RPA and cognitive automation are sometimes used interchangeably. While they are both important technologies, there are some fundamental differences in how they work, what they can do and how CIOs need to plan for their implementation within their organization.
Key distinctions between robotic process automation (RPA) vs. cognitive automation include how they complement human workers, the types of data they work with, the timeline for projects and how they are programmed.
CIOs also need to address different considerations when working with each of the technologies. RPA is typically programmed upfront but can break when the applications it works with change. Cognitive automation requires more in-depth training and may need updating as the characteristics of the data set evolve. But at the end of the day, both are considered complementary rather than competitive approaches to addressing different aspects of automation.
Taking out vs. putting in
"RPA is a technology that takes the robot out of the human, whereas cognitive automation is the putting of the human into the robot," said Wayne Butterfield, a director at ISG, a technology research and advisory firm.
RPA is a simple technology that completes repetitive actions from structured digital data inputs. Cognitive automation is the structuring of unstructured data, such as reading an email, an invoice or some other unstructured data source, which then enables RPA to complete the transactional aspect of these processes.
Structured vs. unstructured
RPA has been around for over 20 years and the technology is generally based on use cases where data is structured, such as entering repetitive information into an ERP when processing invoices.
"It's fairly straightforward and the technology often relies on such approaches like screen scraping," said Tom Taulli, author of The Robotic Process Automation Handbook: A Guide to Implementing RPA Systems.
Cognitive automation, on the other hand, usually works with unstructured data, such as emails, voice messages, videos and so on. Cognitive automation can scan an invoice to find the right amounts to pay, identify the payee and even detect inconsistencies, which could be an alert to fraud.
Knowing the differences can be quite important for CIOs, Taulli said. Traditional RPA usually has challenges with scaling and can break down under certain circumstances, such as when processes change. However, cognitive automation can be more flexible and adaptable, thus leading to more automation.
Short-term vs. long-term ROI
RPA provides immediate ROI, whereas cognitive automation often takes more time as it involves learning the human behavior and language to interpret and automate the data, said Deven Samant, director of enterprise solutions and head of enterprise data and cloud practice at Infostretch, a digital engineering professional services company.
RPA automates routine and repetitive tasks, which are ordinarily carried out by skilled workers relying on basic technologies, such as screen scraping, macro scripts and workflow automation. RPA performs tasks with more precision and accuracy by using software robots. But when complex data is involved it can be very challenging and may ask for human intervention. This is where cognitive automation comes in.
Tactical vs. strategic functional reach
Another viewpoint lies in thinking about how both approaches complement process improvement initiatives, said James Matcher, partner in the technology consulting practice at EY, a multinational professional services network. Process automation remains the foundational premise of both RPA and cognitive automation, by which tasks and processes executed by humans are now executed by digital workers. However, cognitive automation extends the functional boundaries of what is automated well beyond what is feasible through RPA alone.
Cognitive automation can extend the nature and diversity of the data it can interpret and complexity of the decisions it can make compared to RPA with the use of optical character recognition (OCR), computer vision, natural language processing and virtual agents. These skills, tools and processes can make more types of unstructured data available in structured format, which enables more complex decision-making, reasoning and predictive analytics.
RPA usage has primarily focused on the manual activities of processes and was largely used to drive a degree of process efficiency and reduction of routine manual processing.
Although it is very effective at this and its applicability across all functional domains drives significant value, it is seldom able to drive a truly transformational change in the underlying value chains due to its task focus and inability to deal with complex decision-making.
"Cognitive automation, however, unlocks many of these constraints by being able to more fully automate and integrate across an entire value chain, and in doing so broaden the value realization that can be achieved," Matcher said.
Increased reach vs. increased management
Cognitive automation expands the number of tasks that RPA can accomplish, which is good. However, it also increases the complexity of the technology used to perform those tasks, which is bad, argued Chris Nicholson, CEO of Pathmind, a company applying AI to industrial operations.
One concern when weighing the pros and cons of RPA vs. cognitive automation is that more complex ecosystems may increase the likelihood that systems will behave unpredictably. CIOs will need to assign responsibility for training the machine learning (ML) models as part of their cognitive automation initiatives.
If-then vs. human augmentation
RPA excels at automating rules-based tasks that strictly follow if-then-else logic, whereas cognitive automation is better suited at mining for insights that augment qualitative human judgment, said Chris Huff, chief strategy officer at Kofax, an automation tools provider.
RPA is best deployed in a stable environment with standardized and structured data. Cognitive automation is most valuable when applied in a complex IT environment with non-standardized and unstructured data.
According to Huff, he often sees them deployed together. For example, in an accounts payable workflow, cognitive automation could transform PDF documents into machine-readable structure data that would then be handed to RPA to perform rules-based data input into the ERP.
Straight through processing vs. exceptions
RPA is best for straight through processing activities that follow a more deterministic logic. In contrast, cognitive automation excels at automating more complex and less rules-based tasks.
"Cognitive RPA is adept at handling exceptions without human intervention," said Jon Knisley, principal, automation and process excellence at FortressIQ, a task mining tools provider.
For example, cognitive automation can use AI capabilities like OCR to capture text from a document and natural language processing to understand the entities like users, invoice items and terms and organize them into appropriate fields in a procurement and payment workflow. Cognitive automation can also use AI to support more types of decisions as well. For example, a cognitive automation application might use a machine learning algorithm to determine an interest rate as part of a loan request.
"A human traditionally had to make the decision or execute the request, but now the software is mimicking the human decision-making activity," Knisley said.
Programmatic vs. scalable learning
Comparing RPA vs. cognitive automation is "like comparing a machine to a human in the way they learn a task then execute upon it," said Tony Winter, chief technology officer at QAD, an ERP provider.
RPA is taught to perform a specific task following rudimentary rules that are blindly executed for as long as the surrounding system remains unchanged. An example would be robotizing the daily task of a purchasing agent who obtains pricing information from a supplier's website.
Conversely, cognitive automation learns the intent of a situation using available senses to execute a task, similar to the way humans learn. It then uses these senses to make predictions and intelligent choices, thus allowing for a more resilient, adaptable system. These systems tend to learn on the job. Newer technologies live side-by-side with the end users or intelligent agents observing data streams -- seeking opportunities for automation and surfacing those to domain experts.
"RPA is a great way to start automating processes and cognitive automation is a continuum of that," said Manoj Karanth, vice president and global head of data science and engineering at Mindtree, a business consultancy.
RPA tools were initially used to perform repetitive tasks with greater precision and accuracy, which has helped organizations reduce back-office costs and increase productivity. While basic tasks can be automated using RPA, subsequent tasks require context, judgment and an ability to learn. Cognitive automation can use AI techniques in places where document processing, vision, natural language and sound are required, taking automation to the next level.