Artificial Intelligence

What Does it Take to Train a Chatbot?

Five skill sets for training effective bots

By Minda Zetlin

Spring 2018

Most experts agree that creating an effective chatbot requires collaboration among different functions with different areas of expertise. Here are just a few of the skill sets that go into building a chatbot:

Scripting. Most chatbots can be built without any coding but do require some software development skills. “There is a certain level of customization that occurs where you need a developer,” says Suhas Uliyar, vice president, bots, AI, and mobile product management at Oracle. “Not classic developer skills, but someone adept at scripting, as with JavaScript, Ruby, or Python for the more complex use cases.”

Subject matter expertise. These developers will need to work side by side with those who have specific expertise in the subject matter of the chatbot. Ideally, they should get some help from a human whose interactions the chatbot might replace, such as a customer service representative with lots of experience talking to customers.

Script-writing. Quite distinctive from the computer scripting expertise needed to organize chatbot information effectively, chatbot creation also requires plain old-fashioned dialogue-writing skills. After all, even people who are perfectly aware they’re interacting with an algorithm tend to think of chatbots as having personalities and even feelings.

“Give it a tone, perhaps a sense of humor consistent with the voice of your brand,” says Beerud Sheth, cofounder and CEO of chatbot development company Gupshup. “You would expect an HR chatbot to be more sensitive and a marketing chatbot to be more creative. This is right out of Hollywood scriptwriting and draws on the same skills.”

Data management. “Without integrating data, the chatbot is pretty useless,” Uliyar notes. “How do you enable that mashup of services from the data layer perspective?”

Beyond using data to answer user inquiries, all AI-driven software needs data to learn and improve. “One data scientist I know says we could have done everything we’re doing now 10 or 15 years ago, only we didn’t have the data,” says Michael Fauscette, chief research officer at the software review site G2 Crowd.

User experience. “This is just like any other user experience that you design,” Fauscette says. To create an effective chatbot, build in what your organization has learned about user experience.

“We spend a lot of time optimizing websites—what colors should we use? Where should we put the buttons? Because chatbots are out in the wild, they tend to not be under the control of the UX [user experience] organization.” But you need UX expertise to build an effective chatbot. “You’re still trying to optimize the experience for the outcomes you need to get.”

Action Items

How to Train Your Chatbot

How to Build a Better Chatbot

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

Photography by Nicolas Thomas, Unsplash