In the age of Alexa and Siri, how come most customer-service voice assistants still suck?
We’ve all been there. You urgently call your bank to ask about an unrecognized withdrawal or unexpected overdraft fee and the voice assistant asks you to repeat your name again and again, as well as your account number, as well as simple words like “representative” or “checking account.” You angrily hang up, slam down your phone, look across the room, and see your kid happily chatting away with Alexa.
Why does it seem like so many customer service voice assistants still do a terrible job of understanding basic English? Why haven’t these systems progressed, despite the fact that they’ve been in use at least 10 years, and despite the fact that we’re in a voice-computing renaissance exemplified by Amazon Echos and Google Home devices? There are some exceptions, but almost all Americans recently surveyed said they’ve experienced frustration when trying to communicate with a company using Interactive Voice Response (IVR) systems. More than half of them hate having to repeat themselves and almost half didn’t like having to wait to talk to a live agent.
The discrepancy in quality can be tied to a number of factors, including the difference in technologies, an over-reliance on legacy systems that have not developed their technology, long-term contracts with vendors that are hard to break, and the research commitment to innovation at tech behemoths like Amazon and Google. In terms of voice recognition and natural response, Alexa, Siri, Cortana, and Google Assistant are “leaps and bounds better than anything the telecom industry has produced vis-à-vis interactive voice response (IVR),” says Tom Roberto of Core Technology Solutions.
Obviously, the virtual customer assistants developed for companies are designed to handle simple questions and interactions, while home voice assistants like Siri and Alexa have been designed for a broad range of topics and are tied to massive databases, notes Annette Jump, a senior director analyst at Gartner. “It’s still quite the early days in developing the technology of these virtual assistants,” she says, claiming that there have been a lot of advances in the last 12-18 months, but that some of them haven’t yet been adopted by vendors.
One key difference is that while home voice assistants are connected to the internet, the customer-service technologies also need to be linked to internal systems at individual companies, says Jump. “And depending on how well that linkage has been done, that assistant is able to do so much more or very little.”
“In my gut, I think it’s because no one’s better than Amazon and Apple and some of these vendors have large development teams but they’re not willing to spend to make that leap in quality,” says Micah Solomon, a customer experience consultant, who lays the blame at CEOs who don’t prioritize customer service. “Executives almost all say that the customer experience is very important, but if they want to prove that, they need to improve it. They’re not focusing on what their customers really want, they don’t understand how important it is for customers.”
There are a few exceptions, notes Solomon, pointing out Domino’s Dom, a version of Siri that was developed for them by Nuance, one of the top vendors in the IVR space. It’s a personable voice-ordering platform that was launched by the pizza chain back in 2014.
Sorry, that’s an invalid response
Two of the biggest headaches for consumers is having to repeat your name or account number to a live agent after already typing it in or saying it to the IVR system and being rerouted to different agents who all ask you to verify your identity over and over again. That might be due to extra security precautions, explains Solomon, but it could also just be poor user interface. To address those problems, vendors are developing innovations like Next-OS, which provides each agent with a 365-degree view of the customer and sentiment analysis which assesses the mood of the customer and guides the agents’ responses.
Further advancements in artificial intelligence, such as natural language processing, will improve the quality of such systems, says Mari Ostendorf, a professor of electrical engineering at the University of Washington and a leading scientist on speech and language technology. She notes that one of the reasons that home voice assistants are so advanced is because they’re expecting children to use them, and since their voices are so different, Google and Amazon and Apple have spent years trying to do better, investing millions to use billions of interactions to further improve the recognition model, and they’re under intense competitive pressure to be constantly updating.
That pressure is lacking when it comes to most customer service IVR vendors, who haven’t been investing in the same way. And the tech giants’ developers who do speech recognition “are very visible in the research community, contributing to studies,” and thus they’re able to keep advancing the technology, notes Ostendorf.
At the highest level, the competition keeps heating up. Alibaba showed off its AI customer service agent for its logistics company Cainiano at the 2018 Neural Information Processing Systems conference, demonstrating how it effortlessly was able to handle millions of complex requests a day from customers. In a 30-second pre-recorded demo call, the agent “smoothly handled three common, and tricky, conversational ingredients: interruption, nonlinear conversation, and implicit intent,” reports MIT Technology Review. Though those are routine elements of an everyday human conversation, they’re too complex for most voice assistants. Alibaba is also prepping a virtual agent to take food orders, even in noisy restaurants and stores and a “price-haggling chatbot that is already used by 20% of sellers on Alibaba’s resale platform.”
Those advancements might be coming soon. But in the meantime, in the time it took you to read this article, I’m usually on hold with UPS, desperately trying to figure out the right way to connect to a live agent before the IVR hangs up on me. Help!