AudioCodes Ltd. (AUDC) Earnings Call Transcript & Summary

Well, hello, everybody, and welcome to this AudioCodes' live webinar. My name is Daniel Raye. I'm a Product Marketing Manager at AudioCodes. And I'm very happy to be joined today by my colleague, John Corbett from the [indiscernible] John, maybe you'd like to introduce yourself.

Yes, sure. Good afternoon, everybody. Good afternoon, Daniel. So my name is John Corbett. I'm a Director of Recording Solutions, predominantly for EMEA and APAC, which effectively is a business development type role supporting the local sales teams when they want to position our recording and productivity products to their partners and end customers.

Great. Thanks, John, and we look forward to hearing from you. Before we get started, I just wanted to run through a few housekeeping items. All of your microphones have been put on mute today, but just to keep things smooth -- smoothly running, but we do give you the opportunity, and we encourage you to ask questions at any time during the session. Just by clicking on the Q&A button on your screen and entering in the questions. We'll do our best to answer as many of the questions as we can during the session, and we'll also discuss some of them verbally later on, we'll have a section devoted to register that at the Q&A. We're also recording the session. And in a few days' time, anyone who registered for this session will be sent an e-mail, which will include a link to the recording, which you're welcome to listen to again or share with your colleagues or anyone who think who may be interested in this topic. Okay. So let's get started now. I think before we start, we're talking about the subject of voice analytics. And I think before we start, let's just -- so everyone's on the same page, John, maybe you can just give us a good definition of exactly what is voice analytics.

So voice analytics is kind of a subset of interaction analytics. So you'll see particularly in things like Microsoft Teams environments, now analytics being performed on voice, video, chat, screen sharing sessions. And within that, we've now got voice analytics. So voice analytics is a way that we can analyze calls, possibly transcribe them. And then utilize some sort of AI/ML type engine to analyze those calls and give you some kind of real business data insights based around the information that's contained within those calls. And you can use those insights for then help you deliver a better service, deliver a better product and also train your staff as well. So it's really just organizations looking to deliver the best service, product that they can and utilizing their repository of recorded calls to get some data-driven business insights out of them.

Great. Thank you very much. That's a good definition. That's a good start. So we know what we're dealing with now. When a company wants to implement voice analytics, John, why is it important that they should identify the need, first of all? And how can they justify the investment in an analytics solution?

Yes. So when you start looking at analytics platforms, some of them are very, very technically advanced and technically debt and it was very easy just to become enthralled by the technology. We always recommend to any of our partners or customers that are sitting here selling or considering some sort of analytics platform is, first and foremost, identify the need. Quite often -- well, almost all of the time, businesses and organizations will have to justify the investment in any expenditure on technology. So that technology has to be able to effectively pay for itself in some sort of return on investment model. So when it comes to analytics, absolutely one of the key criteria we get people to consider is why? What's been the -- what's your problem statement and sometimes it's easy. Other times, it's more difficult. But let's just say you're an organization offering some sort of online service or insurance or something like that, and you begin to churn customers. And you've got no idea why. You can use advanced analytics to investigate your customer service calls to try and understand, based on what your customers are telling you, why they're potentially leaving. So once you do that, you can put processes into place to address that and maybe reduce that churn. And by doing that, you immediately create a cost benefit case for you, which can then help you justify the investment in that technology. And one of the beauties of doing that is if you can identify the problem statement in one particular area of the business, that can be used to justify the investment, then that technology can be used elsewhere in your business and deliver benefits, which are now going to be over and above the original requirements of the analytics platform. So you're not limited to just using an automated analytics platform in one particular area of your business. You might want to use that one particular area of business to justify the investment, but then you can expand that into other areas of your business. So identifying the need, identifying the problem statement allows you to monetize that output, which will then justify the investment of the analytics platform itself.

Right. That's very interesting. Now when any company that employed has a call recording system in place is going to have a huge amount of recorded data available to them. It's a huge resource that they can make great use of in defining and developing insights for their business. But why can't the company just listen to the calls that they've recorded to understand what the customers are thinking or how their employees are performing? And do you think an automated platform will do a better job of this?

