Evernote founder Phil Libin is betting big on conversational bots. He left the now-famous startup he launched in 2007 last year to join venture capital firm General Catalyst, and he sees conversational user interfaces becoming a game changer on the level of the iPhone.
“In 2007, I had this vision when I first touched my very first iPhone where I kind of understood what the next five years would bring, and I haven’t had that kind of clarity since,” said Libin in an interview with Business Insider. “And now, I have the same kind of feeling about bots, about conversational UIs.”
Such technology could revolutionize customer service and upend contact center opertions. If chat-based bots take over, the need for human representatives would disappear over night, allowing companies to cut out their most unpredictable cost and, at times, source of frustration.
Mayur Anadkat of Five9 has heard this before — many times — during his 15 years in the industry. But the concept of full automation through artificial intelligence is something he sees as a non-existent Holy Grail that the industry continues to chase, pouring untold millions and countless hours into a dream that will never come to fruition.
Anadkat, who is the San Ramon, California-based company’s VP of product marketing, recounts the past of interactive voice response (IVR), artificial intelligence capable of responding to emails, and other flawed solutions as a warning for those looking for a Holy Grail of automation. “We went through this phenomenon with IVRs,” he said. “We also went through this phenomenon to a certain degree during the phase when everyone was doing web avatars — automated chat robots. And IVRs still represent — to a large part of the population — the most frustrating part of the call center industry.”
He recognizes that the technology is getting better and better. Some automation goals that industry companies are seeking are reachable. But it is simply hard to make a cost-savings argument for many of the realistic options since the investment — on top of all the ongoing costs for data engineers, speech scientists, and AI gurus to maintain systems — are so much more expensive than phone reps.
“The further we run down the road with IVRs and speech automation and natural-language understanding, the more and more it became cost-prohibitive,” said Anadkat. “It’s not that it’s not achievable. But you’re taking money out of one bucket, which is the idea that you’re saving on full-time employees in your call center, and you’re putting it into a data-scientist, speech scientist, artificial intelligence scientist role, which becomes very expensive and hard to maintain as technology evolves.”
Even the best systems, those multi-million-dollar AI engines that can understand free-form voice or text, still requiring ongoing investments. One aspect of having a functional system is ensuring that “yes” and “yup” and “yea” and “sure” and “OK” and “fine” are all interpreted as the same affirmative response. But this process is never fully complete due to the evolution of language and the experience that companies get as they receive newer and stranger complaints. And the pathways to reach the right solution must continually be updated, particularly as the company beings offering new products and services.
“It becomes cost prohibitive in that you can’t package up all the possible questions and answers and then not expect it to evolve,” said Anadkat. “And that process of evolution you’re taking becomes very expensive because it usually requires new technologies and Ph.D.-level resources. You’re going to have to do that every year to keep it relevant to your customers.”
Neither Anadkat nor his company are against the use of new, smarter tech tools to improve operations. In 2013, Five9 acquired a company called SoCoCare that does natural-language processing for text channels. It processes voice calls and has an artificial intelligence component that then suggests a “next best action” for the call. But even though it is a dynamic system, sometimes one of those options is comically far off from anything that would help the customer. So it is up to agent to make that choice, and Anadkat believes this component is vital to providing proper support.
“If you left it to artificial intelligence … that error is so much worse than asking the question again through an agent.” said Anadkat, noting that “customer experience is way too important to make an error of automation and lose a customer just based on savings at the call center.”
This view has been growing, slowly but surely, over the past decade. There was a time when companies viewed customer service as a necessary evil on the expense sheet. It was like accounting or even the light bill — not core to the business and a cost to be brought down as close to zero as possible.
That is changing. Even the evolution of terminology — from customer service to customer care to customer experience — reflects the mindset shift. Now, mobile phone service providers, for example, are realizing that their business is largely providing a commodity and their customer service can set them apart. With bill pay, and virtually everything else, being done online, the occasional call to a service rep to handle a problem is one of the few actual interactions the customer will ever have with the brand. So it’s more important for that exchange to go well than for it to be completed for $1 less than it was last year.
“Social media was a big part of it,” said Anadkat. “It informed large companies to their fear that they are no longer in charge of the conversation they have with their customer base. They can’t dictate all the terms.”
That is why Five9 and others have abandoned the idea that automation can replace the agent and instead begun using it to augment their ability to provide service. Automation is here to enhance, not supplant, the representative’s ability to serve the client. “The way we use it is to arm the agents — it’s all additional golden nuggets that give the agents a better understanding and more information before they start engaging,” said Anadkat. “It’s never intended to be the full answer … There is just too much to risk.”