If you listen to the examples of Google phone calls to a restaurant and a hair salon that are posted on the Google AI Blog, you’ll be surprised by how natural the conversations sound. The humans on the other end of the call are apparently unaware that they’re talking to a machine. The exchange of making an appointment goes off as it should.
If you’re designing an IVR (interactive voice response) system, that’s exactly what you’re looking for. But it’s worth noting that artificial intelligence-driven IVR systems are not new. What’s happened is that Google has developed one that works better than most. But while it’s better in terms of its flexibility and its level of understanding than what’s out there, it’s not unique.
In fact, Google Duplex isn’t even the first outgoing IVR system that’s AI based. Robocallers have using such systems for several years now, and they’re used in multiple ways. For example, there are several polling companies that will ask you a series of questions on your political feelings using very similar-sounding technology when they ask if you’re a Democrat or a Republican and then move on to the next question.
Further reading
Some of the more sophisticated of those IVR systems even insert fake typing sounds into the silences so that it seems as if you’re talking to someone who is actually recording the answers. Similar IVR systems are populating call centers, usually acting as a first line of customer service. If you want to know how well it works, try calling Verizon customer service, where such a system has been in use for a while now.
What’s really changed with Google Duplex is that the voice response has become marginally better and slightly more human. The competence of the AI that operates in the background has also improved, making responses to human interaction more natural. The restaurant call provided by Google in which the human has a poor command of English is a good example. The artificial intelligence behind the voice was able to adapt.
While Google’s technology was in a very early stage when it was demonstrated at Google I/O this past week, it gives you a good idea of where it’s headed. Right now, Google is focusing on using Duplex with Google Assistant, which works in its Google Home speaker as well as a number of other devices ranging from mobile phones to television sets.
But this same technology is also on the way to your call center, whether it’s from Google or some other company. The IVR systems that exist in today’s call centers have limitations that make them less effective than they might be. The limitations mostly revolve around insufficient programming to handle input from callers. You’ve probably experienced that when you’ve called your cable company.
The insufficient programming means that whoever set up the IVR system didn’t include a means of handling a response outside of what’s expected. The easiest way to handle that is to include a response such as “Operator” or “Agent” in the list of accepted responses so that a non-standard response goes to a human operator. When that option for getting out of the IVR menu tree doesn’t exist, the response is frequently angry callers because you’ve wasted their time without helping them.
A well-designed, AI-based IVR system will have real potential as a customer service tool when it can use machine learning to handle non-standard or unexpected input from callers or call recipients. For example, I used to have fun with those automated pollsters by answering “No” when it asked my political party. Now, sadly, they’ve learned to adapt.
But for your call center, whether it’s handling incoming calls or placing calls to customers, the ability to adapt to unplanned responses is a gold mine. You have a way to collect useful responses from customers, which will help you improve satisfaction, while also improving your IVR system.
Doing it right includes disclosing the fact that you’re using an IVR system and making sure you’ve thought out your programming so that callers don’t get stuck in a menu loop. This includes providing access to a live operator when it becomes necessary to handle unexpected circumstances, which will do a lot to improve your company’s image in customers’ minds.
Do it wrong and all you’ll do is annoy your customers as too many IVR systems do already. You will only do the one thing that you don’t want, which is to have your customers hang up the phone in frustration. It’s important that the goal of any sort of telephone interaction is to engage your customers, not to alienate them. A good IVR system can make your customers like you and having AI in the background makes that more likely.
For reasons that remain obscure, Google Duplex was seen negatively by some attendees at the Google I/O conference at company headquarters in Mountain View, Calif.
Apparently these people were unaware of the current state of IVR, and were uncomfortable when they heard a good system for the first time. This speaks more to the lack of sophistication on the part of those observers than on the state of AI. Had the same demonstration taken place at a telephony show, all you would have heard would have been cheers of delight.
