Loraine Lawson:
Good day and welcome to Bank Automation’s podcast series, The Buzz. I’m Editor Loraine Lawson, and recently I spoke with Amit (AHmit) Gupta, who is the vice president of product management, research and engineering at Pindrop, a voice-based security, identity, and intelligence platform. The vendor recently conducted a survey that found 14% of people would rather be stuck in traffic at rush hour than reset their passwords. I asked Mr. Gupta how voice data can address these security pain points.

Amit Gupta
Yes, we absolutely found that out. A lot of the businesses, if you look at when the users are interacting with those businesses through the phone channel, rely on asking them certain questions before they can even address or listen to the issue that that user is calling about. And that’s the reason why I’d get frustrating on top of it. 30 to 60% of the users are not able to answer those questions correctly, while 40 to 60% of the fraudsters can bypass those questions. So essentially, those organizations are adding the friction to solve a problem that this is not even a solution to solve that problem for. And that’s where voice absolutely helps. But more than that, an enhanced authentication solution can help alleviate that pain and that bad customer experience. And in fact, convert that to a great user experience. And I’ll give you an example, the enhanced authentication platform that that Pindrop provides. It leverages voice for sure as one of the factors, but it’s a multi factor authentication platform, it looks at various different attributes we received from the audio itself. To do the device authentication, we leverage the voice attributes to perform a voice authentication, which is extremely passive. And we don’t require any friction to be added, as well as on the call channels when the user is interacting with the IVR [Editor’s note: Interactive voice response] and pressing keys to answer certain questions in the IVR, we are able to do behavioral analysis as well. And we also have spoof detection and fraud risk assessment built into the plan, or built into the platform. So when our customers leverage all of these, all of these different factors together to start assessing the risk and the confidence for the incoming callers, very, very quickly, in the call, they know whether the caller ID is valid, or it is being spoofed. Very early in the call, they know whether the fraud risk assessment, or fraud risk here is high or low. On top of it with very little speech, we are able to tell them, is it the same voice. And with the audio attributes, we are also able to tell them is it the same device, all of that comes together to create an authentication profile for the user that our customers can consume extremely quickly, in the first few seconds of the call, do now make a decision on how much friction is really required to be added. And a lot of our customers find that over 90% of the calls can be offered a lot lower friction than what they’re doing today. It’s pretty much the top one person risky calls where they may have to improve or increase the friction, and rest of them can just go through the standard friction. And with that, they start providing a very good customer experience to majority of their callers on the call channel. Who you know, again, the survey wasn’t surprising who don’t want to answer those questions. On top of it. What we hear back from our customers is it’s not just a customer who gets excited with this passive authentication experience, even the call center agent, they don’t like to ask those questions. They don’t want to ask those questions. They actually understand that they are just adding friction to the user by asking those questions. A lot of times I have personally experienced those contact center agents apologizing to me saying I’m sorry, but I do have to authenticate you before I can address your concern. But they also have an extremely pleasant experience in the corner. Decisions completely change where they can be a trusted partner for the user and the businesses just thrive by providing that better experience. So it is more about a passive way of enrolling a passive way of authenticating that allow the users to naturally engage with the IVRS. Or with the contact center agents, one of our customers said, It’s too bad that you’re making them talk to a machine at least make the experience human like, and that’s the customer who enabled only one question in their IVR that says, How can I help you? And that’s a question that everyone would love to be asked. And the more important thing is, once the answer is provided, we our technology enables the organization to start servicing the collaborator.

Loraine Lawson
So can you give us a little peek, we’re all about the technology here, a little peek under the hood, obviously, nothing proprietary, unless you want to, but just a little peek at understanding how this works.

