Thu, Feb 20, 2020
ByProtiviti KnowledgeLeader

A Beginner's Blueprint for Implementing Intelligence Automation

Discussions of robotic process automation (RPA) and artificial intelligence (AI) tend to follow separate tracks. This has been a function of the way these technologies have been marketed and sold. This is changing as companies realize that both RPA and AI are required to achieve any significant degree of end-to-end process automation.

For example, even a company that is not ready to delegate decision making to an intelligent robot and prefers to automate certain processes using RPA will still need to invest in tools like computer vision, natural language processing, voice recognition and others to achieve the desired level of automation. These tools may not, in and of themselves, be intelligent but they use techniques classified as AI. Alexa is an example of a tool that does not make decisions per se but does evolve its interaction with humans based on input. While the user may not see it, these tools are being trained continuously and they get better over time, learning from the mistakes they make and the corrections by users.

That’s not to say that there aren’t many processes that can be automated relatively easily with RPA alone, but RPA without some elements of AI is quite limiting. Practically speaking, most processes have elements that require either visual or auditory interpretation. Even something as basic as posting a purchase order requires the ability to visually scan and extract the required information from documents that vary in format from company to company. RPA alone cannot do that.

Similarly, AI-enabled virtual agents may be great at carrying on conversations, but at some point the work – creating a purchase order, generating a shipment label, etc. – is going to have to be handed off to either a person or a bot.

AI and RPA most often meet at the intersection of two tasks, where some kind of hand-off or judgment call is required, converging into what’s called intelligent automation.

To illustrate, let’s take a physical process, such as responding to inquiries via email in an automated way. To do this, you (or an intelligent bot) need to be able to derive the meaning of the email and extract the key information from it to determine what action to take. Only when you (or the intelligent bot) have done this, the bot can take the appropriate action.

Most processes include at least one step where data is received in an unstructured form (whether voice, chat, email or a nonstandard structured form such as an invoice, which differs from one company to another). Deriving meaning from this data is the territory of AI, while following up with a specific predetermined action can be handled by RPA.

The distinction may seem like an unnecessary technical detail to a manager whose goal is to “increase the level of automation in the business.” The important takeaway is that, as organizations move forward with RPA implementations over the next couple of years, they need to ensure that the implementation team understands the key interaction points between execution and intelligence and have people with AI skills on the project team. Whether this means building a team in-house or partnering with an outside consultant, companies need a plan to acquire talent with the AI skills, as so many transformational goals depend on it.

To read more on this topic, download and explore the following content on KnowledgeLeader: