Using Natural Language in your Contact Center
"Press 1 to access your balance, 2 to place an order, 3 to report a lost card, 4 to ..." How many times have we had to listen to a long list of options, waiting for our need to be mentioned, wondering if that would ever happen, then pressed # to listen to the menu again and finally in desperation pressed 0 hoping that this would transfer us to an operator? What a waste of time and stress multiplier.
And see it from the other side. Even though we have many channels to contact us, voice is still going strong. Callers stay on the IVR for a long time. 800 numbers are cheap these days, but the high volume makes even a few seconds more per call count. And what should we do with these pesky zeros coming in? Disable the shortcut? After all, the call can go to an agent who doesn't have the right skills and so it may need to be transferred again. Bad user experience, and waste of money. Not to mention all the callers who get frustrated and abandon the call. They are unhappy, and we pay a toll-free call for nothing.
There must be a better way.
Enter natural language processing and virtual agents. The technology is maturing at an accelerated pace, with consumer services - Amazon's Alexa, Google Home, Microsoft's Cortana - understanding what people say and usually able to provide answers. If it could be translated into the CX world, it would mean that customers calling into the Contact Center would just say what they need. In their own words, using their own expressions, naturally. This would cut both their frustration - providing a truly great CX, and the company expenses - the call would be distributed to the right queue quickly, with the right metadata, and with much fewer abandoned calls.
I recently tried it out with Google. I said: "OK Google, what is the best mortgage refinancing rate today?" What I got was "My apologies, I don't understand". I am not surprised: I don't expect that Google or other cloud-based, consumer-oriented natural language processors will have an answer to this type of very specific questions anytime soon. They don’t have access to the right data, which is hidden in the banks’ intranet, and the AI may also not be trained for it since it’s a question that’s not asked often.
For a CX application the understanding success rate must be much higher than for a consumer service. But the good news is that when someone calls into your Contact Center they are not going to ask what the weather will be like tomorrow. Instead (in 99.999% of the cases) they will ask for something related with what you do. So, the natural language engine works on a domain that needs only cover your products and services, and can be structured to mirror the services that your Contact Center offers to your customers already – and nothing else needs to change behind it.
In addition, the system can be flexible within the limited domain. If it does not understand the caller’s intent the first time, the system is able to ask a follow-up question, possibly with a short list of alternatives based on what it did understand. For instance, after the system knows that the caller wants something related with "my account" there are only a few services that can be provided and so the system can offer a list of options. It is very rare that the exchange should go to more than 3 levels - back-and-forth questions and answers. This makes for a very natural, fast, efficient and delightful experience for callers and companies alike.
Recently I joined Interactive Media, a company that specializes in creating the experience of natural language interactions, just like the ones I described. Interactive Media (http://imnet.com) is based in Italy and is having a remarkable success there, thanks to their semantic methods that substantially cut the time and effort needed to match the callers’ interactions with the service structure.
We are now offering the same for the American market, and I would be delighted to talk about it with you if you are interested.