That's actually a really good question, Daniel. It's one we get asked quite a lot. So generally speaking, in my experience around recording, and I've been in recording for a number of years, most organizations, if they are going to analyze their calls and have someone listen to them, maybe listen to less than 5% of their calls over a year or so. So that's a very, very small data set that's being utilized. Now let's just say they've got 1,000 hours worth of calls. Is it realistic to expect the member staff to go through 1,000 hours -- listening to 1,000 hours worth of calls in order to get some business-driven data insights out of it. Now realistically, that person is not going to do that in one go. It will be over a period of time. And clearly, if you're doing that sporadically over a period of time, even if you were to listen to all of those calls, it would be almost impossible because of the duration of the time you're listening to them and the gaps in between for you to identify any real trends that are appearing on those customer calls. Also -- and again, I'm sure we've all experienced this. So if you play the same call to 3 different people, because of the subjective nature of how humans interpret calls, you'll possibly get 3 different interpretations of that particular call. So one of the advantages or one of the key advantages of what we make in analytics is it kind of takes the emotion out of it and the subjective nature of the interpretation of the call. Of course, the key takeaway is that you can analyze call at scale. So up to 100% of your calls can be analyzed versus the 5% if you do it manually. But also because of the much more objective nature of the AI and ML engines that are being used to perform this analysis, you're taking any subjective bias away and just getting an opinion based on the analytics platform doing the analysis, which may not have some inherent bias, but by and large, you're removing the emotion from that and then just getting trends identified and data insights delivered to you, but hopefully, that you can enact upon. So if this is something you're considering doing any sort of scale, then clearly an automated platform is much more beneficial to you as an organization.

Right. Yes, that makes a lot of sense. It's just a huge task, which is humanly impossible to do efficiently, really cost effectively and time-wise as well.

Absolutely.

Yes. Okay. So we've seen the value of automated analytics solutions. But as we have a look now at some of the practical examples of ways that voice analytics can actually be used within a business or in an organization to give benefit in different aspects. I think one of the -- maybe one of the most obvious cases would be compliance. A lot of companies, in particular sectors are required to have call recording in place for compliance and regulatory reasons, financial services, insurance, health care and many other industries as well. But how can they leverage that information with analytics to provide benefits -- additional benefits to the company?

So one of the big challenges that we find customers come across with analytics, certain analytics platforms out there are incredibly feature-rich and complicated and quite often produce an output, which is very difficult for customers to understand and utilize in a manner that's going to benefit them as a business. And quite often, this can result in the customer's concern having to go away and hire a business consultant to come in, analyze the output from the analytics platform and then trying to overlay that onto their business to produce the tangible results that I mentioned previously, which then justifies the investment. So it's kind of like a circle, if you will. One thing that we try and get customers to do, when we talk to them about analytics and things like that, is to maybe consider kind of putting advice chunk size specific user cases and one you've just follow up compliance is actually one of those. So you can use analytics in a compliance type user case to do things like identify whether a regulated organization is offering something like financial advice. Now quite often what you're doing is, you're going to be setting up like a custom dictionary of words or phrases both spoken and not spoken that you want to be notified from. So you can set up a dictionary using words like advice and my recommendation and things like that. Because many financial or regulated financial organizations aren't able to offer advice. So you don't want those agents offering any sort of financial advice for the customers. So you can be notified of that. One of the key things around compliance environment is very often a financial organization will be required by regulation to make some sort of regulatory statement at the start of the call. Absolutely, you want to be notified if someone doesn't make that statement. And when you set up these custom dictionaries, it isn't just a case of you'll be notified if any of those words are used, you can set them up so you can be notified where those words are not used. Thereby, you can be notified if that regulatory statement isn't ever offered at the start of the call. You can look for particular trends in customer complaints about service or products. And then particularly in compliance, there's the whole security aspect of it. There's certain words that you should never hear spoken really on a compliance like caller, keep this confidential or it's secret or I don't have my ID and things like that. So being able to be notified of those types of calls absolutely is something that's very easy to understand within the compliance environment and very -- if you can recognize that benefit. And then if you kind of move on from that, there's 2 or 3 other kind of bite-sized user cases that people can use. And maybe the most common one, the one that people certainly seem to come across most often these days is the quality recording environment. So I'm in the U.K., and my understanding is across Europe and America, this is now very, very common when you ring a customer-facing organization, the first thing you hear is a recorded message saying your call may be recorded for quality and training purposes. Well, what does that actually mean? Why quality and training purposes? Why is that? And in a nutshell, any of these customer-facing organizations are looking to offer the best service that they can. Now whether that services is literally as a service or service support in a product, they need to be able to identify that they're offering the best service they can. And if not, what things they need to do to offer a better service. So you can use automated analytics platform in a quality environment to do things that identify the customer experience. Is the customer having a satisfactory experience dealing with our organization. And then much like the regulatory statement they have to make, if agents are supposed to be following a script, whether it's a customer support script or a sounds type script, is the agent following the script properly? Are customers finding it difficult getting through to your organization? Are there problems in your voice routing environment, which means customers rather have had to go through 5 different levels of all our [ attendant ] And by the time someone actually speaks to them, they're annoyed and they're upset. So being able to identify that allows you to go back and maybe simplify the inbound voice routing. And that kind of dovetails with friction drivers. Are you getting a consistent customer complaint type coming into you where the customer is complaining about the service of your agents or the call routing or other performance of a service or product, being able to identify trends like that, again, allows you to put processes in place to ensure that those friction areas are addressed. And then lastly, it's very easy to get completely focused on your customer engagement coming in, but your agent performance in dealing with these customers is incredibly important as well. So being able to analyze your agent performance, both good and bad. So all made analytics platforms can be used to identify good examples of customer service, which can then be shared for everyone to use. But also then poor examples of customer service, which I'm not suggesting like I shared but they can be used to develop a bespoke training plan for that particular agent. Again, all with the open outcome of providing a much better customer experience and service to your end customers.

