One of the fastest-growing areas of artificial intelligence—at least if that term is defined broadly—is “robotic process automation,” a set of capabilities for the automation of digital tasks. RPA, as it is often called, has some valuable functions, but digital-centric companies may need more intelligence and process simplification to than RPA can currently provide.
Let’s review the attributes of RPA as it stands today. It’s a combination of multiple presentation layer interfaces (that is, it can connect as a “digital user” to multiple different systems), basic rules, and simple workflow. It can mimic data flow tasks performed by humans and automate those tasks digitally. RPA systems are easily programmed and modified by nontechnical workers, although implementing dozens of process robots requires significant involvement from technical specialists.
Each “robot” (actually an instance of a program on a server) can do one or a few tasks. Doing multiple tasks requires multiple robots. Some companies—typically large banks with a lot of back-office financial tasks—already have over a thousand robots. RPA implementations typically yield rapid ROI and improvement in cost/time productivity. In some cases, RPA improves the quality of output vis-à-vis human-created output.
The good news and bad news about RPA is that it doesn’t change the underlying systems to which it connects or the process tasks it automates. This is the key to its easy implementation, but it limits the ability to simplify the processes and to modify the underlying systems architecture. In a sense RPA is pouring cement—albeit quick-set cement—around existing systems. Its simple architecture also limits the ability to create and act on intelligence.
Perhaps the key shortcoming of RPA is that it is simply not very smart. RPA as of now doesn’t have much capability to eliminate unneeded process steps, create intelligence, learn, or act intelligently. It is possible that vendors will add intelligence to RPA over time. Already there are some vendors that have incorporated a capability to “observe” human co-workers and take similar actions. And one leading RPA vendor, Blue Prism, recently announced that it was partnering with IBM and other vendors with the goal of adding intelligence to its RPA offerings.
What would it mean to have an intelligent RPA solution? In effective digital organizations, smart machines should be able to:
Eliminate process steps or processes: Perform complex tasks and altogether eliminate the steps performed by humans. e.g. automatically gathering and computing data from multiple sources.
Create intelligence: Create intelligence through interpretation of structured and unstructured information, and facilitate decision making based on the information. For example, an automation solution for the “front desk” in the insurance industry should be able to interpret contracts and invoices, and automatically reconcile them with claims for a cost audit.
Learn Learn from past performance and human behavior to automate exception cases. Smart machines should also be able to learn from the structured/unstructured information to identify new patterns/intelligence. For example, it could develop intelligence about customer preferences from emails, CRM notes, attachments, and so forth. Ideally it would know why a customer is contacting an organization.
Act intelligently: Automate certain tasks based on the intelligence created by the machine. In an order fulfillment process, for example, it should be able to determine whether a delivery truck should be at the warehouse or not. It could also compare the truck plate number with the order management system and send a signal to open the warehouse gate.
It may be difficult to ever accomplish these intelligent capabilities with RPA as the sole or primary technology. Intelligent platforms like RAGE Frameworks (recently acquired by Genpact, where I have done some paid speaking), combined deep automation capabilities with AI skills like natural language processing from the beginning. Other vendors like LoopAI Labs (where I am an advisor) are also combining RPA-like capabilities with intelligent features like deep learning.
Intelligent platforms with the desired smart capabilities may be more difficult to implement than RPA alone, but they typically provide both higher value and greater integration with existing technology architectures. If you find this type of integration appealing, try to make sure that the intelligence capabilities are suited to your needs. RAGE’s primary intelligent component, for example, is based on computational linguistics. LoopAI’s is based on deep learning. Other vendors may employ machine learning, neural networks, or even rule-based expert systems as the primary underlying intelligent capability.
If you’re in the market for digital task automation, I have several recommendations. First, use RPA sparingly unless you are very comfortable with your existing systems architecture. And since RPA will change rapidly over time, don’t make a major commitment to a particular vendor’s RPA in your architecture. You don’t want your flexibility to be limited. Some vendors (WorkFusion, for example) are even beginning to offer open source versions of RPA. It seems likely that basic RPA will become increasingly commoditized, so you may want to explore open source options.
It’s also important to think carefully about how much intelligence you want in your task automation solution. If you believe your requirements will involve more than a few logic rules, you may want to explore more ambitious platforms for intelligent digital work. These may eventually be added to RPA solutions, but there is no guarantee of that. And even if RPA does get smarter, it’s not clear in what ways it will become smart.
This article was originally published in Forbes and LinkedIn Pulse.