Earlier this year there was a revolution: we launched our ‘single voiceprint across channels’ capability. , Until today, I get questions about how we achieved it, and why NICE is the only vendor offering that capability. People, it seems, are as curious as they are cautious, as their questions often revolve around whether or not there is a catch.
To put it clearly and concisely, there is no catch. NICE Real-Time Authentication uses one single voiceprint for authentication across multiple channels. The system creates the voiceprint based on a conversation between a customer and agent and can use the VERY SAME voiceprint for authentication via the IVR, mobile app or any other channel that allows voice capturing. See, no catch!
To make it as clear as possible, I had a 1:1 chat with Nexidia’s Head of Research who led this technology breakthrough, Dr. Jon Arrowood.
Jon, can you give us some background on the single voiceprint development?
Jon: When we look at customer voice authentication in the contact center, organizations use Voice Biometrics in two manners – passive authentication, meaning that the customer is authenticated during a natural conversation with an agent, or active authentication, which is normally used in the IVR. You probably know that - when you are asked to say a specific phrase to verify your identity.
The process of creating a voiceprint for someone is called enrollment. If you use passive voice biometrics, the enrollment is done automatically without you noticing it. But when using active voice biometrics, the end-customer needs to actively enroll, so they need to repeat the same phrase over and over again, for 3 times when talking to the IVR.
3 times? I usually hang up the phone if I am asked to repeat the same thing twice!
And your behavior is very common! The fact that customers have to actively enroll is a real barrier to adoption. As an example, a friend of mine was calling their brokerage recently. And as you can guess, they wanted to get to an agent pretty quick. But the company had it set up to get you to do an active enrollment first. He was in a hurry, and didn't really want to do it. Well, the company knows that if they don't get it now, you're probably not going to call in later when you finally have the time -- because you never will!
This is just an example of the dilemma companies face: if given the option, customers don't enroll, but if the option is forced on them, their all-important Net Promoter Score (NPS) goes down. In my friend's case, the company chose the latter approach, and now they've taken something that could have been positive experience (better security, faster authentication…) and turned it into something negative and frustrating for him.
Let's say they had instead taken the other approach of making it optional. He wouldn't have done it! And how can an organization gain value from voice biometrics if their customers don't enroll? It's a lose-lose situation.
To go back to our active enrollment story: it doesn’t really catch consumers and only some of them enroll, which is a real barrier to organizations that want to apply voice biometrics and gain value from it. If they have this technology in place but their customers don’t use it, then they don’t achieve anything and it all becomes pointless.
So they can have the best technology in place, but it is worth nothing if their customers aren’t using it?
Exactly. Now, the single voiceprint capability addresses exactly this challenge. In practice, it allows using the same voiceprint for multiple channels, without a need for active enrollment.
So with this development, you basically allow passive enrollment for self-service channels?
Yes, that’s exactly what we do. We use voiceprints that are created passively based on a customer’s call with an agent for voice authentication when calling the IVR and other self-service channels.
It sounds almost impossible. Can you shed some light on how the technology works?
It’s all based on machine learning, and huge amounts of data that we have access to. The fact that we can use more than just one call per customer for enrollment, allows us to create one voiceprint, that is so rich and accurate, that we can use it across multiple channels, both for passive and active voice authentication.
Did you base your research & development on certain principles or previously existing technology?
Actually, we did. We used a brand new technology, straight out from the academia for the core algorithms, and we added machine learning. Yet, all this technology would be useless if we didn’t have those huge amounts of data. These machine learning algorithms can only work if they have a lot of data, and that’s what really helped us get to this breakthrough.
How come no one else did it before?
To be honest, I don’t think that anyone else has access to that much data. NICE and Nexidia together have the largest amount of contact center data in the industry. And partnering with our customers is what allowed us to get to such accuracy and to the single voiceprint capability.
So it’s mainly the data that made the difference?
It’s not just the data. The development also required a lot of application design, which we could easily do with the NICE recording system. For instance, we get much better results if we have more than one call of the same individual. This is a non-issue when voice authentication is based on the Recording platform.
During your research work, were there any moments you thought it would be impossible to successfully achieve the passive enrollment goal?
Absolutely! We have worked in the area of speaker authentication a lot in the past. 12 years ago already, we started working on such short segments of speech. I genuinely didn’t think you can do it.
The new algorithms published from the academia provided a strong breakthrough, and then the improvements we implemented based on machine learning gave our work an additional flavor. And then, because practice makes perfect, we really benefitted from our access to the data to train really good models…no one in our group expected it to perform that well!
What is the next technology you are working on?
Well, the hip thing is currently artificial intelligence for fraud and authentication, better call routing, and completing more actions in self-service. We focus on creating technologies that will impact millions of people and make for the best possible customer experience. For that, you need smooth flow and seamless authentication to be as simple as possible, and that’s what we are working on.
Interested to hear more from Jon? Check out this video where he explains more about the single voiceprint.