Okay. That's a good practical example there, good use cases for voice analytics. One of the buzzwords, I think when we talk about voice analytics and this kind of solution, one of the buzzwords that keeps coming up is sentiment analysis. Can you throw a bit of light on to that and why it's important in a voice analytics solution?

Absolutely. You'll see sentiment analysis mentioned by just about every analytics platform out there. And a sentiment analysis is really just trying to determine within a call if the caller is happy, unhappy or kind of middle of the road. And it tends to be the 2 extremes that organizations focus on more. And again, by the way, I'm talking about customers, but also your agents, if there's something that's upsetting your agent's view, it's good to be able to understand that as well. But effectively, there's a couple of ways of approaching this. So some analytics platforms will do things like interpret the words that are spoken based on the transcription of the call and a predefined custom dictionary, which will determine what words indicate good and what words when spoken, indicate bad? And then everything else is kind of in the middle. So you can set up those custom dictionaries. And then the automated analytics platform will analyze the transcription of that call and then give a score on that call, highlighting it, which the platform itself has examined this is an unhappy customer and also where the customer is happy and everything else is in between is where the customer is just so, so. And that's one way of doing it. That's certainly probably the most cost-efficient way of providing sentiment analysis. There are analytics platforms out there that do this in a very different way. It uses a much more comprehensive kind of AI/ML type engine to now start looking at the tone, the intonation of the voice, as well as the words spoken, to detect things like sarcasm and gives you a much better -- arguing much more accurate view about happy and unhappy customers. The kind of offset to this, though, is that those platforms that do that tend to be significantly more expensive than those that do pure sentiment analysis based on the text of a call, the transcription of a call. So it's really up to you to work with your customers or your customers to decide how -- exactly how comprehensive do we need the sentiment analysis. And it's a really useful tool and it can be used with other things like categorization so you can categorize calls based on whether happy or unhappy or whether certain things have been discussed on a call. So you can combine the sentiment analysis with categorization to give you a really good feel for the types of call that is coming in from your customers and allow you to put processes in place to remedy or repeat the good and bad things that you're experiencing within that. So being able to get a visual cues when you're playing back your calls, okay, Here's an analytics score. 80% of this call is unhappy, right? If you're looking for complaint calls or you want to address unhappy customers, immediately, you can see that that's an unhappy call. You can listen to it, validate that indeed is an unhappy customer and then put remedial processes in place accordingly. So having that sentiment analysis on your customer calls is a great indicator of how you're performing against your customers' expectations or what, if anything, is that in your customer, and as I say, you get visual cues on playback that allows you to immediately drill down on that. So as a tool, it's kind of invaluable in analytics of voice calls.

Right. That's great. Yes, it sounds really important, and I think you can get a lot of value out of sentiment analysis. One other, I think, buzzword, everyone is hearing about that, not just in voice analytics, but nearly everywhere is the subject of ChatGPT and generative AI, large language models. Do you see applications for that kind of technology within the voice analytics world?

