of large organisations globally will have adopted RPA in some form by 2022”.
Automating repetitive clerical tasks with RPA has delivered benefits including reduced errors, greater process efficiency, and improved employee decision making, according to an IDC survey of RPA adopters sponsored by UiPath and PwC. In a case study documented by LSE professor Leslie Willcox, a European utility provider used RPA to cut development times by up to 40%.
Crucial to RPA's success has been the fact that, by mimicking user behaviour, it allows organisations to automate processes without re-engineering their legacy systems. "RPA’s outside-in approach avoided the overhead of changing the internals of legacy software and as a result, its adoption rate has been increasing, leading to its multi-billion dollar valuation," a report by IBM Research AI explained last year.
Indeed, modernising operations is an explicit aim of automation initiatives. A survey of 400 executives at Global 2000 organisations by analyst company HfS Research found that 65% are 'aggressively' pursuing automation to modernise their legacy practices.
But because RPA circumvents the need for architectural changes to an organisations' infrastructure, it can lead to siloed and piecemeal implementations, critics say. "That's why we declared the death of RPA in 2019," says Reetika Fleming, research leader at HfS Research. "We saw this type of tooling being used to band-aid legacy [systems], in such a siloed way. And now we've created new legacy, because there are all these 'bot estates' that need to be managed over time."
We saw this type of tooling being used to band-aid legacy [systems], in such a siloed way. And now we've created new legacy, because there are all these 'bot estates' that need to be managed over time.
Reetika Fleming, HfS Research
Once a company has achieved some 'quick wins' with RPA, it often becomes a hammer looking for a nail, Fleming adds, with RPA teams seeking out processes to automate, regardless of whether RPA is the best approach.
Dave Elliman, global head of technology at consultancy ThoughtWorks, reports similar dissatisfaction with process mining tools, which analyse system logs to map out a company's business processes in preparation for automation. "I've seen variable success – and by variable, I mean [from] abject to relative failure," he says. Process mining is often met with opposition from enterprise architects, Elliman says, who already have a detailed understanding of their organisations systems.
What is intelligent automation?
These are the shaky foundations on which the industry is teeing up a new wave of "intelligent automation". Visions for this upcoming wave vary in scope and nomenclature, but a significant shift in enterprise IT is widely anticipated.
At one end is the narrowly defined concept of intelligent process automation (IPA), which describes the application of AI to existing process technologies such as RPA and process mining. According to the IBM Research AI report, IPA extends beyond RPA to "automate complex tasks which require decision making, insights and analysis or the composition, coordination and collaboration of multiple IPA solutions". Examples include using natural language processing and deep learning techniques to improve process discovery, to predict business outcomes, and to provide decision-support for human workers.
Adoption of IPA has been limited so far, however. This may reflect a lack of trust in AI systems, IBM's researchers write, and there is work to be done on improving the transparency of AI-powered process automation. But IPA is also costly to implement and maintain. It "requires data preparation and feature engineering, before building and validating the AI capabilities," while "AI models must be retrained in response to changes in the business process or changes in the data."
Furthermore, the researchers warn, "handling more complex tasks will require the composition of multiple IPAs, as well as the collaboration and co-ordination of these IPAs," requiring frameworks that have yet to be developed.
Ops teams are dealing with modern application architectures - real-time apps and event-driven architectures - and they cannot configure and deploy these apps manually. It's all got to be automated.
Charlotte Dunlap, GlobalData
Meanwhile, analyst houses present a more expansive vision of the future of automation, which includes not just AI but also low-code platforms, platform-as-a-service, traditional business process management systems and new working practices for IT teams. Gartner describes this phenomenon as 'hyperautomation'; Forrester calls it 'digital process automation'. HfS Research has coined the term 'NATIVE automation', standing for 'nextgen automation technologies integrated very easily'. A recent framework tender for the NHS refers simply to 'intelligent automation'.
Demand for this suite of tools comes in part from IT operations teams, who are wrestling with the complexity of multi-cloud architectures and under pressure to accelerate digital transformation, says Charlotte Dunlap, senior analyst for application platforms at GlobalData. "Ops teams are dealing with modern application architectures – real-time apps and event-driven architectures – and they cannot configure and deploy these apps manually. It's all got to be automated." This explains a recent spate of acquisitions in the space, including Salesforce's purchase of RPA provider Servicetrace earlier this month, Dunlap says.
Getting started with intelligent automation
Whatever shape it may take, organisations pursuing intelligent automation are at risk of repeating the mistakes that blighted many RPA implementations. "Initiatives to add AI-enabled intelligence to business processes are often delivered in silos without an integrated strategy," Gartner warns. "This results in future scaling challenges." A lack of guidance on how to integrate RPA with other platforms deprives enterprises of strategic gains, it adds.
So how can organisations lay the groundwork for effective intelligent automation? The key, says HfS Research's Fleming, is not to view it in isolation. Automation is one pillar of digital transformation, she says, alongside data analytics and people and culture.
Organisations must manage these in concert to find the right balance of automation and human work, asking "how much of this work actually needs to exist and of that, how much of that can be automated? And then of that remaining work, where should this work be done and by who?"
Elliman predicts that, whatever the latest platform may be, enterprise organisations will continue to create siloed systems until they change the way they approach technology problems. "If people don't solve problems together, then the accountability for systems being accurate, correct and usable gets split up between the business request and the IT supply. The collective is at fault."
