The Return of Expert Systems

By Ric Kosiba, Sharpen Technologies

In the last edition of On Target, we discussed the concept of Micro-Bots, small nuggets of AI or automation code that performs simple, but smart tasks. In our example, we examined a “likely upset customer” bot, that would change that customer’s experience based upon their calling history or an AI model result. This micro-bot would start a few processes, one to route the call to an expert, another to tell the customer he is being treated specially. It would send the customer’s previous and unsuccessful contacts straight to QA for evaluation, and it might add the customer to a “delight list” for extra service items.

Also in this previous article, we discussed the concept of expert systems, and I would like to discuss them further here. I am excited about their potential in contact centers.

AI Confusion

AI is one of those topics that is pretty tough to understand, not because the topic itself is all that difficult, but because the folks that talk about it most have a vested interest in making it sound high tech and over our heads. Mystery makes some things seem much more important than they might otherwise seem. Further, some products are touted as AI, but really just use the same (good) math that has existed for decades. You can’t tell if it’s really AI, because the people telling the story either don’t understand the products or they might be trying to make them seem more “highfalutin.”

It should not be that way. None of the AI concepts are really that tough, especially for us mathy WFM types.

Lately, machine learning has been all the rage, but to be honest, there are not that many machine learning applications in WFM. As a matter of fact, I would argue that there are few to none. But in the contact center proper, there are some solid AI applications, the most obvious being voice transcription and voice analytics, and maybe chatbots. But that doesn’t mean there aren’t other great applications for high powered math. It’s just that I wouldn’t call most of them machine learning.

What is an Expert System and Why is it Relevant Today?

In the eighties and nineties, AI was becoming a thing. Because computing power and memory were limited, the algorithms themselves had to be well constructed; for us algorithm specialists, a major source of research and development was in building extremely efficient algorithms. They had to be simpler.
The first types of AI were limited by the number of searches they could do and the amount of data they could tap into. And the thinking went, “if we are trying to make our systems as good as intelligent people, maybe we should just ask these smarties how they solve problems, and code that up.” These first AI programs were called expert systems, and relied on the logic that experts, or teams of experts, used to solve problems. There were all sorts of techniques that programmers used to elicit this expertise and to test whether the decision steps related by the expert was correct (because even experts don’t always know why they do the things they do).

There were many expert systems successes, but the more complex the program, the more likely the expert was not able to convey all that they knew, and the sketchier and more convoluted the AI rules were. For a while, after some significant failures, AI fell into disfavor.

But as computers got faster and data became richer, more complex forms of AI became viable. But here is something interesting. Even in the age of machine learning, there are many, many problems that it doesn’t solve well or economically. In our micro-bot example, where we are trying to handle a likely upset customer, do we need an expensive piece of AI to help us? Probably not. In this example, the age-old expert system is the way to go—take an expert, or solicit opinions from your team, and ask, “how should we handle an upset customer?” I fully expect that your team’s solution will beat some machine learning algorithm’s. Similar for “which agent should handle which type of call?” Do we need an expensive system to tell us who our best agents are for handling specific types of calls? I expect simple data analyses would answer that question without the added cost.

I would argue that most issues in the contact center can be handled quickly and effectively using micro-bots—chunks of code, with a trigger, model, and action, that solve small and frequent customer issues.

Some benefits of the micro-bot:

  • They can be rolled out in less than an hour
  • They can be tested, using A/B testing, immediately
  • They can be turned off, kept on, or improved over time
  • They are cheap and effective

What do you Call a System Made up of Micro-bots?

In the last On Target, we discussed the idea that micro-bots were the future of AI for contact centers. Sure, there were a few grand AI/machine learning applications that were worth the long implementations and the cost. But free, immediate, and testable AI/automation is the cat’s meow.

When you have one micro-bot, you have a cool function that solves a simple problem. When you have ten or thirty of them, made up of the best ideas of your team, the best practices of the industry, and micro-bots evolved from cycles of tests, measures, and improvements, you have turned your platform into something very interesting. Your contact center platform *is* an expert system.

In the current paradigm, contact centers are as they always have been—ways to handle contacts efficiently, with some add-on systems with maybe some AI, to help manage specific problems. When your contact center is managed through micro-bots, however, it is an expert system, and by itself is AI. This may be the future for managing contact centers.

Ric Kosiba is a charter member of SWPP and is the Chief Data Scientist at SharpenCX. He can be reached at or (410) 562-1217.

SharpenCX provides unified contact center software that empowers agents to deliver engaging customer experiences with an all-in-one, customizable platform.