We already hear a lot about artificial intelligence (AI) systems being able to automate repetitive tasks. But AI is such a large term that encompasses many types of very different technologies. What type of solutions are really able to do this?
Robotic Process Automation (RPA) configures software to mimic human actions on a graphic user interface (GUI) to carry out a business process. For example, an RPA system could open a relevant email, extract information from an attached invoice, and input it in an internal billing system. Although modern RPA solutions are already relying on various AI-powered technologies like image recognition to perform their functions, positioning RPA within the spectrum of AI-powered tools is still somewhat premature: on its own, RPA is basically just an alternative to scripting for non-technical users.
Enterprises that are currently beginning with automating prescribed tasks hope to adopt more advanced capabilities like data-based analytics, machine learning, and ending with cognitive decision making; they should however realize that existing RPA solutions might not yet be intelligent enough for such aspirations.
Filling in the Gaps
If RPA sounds limited, then you are correct; it is not a one-stop-shop for intelligent automation. RPA only automates the button clicks of a multi-step process across multiple programs. If you’re under the impression that RPA can deliver end-to-end process automation, pause and reassess. RPA can do a limited and explicitly defined set of tasks well, but faces serious limitation when flexibility is required.
As soon as any deviation from the defined process is needed, RPA cannot and does not function. However, it can be part of a larger business process orchestration that operates from an understanding of what must be done instead of how. RPA delivers some value in isolation, but much more is possible when coordinated with other AI systems.
The weaknesses of RPA systems overlap nicely with the potential that machine learning (ML)-based AI can offer. ML happens to be capable of adding flexibility to a process based on data inputs. Solutions are coming available that learn from each situation – unlike RPA – and produce interchangeable steps so that the system can assess the type of issue to be solved, and build the correct process to handle it from the repository of already learned steps. It widens the spectrum of actions that an RPA system can make.
Synchronization Adds Value
AI does have strengths that overlap with RPA weaknesses like handling unstructured data. An AI-enabled RPA system can process unstructured data from multiple channels (email, document, web) in order to input information later in the RPA process. The analytics functionality of ML can add value to an RPA process, such as identifying images of a defective product in a customer complaint email and downloading them to the appropriate file. There are aspects that the pairing of RPA and AI do not solve, such as end-to-end process automation, or understanding context (at least not yet).
Overall, RPA’s value to a process increases when used in combination with other relevant AI tools.