Amit Gupta
Absolutely. And I touched on that a little bit in terms of the different factors that we we assess, if you will, when the calls come into the platform. So but essentially do dig a little bit deeper there. When the users call into any contact center, any large organization these days have an IVR that IVR, maybe natural language enabled IVR, where you can speak your responses, or it may be a no voice IVR where you have to key in the digits to actually answer the questions there. But many of these ideas have some sort of questions to be asked answered first before they can even understand how to route that call, how to address that issue of the user. And what Pindrop is able to do is we are able to assess all the different attributes that come into these ways channels. So for example, when a call is made, the audio attributes in the call can be very different based on the device type, or the carrier or the combination of these two that has been used. The what we found what our research organization found was that the key presses that come in the tones and the attributes around that that come in, we created a patented technology called tone printing, which is very unique to a device statement or carrier. So with all of this assessment, even without the voice itself, we can create a very strong multimodal device profile for a user on top of it, once the user starts speaking, we are able to assess the voice features. And we are not necessarily listening to what the user is saying. These voice features that we extract, they’re just arrays of floating numbers completely irreversible to what the actual speech was completely language independent, we don’t have any friction in terms of that they have to have certain passphrases, the user can literally be talking naturally. And we are able to assess that voice behind the scenes completely passively and create a voice profile for that user. When you call financial institutions typically, and even some utility companies have this, they make you dial in your user ID or unfortunately, many of them still have a social security number that you have to enter or your phone number that you’re entering. When you’re pressing those keys, the human brain is wired to press those keys, the patterns around those key presses to be a certain pattern. And we are able to do that heuristic analysis as well to create that behavior profile. And on top of it, it’s the spoof detection capability and fraud detection that I talked about earlier. What we found is even though the KBase can be bypassed for 40 to 60% of the fraudsters, just our low risk assessment alone is four to six times better than that. So if Pindrop is marking a certain call to be lower fraud risk, just based on that our customers are able to reduce friction, you’re not going to eliminate the friction because all you need is it’s a lower fraud risk call, not necessarily who it is. But you can reduce the friction, you don’t have to ask answer all or you don’t have to ask all for six or 10, or however many questions that customers have, you can ask one lesson to lesson start reducing friction. On top of it when you know that the caller ID that’s presented on the call is not spoofed. And that is the phone number that you have for the user on file, there is a very high confidence that you can match those two things and have a very high confidence. It is the right user that’s calling back and therefore lower transactional risks or transactions. You can eliminate friction, or you can reduce friction even further and ask them one more question if you will, for example, but for the customers or the users that have repeated themselves, where Pindrop was able to create a voice device behavior based profile earlier, and now they’re calling back, we can now on top of these to match the voice device behavior to what was originally stored. And if that matches, that gives you the strength of multi factor authentication, and allows you to start eliminating friction altogether. So that’s how all of these different factors behind Under the hood work, and then work seamlessly together. In fact, we very strongly recommend to our customers to not enroll or authenticate any user, that we have not assessed as low fraud risk at the minimum. On top of it, if they pass the customer authentication, they can enroll that if on top of it, if they have any other way of trusting that user we can, we can consume that in our engine as well. But at the minimum, the we believe that our platforms fraud risk assessment makes the basis of a minimal level of trust that you want to have before you reduce friction on anything.

