SAN FRANCISCO -- Using chatbots for enterprise applications may be fairly commonplace, but for organizations not yet using them, creating or simply introducing chatbots into their systems can be a daunting task. To help ensure a successful chatbot strategy, best practices, as well as some potential issues, should be considered.
A key ingredient in the recipe for successful bots, and one of the earliest that should be added, is having the right chatbot team, according to Jordi Torras, CEO and co-founder of Inbenta, an AI vendor that specializes in developing and deploying bots and conversational user interfaces.
Make a team
At the least, an enterprise hoping to build its own chatbot should have a team comprised of a botmaster, a middleware specialist and a domain expert, Torras said, during a keynote on chatbot strategy at The AI Summit 2018.
"Middleware specialist," and perhaps "domain expert," are familiar job titles to organizations with technical expertise. "Botmaster," sometimes written as "bot master," is more native to the chatbot and botnet space.
With a team in place, an organization should set its expectations and key performance indicators (KPIs) for the planned chatbot.
Lay out a chatbot strategy, a plan, for how the bot will be used. Can the bot be integrated to back office systems? Will it have a web-based only deployment, or multichannel? What about the future -- will the chatbot use cases expand over time?
These are important questions, Torras noted. During workplace discussions, however, it's also good practice to be realistic about the limitations of the bot, he said.
"There's always going to be questions or conversations that are going to need a human to deal with," Torras said. So don't have the expectation that a bot will eliminate the need for human employees working, for example, in customer service.
When considering integration, an organization should evaluate its architecture, and think about the permissions it is willing to give the bot, Torras continued. A chatbot with limited integration is, in turn, "only going to perform basic functions," he said.
Test it out
For a successful chatbot strategy, users should do a lot of testing. Human languages are complex, Torras said, and effectively applying natural language processing (NLP) or even basic conversational capabilities is difficult.
Jordi TorrasCEO, Inbenta
"There are infinite ways to ask just one question and we cannot try them all," he said. "But, we can try some."
It's a good idea to also try using the chatbot in various channels. "Different channels might have different flavors, different experiences. Be sure to try them all," Torras said.
For organizations with a chatbot strategy that involves buying chatbot tools from a vendor, it's ideal to try different vendors to get a feel for the tools before committing, Torras said. Check for self-service and ease of use, and make sure that a technology's capabilities line up with what the organization needs, he advised.
To help get a feel for different types of bots, look some up and experiment with them, Torras said. He offered some examples of bots that are different, but work well, including the U.S. Army's Sgt. Star chatbot for recruiting, and the highly conversational Mitsuku bot.
Bringing good chatbot practices back to NZ
Felicity Midmer, a finance function and automation specialist at professional services provider PwC, and an attendee at Torras' keynote, had flown in from New Zealand for The AI Summit.
"We're starting to do a lot more automation work out there, so we wanted to find out what else is going on in the rest of the world," Midmer said.
While she and her co-workers went to numerous presentations at the conference, Torras' was particularly interesting as Midmer said she hasn't seen many successful bots in New Zealand.
"We're hoping to take some learnings from that and go to the next stage, not just as FAQs," Midmer said.
The AI Summit was held Sept. 19 to 20 at The Palace of Fine Arts.