Innovation

The business value of chatbots: What tech leaders really need to know

At Mobile World Congress Americas, a panel of chatbot experts from Twitter, Facebook, IBM, and more discussed where the tech stands and where it needs to go.

Chatbots have gained a lot of hype in the past couple of years, but many businesses still struggle to figure out how to integrate them for employee and customer support. At Mobile World Congress Americas in San Francisco on Wednesday, a panel of chatbot experts met to discuss where this technology currently stands, and where it needs to go to have a true impact on business operations and revenues.

The panel included:

Anand Chandrasekaran, director of platform and product partnerships, Facebook Messenger

Beerud Sheth, founder and CEO, GupShup

Ram Menon, founder and CEO, Avaamo

Rob High, IBM Fellow, vice president, and CTO, IBM Watson

Robin Wheeler, director of sales, tech and telecom, Twitter

Despite the hype, the panel agreed that "it's still day one in the conversational world," Menon said. "The first wave of chatbots were basically a glorified search engine. What we should expect over the next three to five years is for a conversational UI to be able to perform judgement-intensive tasks."

The first generation of chatbots can respond to requests such as "Play Taylor Swift," or "Tell me the weather," Menon said, but people don't actually speak like that. "The future is about having a real dialogue, and the chatbot remembering what you said last time and responding. That's where the AI comes in," he said.

SEE: Special report: How to implement AI and machine learning (free PDF)

In the enterprise, "it boils down to 'Can you make me money or save me money?'" Menon said. Across Avaamo, a platform designed to help enterprises create and deploy conversational AI for use by employees and customers, 95% of interactions occur outside of Facebook or Twitter, on private networks. The company works with a network operator that uses a chatbot to schedule maintenance, and another who uses it to handle commission for salespeople. A large insurance company uses chatbots to provide 11,000 insurance quotes per day.

Many network operators and other enterprises have their own portals or mobile apps that they have spent millions of dollars promoting, and want chatbot interactions to occur in those channels, rather than on social media sites, Menon said.

A typical network operator receives six to 10 million calls per quarter, Menon said. About 50% of those calls are somewhat repetitive, perhaps involving customer service issues. "If a network operator looks at this from a pure end user automation perspective, if they can deflect 25% of those calls, they make money," Menon said.

Business cases for chatbots are typically formed around the assumption that these assistants will offload 30% of human work, High said. While customers go to a chatbot with a basic need, such as "What's my account balance?", it may solve an immediate problem, but does not get to the real question, which may be "Can I afford to buy this car?" or "Do I have money in the right account to pay my bills this month?"

"If a conversation agent can bring a user into the depth of the real problem, that will create more engaging relationships," High said.

One challenge is that customers often express the same question in different ways, Menon said. For example, when working with a city garbage operator, he found that to express that the garbage had not been picked up on a certain day, some people would call in and say "The bins are still full," while another would say "The truck didn't show," and so on. "The challenge at the front end is to understand that implicit intent which the human being gets," Menon said. "On the back end, it's a judgement-intensive task."

For example, if someone calls into a phone carrier and says "My phone isn't working," it would be helpful to the carrier if that chatbot can learn why and whether or not to try and sell that person a new phone. "How AI can help a chatbot to make a judgement and respond appropriately is really where the future is," Menon said.

It's important to have a human backup for customers that need additional help or have a customer request, Wheeler said. You can feed that information into the bot as well to drive its NLP improvement.

One operator that verifies IDs has customers upload the ID to the bot, which then seamlessly hands it over to a human agent for verification, Menon said. "A lot of people are separating out clearly that the part of the business process that is repetitive can be done by the bot and seamlessly moved to the human," he added.

Facebook Messenger is proving to be a fertile ground for businesses to integrate chatbots, Chandrasekaran said. The platform was originally intended as a way for people to contact their Facebook friends, Chandrasekaran said. "That behavior is evolving—people are messaging businesses the same way they message friends."

Facebook currently has about 70 million business pages in operation. Of those, between 20 and 25 million are messaging their customers back and forth each day, Chandrasekaran said. "That means that customers now have an expectation that if they message a business or brand, they are expecting to hear back," he said. "We're past the early stages, but there's a lot of room for growth."

T-Mobile adopted social customer care and activation via Facebook—if you purchase a new T-Mobile device, you can activate it with a couple of questions on messenger, instead of a 20 minute phone call, Chandrasekaran said.

Over time, for chatbots to grow more sophisticated, they will inevitably need to communicate with each other, Sheth said. "It's inconceivable to imagine one chatbot that handles all possible use cases," he added. "There's going to be a need where instead of individual intelligence, you have collective intelligence."

The next step in chatbot evolution will involve moving from a single-turn interaction to something more conversational and engaging, High said. "Where there's a lagging for me is in the developer experience, in enabling the methodologies and the techniques for creating those rich conversation agents beyond that simple single-turn interchange," High said. "There's a lot more to be done there, understanding how to take a natural conversation for human beings and map that into an AI that's able to recognize the kinds of variation that occur in those conversations, and be able to respond in a meaningful way with the right level of reasoning and judgement."

istock-667050642.jpg
Image: iStockphoto/Natali_Mis

Also see

About Alison DeNisco Rayome

Alison DeNisco Rayome is a Staff Writer for TechRepublic. She covers CXO, cybersecurity, and the convergence of tech and the workplace.

Editor's Picks

Free Newsletters, In your Inbox