Artificial Intelligence

How to Train Your Chatbot

Automated chat is more prevalent every day; here’s how to manage it.

By Minda Zetlin

Spring 2018

When Star Trek’s Captain Kirk wanted to take notes, he said, “Computer on. Record.” When Dave, the astronaut in Stanley Kubrick’s classic 2001: A Space Odyssey, wanted to get back inside his spaceship, he said, “Open the pod bay doors, Hal.” Even in the 1960s, when computers were giant machines operated by punch cards, science fiction writers knew that humans would be conversing with them if they could.

Now they can, and they are. More and more users are turning to automated chat as a convenient and sometimes fun means of accomplishing tasks or obtaining information. Last year, the messaging and chatbot company LivePerson surveyed 5,000 people about their experiences with chatbots. Among the respondents, 67% reported using chatbots for customer support and 14% said they’d used chatbots to help with productivity. About half of them had no opinion one way or the other about chatbots, but of those who did, 38% reported positive experiences while only 11% didn’t like them.

Clearly, chatbots are here to stay, and their uses will only proliferate. But as with any new technology, as chatbots gain in functionality throughout the enterprise, business and technology leaders are facing complex choices about how chatbots should be trained, managed, and deployed.

“Every decade or so, the way we use computers changes,” says Beerud Sheth, cofounder and CEO of chatbot development company Gupshup. “In the mid-1980s, we started using desktop software on personal computers. In the mid-1990s, we changed to the web—all software migrated onto websites. In the mid-2000s, it migrated to mobile apps. In the mid-2010s, we’re moving again, to conversational interfaces.”

When you get into the higher skill sets for bots, you get into security and access management issues and cybersecurity. So you have to have some governance-based automation in place. ”–Kaushik Iyengar, Senior Manager and Innovation and Product Development Lead, Deloitte

Chatbots are defined as computer software that simulates human conversation when interacting with users. Though they might not realize it, most Americans have interacted with chatbots, whether by using a company’s interactive voice response (IVR) system over the telephone; asking questions via a chat interface on a website; or talking with a voice-controlled intelligent assistant such as Alexa, Siri, Cortana, or the Google Assistant. These chatbots, like many others, are powered by AI. But there are also many useful chatbots that operate without AI by answering a preselected set of questions such as answering queries about customers’ bank balances.

Growing interest in using chatbots for all sorts of functions goes hand in hand with rapid advances in AI and natural-language programming, which enable chatbots to answer an ever-widening variety of questions, perform increasingly difficult tasks, and seem more and more natural as they communicate.

“You start out with a very simple set of predesigned questions and canned responses, the evolution from an automated email response,” says Kaushik Iyengar, senior manager and innovation and product development lead at Deloitte. More-sophisticated solutions can recognize parts of speech, know the history of interactions with a customer or user, and offer curated responses.

“Current research is focused on building out advisors that can understand and answer intelligent questions,” Iyengar says. “For example, ‘What are my taxes going to be for 2019?’ or ‘Prepare some life insurance options for me.’ That’s the further end of the spectrum in terms of complexity.”

Beyond Customer Service

The first known chatbot was a program called ELIZA created at MIT in 1966 as part of research on human/machine communication. It could simulate Rogerian psychotherapy (which consists of asking open-ended questions) and fooled at least one person into thinking he was conversing with the software’s creator, thus unintentionally meeting Alan Turing’s famous challenge to create a computer that could pass for human.

Since then, chatbot use has continued to grow, freeing up human workers in areas such as customer service and sales. Today, chatbots are popping up in all sorts of new places. Some of us might think of chatbots as a customer service tool that shows up within a chat window on the organization’s website or that fields customer calls in its IVR system. But chatbots are increasingly used for tasks from scheduling to collaboration to performing basic HR functions.

They’re operating outside the enterprise (and beyond enterprise websites) as well. Marketers now face a conundrum: more and more customers use mobile devices as their primary device to access the internet, and they are beginning to feel “app fatigue” that makes them reluctant to download yet another app just to interact with one organization. But with the massive growth in the adoption of messaging solutions, chatbots provide an obvious way to interact with customers where they are: Facebook Messenger in the US, WeChat in China, Line in Japan, WhatsApp (acquired by Facebook) in a variety of other markets, and Slack across workplaces. These messaging providers have made it easy to integrate their tools with those of chatbot vendors to deliver rich chatbot experiences to their users.

And then there are voice-activated digital assistants, such as Amazon’s Echo/Dot powered by Alexa, Apple’s Siri, Microsoft’s Cortana, and the Google Assistant used by Android and Google Home devices. (See “Raise Your Voice,” for a more in-depth look at voice technologies.) Your customers are communicating over all these channels, and so should you, says Suhas Uliyar, vice president of bots, AI, and mobile product management at Oracle. “You can’t dictate to customers where and how they interact with you, as you did in the past,” he notes.

Who’s Running the Show?

The proliferation of chatbots across channels, performing both internal and customer-facing functions, leads to an obvious question: Which department within an enterprise should be in charge of managing, maintaining, and deploying chatbots?

Some experts believe automated chat is growing so quickly in importance that there should be a new department within large organizations to manage it, as has happened with data in many enterprises. “Large organizations should start building those now,” says Rob May, founder of intelligent assistant software company Talla. Deloitte’s Iyengar says chatbots are “one other tool in the automation toolkit.” He thinks large organizations should consider creating centers of excellence around automation, charged with overseeing all automated functions.

Uliyar believes that enterprises should create their own departments for delivering engaging experiences across multiple channels: mobile, chatbots, and future immersive technologies such as augmented or virtual reality. He says it is critical that companies build centers of excellence for designing chatbots that can be used across multiple businesses.

Michael Fauscette, chief research officer at the software review site G2 Crowd, agrees. “I don’t think you can centralize platforms within IT the way we used to think about it,” he says. “But it’s still really important for IT to pick the right platform for the organization.”

You really have to think of them as if they were a new employee. Training a chatbot is quicker than training a person, but it still takes some time. But it’s well worth it.”–Rob May, Founder, Talla

As chatbots come into use for internal functions such as collaboration and HR, your organization will also need to consider concerns about privacy and information security. If an authenticated user asks for information that a chatbot can provide, should it always comply? In some cases, the answer will be no, and chatbots might not pick up on the nuances that might cause suspicion in a human employee.

“When you get into the higher skill sets for bots, you get into security and access management issues and cybersecurity,” Iyengar says. “So you have to have some governance-based automation in place.” He notes the need for a common infrastructure, with policies and procedures driven through a central function that includes IT.

IT should also be involved in crafting an overall strategy for chatbots within an organization. The lack of such a strategy is one of the biggest mistakes companies make in connection with chatbots, May says. Instead, many companies simply try out a few new functions to learn about chatbots with no particular objective in mind. Instead, he says, “you really have to think of [chatbots] as if they were a new employee.”

“Training a chatbot is quicker than training a person, but it still takes some time,” May adds. “But, it’s well worth it. They’ll save you time in the long run.”

Action Items

What Does it Take to Train a Chatbot?

How to Build a Better Chatbot

Watch this Webinar to explore the future of customer engagements with AI, chatbots, and mobile apps.

Illustration by Wes Rowell