Three blogs in already and we have yet to tackle the elephant in the room: what precisely is robotic process automation (RPA)?
Basically, RPA is a new type of software that acts as a virtual workforce. It is configured to replicate the activities previously carried out by human workers, delivered by ‘robots’ or ‘bots’ or ‘virtual engineers’. There are a number of emerging software providers that sell RPA platforms, such as Blue Prism, Automation Anywhere, UIPath, OpenSpan and others – including some of the large outsourcing companies. RPA is ‘robotic’ in the sense that it is designed to replicate and replace human work. RPA is only and always digital – it is computer-based work, hosted on servers. The images of shiny metal robots – so beloved of technology marketing departments – are all very nice but misplaced. RPA has not caused a single metal finger to press a keyboard, and never will.
So far RPA has been deployed most in high volume, digital environments where there are labour-intensive processes to enter or check data, format or analyse information, or monitor and escalate issues – simple transactional tasks that don’t require great intelligence but can soak up vast amounts of labour because they are fiddly. In particular insurance, telecoms and utility companies have been early adopters, deploying RPA to the core operations such as new business, admin, and non-voice customer contact.
So far, RPA may not sound all that different to other types of automation, such as ERP, workflow, BPMS, or other business applications – after all each of these directly or indirectly seeks to replace human workers. However the focus of RPA software is to operate existing applications and systems, recording and mimicking the way that human users login, navigate and interact with these tools (some people use the word “orchestration” – RPA conducts your legacy system orchestra). RPA is not integrated with these systems; in other words the existing applications ‘think’ the RPA robot is a human user. This sounds simple, and it is, but it is different to how most other automation works. In a word RPA is ‘non-integrated’, it sits on top of legacy systems without the need for complex, bespoke interfacing.
This leads to another differentiator: RPA is quick (some might say ‘quick and dirty’). Because RPA software is not integrated in the same way as other software, it doesn’t require extensive use of programmers to develop code and interfaces; instead the focus of RPA development is on modelling the process flows and business rules that the robots need to follow. This is typically a two-to-three-month process rather than the 12 to 18 months of intensive IT-surgery required for an ERP module implementation, for example.
This short, lean development cycle means that development costs are also lower, resulting in quicker payback – and another differentiator compared to traditional automation. RPA projects, particularly pilots, can start small because there is no massive development overhead to recoup. If an RPA implementation can be done in a few months, for say £50k, then it makes sense to target very specific, irritating points of failure in a given process that were previously too small, or expensive to address via automation. So for example, there are use cases of RPA solutions to handle peaks of cash receipts in accounts receivable, of SIM card set up in mobile telecoms, or password admin in IT helpdesk.
But RPA is not just about small-scale point automations – and this differentiates it from the lower end of the automation hierarchy, macros. For years companies have been deploying very rapid, very cheap automation via Excel macros and other desktop automation. Macros can certainly be effective, but are seen as cumbersome and risky for being used on any significant scale. RPA is closely related to desktop automation, but is designed to be scaled up across an enterprise, and I am aware of RPA deployments now delivering work equivalent to hundreds of workers, and rising.
Finally RPA is not artificial intelligence. It is robotic in the sense that it mimics human interactions with your systems, but it follows user-defined rules to deliver this. RPA is based on smart new software, but it doesn’t deploy cognitive technologies to spot patterns, learn or handle ‘fuzzy situations’. AI in services (such as Celaton InStream, IBM Watson, IPSoft Amelia, Wipro Holmes) is the next generation of automation. This will be the subject for many Live Wires blogs in due course, but for the moment is suffices to note it is not RPA. It is also in terms of business impact some years behind RPA – the list of actual public use cases for AI/cognitive automation is rather, ahem, limited.
So RPA is a new type of software that acts as a virtual workforce. It targets labour-intensive, rules-based computer work. It sits on top of your existing applications and systems, is quick and cheap to implement, and can be scaled. So in principle, it is clearly different to traditional automation applications, or digitisation tools, or desktop automation. Stepping beyond the definition, though, there are challenges. In practice different RPA tools don’t always fit neatly into this categorisation. Some RPA tools incorporate elements of other types of automation such as workflow or digitisation technologies. Some new automation tools share most of the features of RPA listed above, but not all. Some vendors of RPA tools rather confusingly don’t use the word RPA (you might see the words virtual worker, robot agent, autonomics).
The reality is that RPA is a convenient label for a family of similar tools and technologies, a family which is still growing and finding its shape. In the last few years, as the number of vendors has increased, and as awareness amongst potential buyers has grown, use of the RPA label has gathered momentum. It seems likely that this will gather pace and in the years ahead RPA will become a more coherent market. At the same time, it is also clear that specialisation and fragmentation will occur, with niche automation tools for sectors such as insurance, local government, pharma, and functional specialisations such as ITPA (IT RPA).