Absolutely, Daniel. No, you have to be living under a rock and not have heard of ChatGPT. It's kind of everywhere these days. And certainly, we are seeing these large language models like ChatGPT now being embedded into analytics environment to provide additional analytics and data insights based on what's there. So there's a couple of things you kind of need to consider when you're talking about these large language models like ChatGPT. Without a doubt, when they are used in an objective environment, so for summarizing meetings, providing the meeting summary, providing bullet points and actions assigned in that meeting, they are an incredibly powerful tool. And I think as we move forward, you're going to see a complete shift change in how we manage our meetings and the outputs of our meetings as we move forward based around these large language models like ChatGPT, they're an incredibly valuable tool that are going to be used in those environments. The other area where ChatGPT is often being used is in a more subjective environment where maybe a very nebulous question is being asked and ChatGPT is going away, accessing the repositories of information it's got access to and then providing you with a particular answer. That's a slightly different way of using ChatGPT. And we've kind of come up with an acronym, just to bear in mind when you're using ChatGPT in this kind of subjective environment. And that acronym is RAVE, R-A-V-E. So when you get your output from ChatGPT, absolutely initially review it, make sure you understand kind of what's there. Then assess it, is it accurate? Is it providing you an accurate representation of the question you'd asked. And then importantly, if you can try and validate that premise that's now being delivered by ChatGPT from another source. And if you can do that and you're not comfortable that is accurate, then you can now employ that across your business, but we would always recommend in this subjective environment, if you possibly can, you validate the output before you employ in your mission-critical processes in your business or before you start changing your business processes based on the output from ChatGPT. But without a shadow of a doubt, we are going to see all of our workplaces change over the coming years with the advent of these large language models. You've only got a look now. There are a plethora of these things in the market. One thing I will point out and often this gets brought up is exactly what we do about kind of value protection and things like that. I would always be looking for an analytics platform that's using one of the commercial large language models. So ChatGPT itself has an open-source version and then a more commercial version. So I would only ever want to in a business environment, look at the commercial models. And then when it's looking at analyzing conversations, you have to bear in mind that human conversation can be quite nuanced. So any determinations and insights that come out of ChatGPT. Again, as far as you can, just think about acronym RAVE and particularly the V part about the validation and the verification. And if you even ask ChatGPT itself how accurate is analyzing voice calls, it will be quite clever in its answer and say the accuracy depends on a number of particular features like the accuracy of the transcription and the quality of the audio, things like that. So in itself, it knows that there are aspects around analyzing calls that can introduce some inaccuracy, which makes it even more important for if you can to get some third-party validation out of the nuanced subjective ChatGPT outlook. The objective of stuff when it's really just summarize in a meeting, producing bullets and action points, yes, you're pretty much good to go with that because it is objective. There's no subjectivity involved in it. But on the subjective piece, and again, look, often, it's very accurate. But as I mentioned, before you start making business processes and deploy them in your business, just make sure if you possibly can validate it before you start changing your whole business processes based on the ChatGPT output.

Right. Great. So [ actually ] space, I think, is the bottom [indiscernible].

Well, it's here now, Daniel. I mean you can watch this space, but it's absolutely here now. It's just going to get more common as we move forward.

Right. Great. Okay. Okay. So we've come on to the Q&A section now. I've had a few questions come in. And John, if you're able -- I want to -- if you don't mind, I'm going to put a couple of them to you. One of the questions is regarding the objective insights that you can get out of voice analytics system. And you talked about how automation is more -- maybe more objective than having humans analyze the calls. But what about the potential introduction of data by algorithmic bias, contextual bias and others built into the actual system itself? How can that be avoided or mitigated?

So the simplest answer to that, probably is it can't. whether you're aware of it or not is another matter entirely. Now this is one of the areas where they use the commercially available models rather than the open source. But unless you can get into the call, and speak to the developers around that, it's probably incredibly unlikely that, that an end customer is going to be able to identify any algorithmic or data bias. At the end of the day, any analytics platform potentially has a bias because if it's a dictionary that the user is setting up to identify key words and phrases then is then a bias that you then create. But when it comes down to the underlying technology [indiscernible] things are bias, it's very unlikely that firstly, you're going to be aware of it. And secondly, you'll be able to demonstrate that bias. which is why, again, thinking about RAVE, it's incredibly important when you start getting the outputs, if you possibly can, you pick a couple of aspects and then validate that against a different data source to ensure it's as accurate as you can get it. Beyond that, unless you're really a sort of development partner with any of the companies offering this then it's very unlikely you're going to get to see or experience any of the algorithmic or data bias that may or may not be present, but you can mitigate that by validating the output before you act upon it.

