In the 2004 sci-fi movie “I, Robot,” detective Del Spooner (played by Will Smith) recalls a moment when a robot needed to provide the ultimate customer support. An accident had caused two cars to sink in a river, and the robot was tasked with saving either Spooner in one car or a girl, Sarah, in another.
The robot chose Spooner.
“I was the logical choice,” Spooner says, angrily. “It calculated that I had a 45 percent chance of survival. Sarah only had an 11 percent chance. That was somebody’s baby. Eleven percent is more than enough. A human being would’ve known that.”
While we’re still a ways from actual robots physically helping us solve our problems, we already have messaging bots inside customer support departments today. Rather than dialing the customer support hotline, customers can fire up a messaging app on their mobile device and interact with a bot. As these bots become smarter, moving from rules-based logic to machine learning, there will be less need for human agents.
Unlike in “I, Robot,” this can be a good thing. When someone has a problem, they don’t want to be caught up in a phone tree or bounced from agent to agent only to have to describe the problem again and again. People in the digital age expect quick and accurate answers to their problems, and messaging bots provide this.
There’s been a lot of chatter around customer support messaging bots lately. Kristina Shen, vice president at Bessemer Venture Partners, began tracking tech vendors in this space nine months ago. She found 20 companies. Today, there are hundreds of companies on her tracking list, she says.
Moreover, a company’s balance sheet stands to prosper from customer support messaging bots. There are millions of contact center agents around the world trying to answer questions from billions of people at an enormous payroll cost to companies. A bot has the potential to reduce the number of agents dramatically.
Consider the case of a utility company in Washington D.C. serving 1.4 million customers: It ran a customer support test pitting a human agent using the phone against another human agent using messaging. Not surprisingly, the phone agent completed only 10 resolutions per hour, while the messaging agent had 15 concurrent conversations going every 10 minutes. In terms of efficiency, messaging kills the phone.
Here’s where it gets interesting, if you’re into bots.
The messaging agent used a Word doc with answers to common questions and responded to customers by copying and pasting answers into messaging windows. An automated bot can do this easily. What about questions that aren’t in the Word doc? A bot’s natural language process system could watch and learn how the human agent responds, so that the bot can answer the same question in the future.
“It makes you wonder, what point does that margin get squeezed to the point where no humans are involved?” says PypeStream CEO Richard Smullen, which helps the utility reduce the number of human agents through messaging.
Conversely, companies might not want to use messaging bots in customer support roles. Brands want to build deeper digital relationships with customers that lead to cross-selling and upselling. If interactions get shunted off to a self-service system or bot, these opportunities disappear, says Forrester analyst Ian Jacobs.
“We’re a long way from the contact center agent apocalypse,” Jacobs says. “The bot experience is still generally very poor and the tasks they handle are very simple.”
So bots can handle simple queries, but more complex situations require a human response. While technology has the ability to determine who should respond—bot or human—the handover can be tricky. “Having a human respond at the same speed that a bot was responding is incredibly hard,” says Jason Smale, director of product strategy at Zendesk.
One way around this, a human agent can monitor chats between customers and bots and jump in when needed. This is a better approach than attempting a handover when a bot fails to a human agent who comes to the chat lacking context, says Mikhail Naumov, president and CSO of DigitalGenius, which builds machine-learning chatbots.
Then there’s the emotionally charged nature of a customer support call that doesn’t play well with bots. When someone has a problem, anger can quickly bubble to the surface. It wasn’t long ago when people would get stuck in phone trees, scream into their phones, and pound the zero key in hopes of hearing a human voice.
Like the movie “I, Robot,” there will be times when a human is needed to deliver an empathetic response in order to calm an irate customer or make a decision that’s not strictly dependent on a logic tree. At the end of “I, Robot” (spoiler alert!), a robot was again going to make a cold, calculated choice, until Spooner ordered it otherwise.
“When you get into areas that require real trust—the ‘look you in the eye’ sort of stuff—a bot won’t cut it,” Forrester’s Jacobs says. “If I am worrying about my health or my wealth and having a major problem with either, I want a real person there to help me.”
(Editor’s note: Kristina Shen of Bessemer Venture Partners, Richard Smullen of PypeStream, Jason Smale of Zendesk and Mikhail Naumov of DigitalGenius were speaking on a panel at MobileBeat 2016 in San Francisco last week.)
Tom Kaneshige is editor of Five2ndWindow, Penton’s independent news site helping marketers and line-of-business executives get ahead of the mobile disruption happening to the customer experience. You can reach him at email@example.com.