Automation in Disguise

December 6, 2019
Automation in Disguise

Robots, automation, artificial intelligence—is there a difference? The short answer is yes. Adam Silver breaks down the main differences between automation and AI, while sharing the characteristics and traits you should be looking for when picking an AI business solution.


Let’s start by raising the wizard’s curtain. If a vendor is trying to sell you an AI-based automation solution, then they’re almost certainly hooking you with hype and not substance.

To ensure you’re investing in an automation solution that will meet the needs of your business, at an appropriate price point, it’s important to understand the options before you commit and know how to differentiate them.

In the automation solutions space, robotic process automation (RPA) is often represented as AI, which is problematic because they’re not the same thing. Not by a long shot.

RPA is a rules-based technology. Robots are programmed to help businesses automate tedious activities—like payroll, procurement, and processing insurance claims, for instance. This technology is great at eliminating human error, reducing compliance issues, and tracking information more efficiently so it becomes easier to diagnose issues or problems. Sounds great, but that’s not AI.

Simply put, RPA is a software-based solution that follows predetermined rules, while AI is designed to simulate human thinking. RPA must be told what to do whereas true AI is capable of learning and thinking on its own.

The automation spectrum

If you think of automation as a spectrum, with RPA on one end and AI on the other, there’s a lot of space in between—and that’s where natural language processing (NLP) and machine learning (ML) reside. Both technologies are more advanced than RPA, but still can’t be considered true AI.

To better understand this spectrum, let’s look at an example of these technologies in action—say, in the area of invoice processing. In a non-automated scenario, suppliers would typically mail in their invoices, which would be gathered into a paper folder, at which point the relevant information would be extracted and the bills would be created using accounting software (one step above this would be to have invoices being emailed instead of mailed, with attachments saved in a virtual folder).

With RPA, the grunt work is taken care of—meaning an RPA solution would retrieve the emails, download the attachments into a defined folder, and create the bills using an automated, repetitive action like copy and paste. Since invoices differ from supplier to supplier, however, RPA would have to be explicitly programmed to extract the relevant information from each one—which would be a time-consuming process.

At its most basic level, NLP is an automation solution that can read unstructured information. So, in the previous invoice processing example, it wouldn’t have to be programmed to read different types of invoices, like an RPA solution would. That said, it would still have to know what type of information to look for—say, specific keywords or number lines.

ML, similarly, needs to be taught information as it goes along. So, if an invoice has new expense lines previously unseen, the algorithm may ask for clarification on how to handle them. At this point, a human would have to show it the correct approach—a lesson it would learn from and ultimately apply in the future.

True AI, on the other hand, would read the invoices and extract all important information—including such things as the invoice number, supplier name, and due date—without prior programming. It would be able to decipher different invoices, regardless of formatting, as a person would.

RPA in disguise

While RPA is a great technology—and well-suited to certain applications—it shouldn’t be confused with AI. To differentiate between the two, it can be helpful to remember AI’s three characterizing qualities:

  • intentionality: AI agents leverage available data—from applications such as sensors, digital data, remote inputs, or a combination of these sources—to intentionally assess situations and reach conclusions
  • intelligence: by leveraging machine learning and data analytics, AI agents can make decisions based on a variety of different considerations—such as efficiency, equity, justice, or effectiveness
  • adaptability: AI agents can learn and adapt as circumstances and conditions change, and make new decisions based on the information at hand

Although AI exists, as evidenced by AlphaZero and IBM Watson, there aren’t really any business automation applications in this area yet. But we’re getting there.

That said, if your business is looking for an AI solution, here’s a list of traits it should possess:

  • knowledge
  • reasoning
  • problem solving
  • perception
  • learning
  • planning
  • ability to manipulate and move objects

The next time you’re evaluating different AI-based automation solutions, ask the vendors what exactly the AI does and how. If their answer still sounds like a rules-based feature, there’s a good chance it’s not really AI.

Be sure to consider what you’re buying and what you actually need

Today’s automation landscape is evolving at a phenomenal rate and, with no trusted leaders in the space yet, it can be difficult to differentiate between the providers that offer truly revolutionary solutions and those that offer empty promises.

Understanding the types of offerings on the market—and the differences between RPA, NLP, ML, and AI—is a good starting point to confidently navigate this terrain. With that information in hand, you’ll be better equipped to ask the right questions, zero in on the solutions best suited to your company’s needs and make technology investments with your eyes open.


Adam W. Silver is the Managing Director of  the Performance Acceleration practice at Farber. The Performance Acceleration practice helps executives and boards overcome operational and strategic challenges to uncover potential and unleash performance. Adam can be reached at 416.496.3734 and asilver@farbergroup.com.