Loraine Lawson
So is it leveraging AI to do all this or

Amit Gupta
is there is there is heavy usage of machine learning, if you will, in various different factors that come into either creating this fraud risk assessment, we, we have processed billions of calls, I mean, Pindrop, started in 2011. So it started as a fraud protection company for the contact centers, we have now expanded to provide the authentication services, we have gone beyond authentic, beyond the contact centers as well for both authentication as well as personalization services. What we have learned over these many years, with analyzing the billions of calls are that fraudsters have certain patterns, they keep changing the keep evolving. So obviously, we are constantly innovating our technologies, our platform as well. But they have certain patterns, they have certain affinities that we can then assess. And that’s something where we leverage machine learning to create multiple risk engines that assess the calling patterns of different phone numbers. It assesses the we we believe we have the largest fraud Consortium on the call channel. It assesses the reputation of every phone number against the consortium intelligence that we have built over the number of years, it assesses several other carrier based risk assessment. And that’s all just on the fraud side itself. On top of it, the entire voice, biometric engine is a machine learning based engine, we have created that completely from ground up in the last five to six years, it has evolved to a level where to authenticate a user we need less than one second of net speech. So we can authenticate extremely quickly. And that enables our customers to maximize the user authentication within the IVR itself. So they can provide the best personalized experience the the expedited services to all the genuine users, while the same technology is also used to identify a different voice. Like if you have the identifier for a person, you wouldn’t expect a different voice to call against that identity. And we are able to identify that for our users as well, who can see that as a risk factor and can increase the friction for that user as well. And all of this is done in real time. We return these through API’s to our customers within milliseconds. So they get all of this intelligence from the complex machine learning based platform in a very simple to consume API’s and policies that they can then start leveraging to operationalize the user treatment.

Loraine Lawson
So BMO has said you greatly improve their fraud profile along with data they got from a consortium, which of course Canada, Canadian banks will share their data about certain fraud events. I wonder, is there a way in which banks use Pindrop to reduce fraud? In particular, like do they use data in addition to your product itself? Or really pretty much the same?

Amit Gupta
Yeah, I’m glad you asked that. So financial institutions actually, were the first ones are first word nickel, where Pindrop saw a lot of success with fraud protection. And the main reason there was there was a very, very sensual quantification of the fraud losses in those organizations. So when we piloted our product, or when we sold our product to those customers, they were able to see the real effect because they were able to stop the fraudulent transactions based on the interview provided. So from that perspective, yes, financial institutions are special. From that perspective, many of those organizations are actually now creating full blown fraud organization where the sole goal of those organizations are to look at all the different user interactions that are coming in and assess on transaction basis, how to minimize the fraud losses. Now, a lot of these institutions, they implemented several protective measures on digital channels on the web channel on the mobile app. However, the contact center channel or call channel was, I don’t want to say left ignored, but it was an afterthought. And what fraudsters saw was a way or path of least resistance because they could call in and social engineer a fraud, contact center agent, and perpetrate the fraud fraud transaction there. So what banks do with our product is, we are able to fortify their fraud prevention aspects, even in the calls or call channel, when the caller calls into based on all the metadata that we get around the call around the phone number, around the audio around the voice, we are able to not just detect the anomalies based on the audio or the voices or the key presses. But we are also able to create fraudster profiles. So because what typically happens is and believe it or not, fraudsters are very, very motivated these days, so much so that they have their own contact center operations going on as well. And once they attack a certain organization, and if they’re successful, they’re not just trying to attack one account and move away, they immediately resort to attacking as many accounts as possible in the same organization, because they have now penetrated the security firewalls of the data organization. And that’s where being able to create the fraudster profile, being able to now match every subsequent call against non fraudster against a known fraudster profile and then alert the institution on any high risk calls is extremely useful. And that comes in addition to all the metadata based fraud assessment that we we can do and provide the provide the protection on the very first attempt of the fraud as well. So what financial institutions do then especially in our product, is they consume the Intel both in real time, where if the call fraud risk score is higher than a certain threshold, they either send Add call to a more seasoned contact center agent or route that call to a fraud analyst. So you’re not talking to a regular contact center. There are other aspects they can implement in real time in terms of asking that user that they need to go to a branch to be service, because they’re not able to service that call on the phone. Very few customers, but they have looked into the option of just dropping the call, I’m not aware of any customer who’s implementing that those are the discussions that we have. But at Pindrop, we are very motivated in improving the experience as well for genuine users. So we typically recommend against dropping the call altogether. However, some of the aspects where you can increase the whole time, for example, for the call, or you can route it to a specific agent or ask them to come to the branch to get service. All of that is a fair game. And the customers do that from real time perspective. And of course, the other part is to increase the friction within the authentication process itself. All of that is real time. But also, there can be POST call analysis as well. And we have a built in fraud investigation portal in our tool, which customers can zoom, where the alerts on the call are creating cases for them to review. And then they can review those cases and make their own assessment by listening to the call, again, by talking to that customer. Many of the transactions are multi day transactions, like most of the ACH transactions need to be cleared within like a couple of days. So the customers do have that much time to review. But then there are some wire transfers that where the money is going out right away. That’s where the real time treatments become very important. But those are typical ways for our customers to leverage the intelligence and operationalize that. With Bank of Montreal, absolutely, they are seeing a huge detection on the fraudulent events in the fraudulent calls, leveraging our product. And they are absolutely leveraging the voice AI and machine learning capabilities that we have in the platform along with the metadata based risk assessment that we are building.

Loraine Lawson
So what’s next in the frontier of I know you do more than voice authentication, but in your your sphere of voice on indication and the pattern recognition? What’s the next challenge for vendors like yourself? What’s the holy grail for you guys?

Amit Gupta
So well, the mission for Pindrop is to provide identity, security and intelligence or identity trusted intelligence on every voice in production. And what we are seeing is voice is becoming more and more pervasive, right. So Alexa and Google homes didn’t exist earlier. Sure, they don’t exist with financial transactions, or a lot of financial transactions yet, but that’s just a matter of time. So the voice is just becoming more and more pervasive, we are talking, we are looking at connected cars, we are looking at personalized experiences on the smart devices as well. What we have and what we believe and is the voice channel and the why security is where which becomes extremely paramount for us. Because once you can gain a high level of confidence that the user is who they say they are, you can absolutely provide them personalized experiences, we launched our partnership with TiVo a couple of years back, where they’re trying to bring our technology into their, their devices that will be connected to the TVs, right and just think about a family of four people where the kids have different needs than the parents. And this came out in our survey as well, where the consumers absolutely believe that voice technology can be leveraged to put the parental controls in place so that the kids don’t get exposed to the material content, for example. And that can be done extremely easily. It’s not just for that security, but even for personalized experience. And we have a demo where our CEO, essentially talks to the TV saying show me some options to watch or show me some sci fi options to watch. And the sci fi movies like Star Wars come up on that. But then his kid does exactly same thing says the exact same thing. show me show me some sci fi options towards and he gets the Sci Fi cartoons, right. So and that’s something because now that ability to get the intelligence from the voice attribute, whether it’s the age, and some of these are protected classes. So we need to be able to do that in a very privacy friendly way. But that is a possibility. And the technology is evolving. The voice is now coming up as an authentication factor on digital channels as well. So thinking about use cases where the communication between a user and a business starts on a chatbot but then it hits a wall and now that user needs to be connected to the call channel. Today most of the organizations have have the experience where the user has to be re authenticated, they have to wait, same amount of time on the call channel when they are transferred. Can that journey be more seamless? Can it be where they can continue the context from the digital channel into the call channel without making the user wait or re authenticate them. When the banks have different line of businesses for credit card and the checking account, and you have questions about checking account, then you want to talk to somebody for credit card account, they are transferring that call to a different contact center agent altogether, that agent talks to them again. And then there is a whole time there is an authentication again, it gets very frustrating for the users, we have an opportunity to leverage one of the most innate properties of a human being which is our voice to really make that experience seamless. So that’s what we are focusing on how do we make these different worlds connected to each other leveraging the voice? How do we provide omni channel fraud protection? How do we provide omni channel authentication experience? How do we provide the seamless user experience on any channel that can leverage voice? That’s really how we are looking at our future and our strategy.

Loraine Lawson:
You’ve been listening to the Buzz, a Bank Automation News podcast. Thank you for your time, and be sure to visit us at Bank automation news.com for more automation news. You can also follow us on Twitter and LinkedIn. Please don’t hesitate to rate this podcast on your podcast platform of choice.