Right. Okay. Okay. We have another question here about -- this is about Microsoft Teams and the built-in analytics provided by my Microsoft Teams. Do you -- can you -- do you have any comments about that? And I think they've just recently announced a new meeting summary feature. How does that compare with other voice analytics of offerings?

We get asked this a lot. So look, here at AudioCodes we have a couple of kind of analytics platforms that we can offer to our customers. One is our compliance recording platform. And the other is our kind of meeting productivity platform code -- so SmartTAP is our compliance reporting platform and Meeting Insights is our productivity platform. Now ironically, SmartTAP uses, as its analytics platform mocks of cognitive services. So we're using that commercially available Microsoft service already. When it comes to the Meeting Insights, and this is the one where I'm talking about the objective ChatGPT integration, things like that, we are using the commercial ChatGPT engine, of which I believe Microsoft is a 49% owner of or something like that. Can we expect Microsoft to include functionality like this into the Teams environment? Yes, we probably can. They don't offer compliance recording on Teams. They absolutely rely on third parties for that. But they do have the native recording platform within Teams where you can click -- record this call, but that's not a compliance recording platform. Likewise, they're doing things like meeting summary, which I believe will offer a decent subset of functionality might be not as extensive as what we can offer with ChatGPT and the features around Meeting Insights. But the one thing to be aware of, I think the meeting summary or the features that just brought out require a Teams Premium license. So the Teams Premium license, if memory serves, is about $8 to $12 more per user per month than it is for a standard E3 or E5 license. So again, as with any technology deployment, you'll look at the features, the functionality and the costs associated with it. And you may find there are some features of platforms like AudioCodes Meeting Insights that you think is really great. But then you'll have to cost up and then compare that to the features and functionality that you get on things like meeting summary in Teams and the cost that you get -- the additional license cost, you're going to have to pay for that to try and determine which one is best going to see you as a business. And if you remember, when I came right at the start of this day, we talked about identifying the need. This is where if you've identified the need and the success criteria, you should then be able to determine what platform is going to best help you meet that success criteria, and then the cost benefit justification will come into that as part of that discussion. So yes, look, we fully expect to see, as I mentioned, the market as we move forward in the next 12, 18 months is just going to be swamped with things like large language model engines and things like that in areas which maybe we hadn't even -- can't consider today. And I fully expect Microsoft being a co-owner of ChatGPT to embed that where they can, but I also don't expect them to do it for nothing. They're going to want a license fee for that, and that license fee may be greater than it would cost if you look at a third-party platform includes ChatGPT. So I think we're in the process of deal side-by-side on a particular platform versus the Microsoft platform. And I fully expect every other analytics vendor out there, that's offering anything like this, particularly around Microsoft Teams. We're doing the same sort of thing. So any company that's considering this, will be able to do the due diligence and be able to see side-by-side comparisons. And then based on the identification of the need, make a selection of the most effective, both from a functional and a cost perspective to meet their requirements of their business.

Great. Okay. Thank you very much for that. So I think that's covered the questions. We're reaching the end of the session now. But John, just before we go, can you maybe just summarize what you think are the key takeaways you'd like our listeners to go away with from today's session?

Yes, certainly. I mean, at the risk of repeating myself, if you're going down an automated analytics kind of route, you think it's something your business is going to need. First and foremost, and I can't stress this enough, I identify the need, identify the success criteria that you want at your analytics package for a number of reasons. First and foremost, to justify the investment. But secondly, so you can ensure you get the outputs that you need. Now having said that, there will be organizations out there that will be looking at analytics platforms without a specific need. They may just be doing it so that they understand their customers a bit better. They can use the analytics platform just to see if they can get any information from their customer base that maybe they're not aware of currently. And that's fine. That's absolutely great because then the process in itself is the justification for the investment. Generally speaking, most other organizations will want to see some sort of return on investment from that analytics. So absolutely identify the need, if you can, and then use that to help develop the success criteria that you need to get out of the analytics platform. Once you kind of decided you're going to do this and you start really getting into the nitty gritty. If you can try and create some of these user cases that we talked about, the compliance user case, the quality assurance user case, maybe a sale user case, something like that. Because if you kind of cookie curry up into bite-size chunks, then there won't be a need for you to engage a business consultant to interpret the data output because you've clearly defined it at the start. So the output should be in line with your expectations and what you want as a business. So if you can put your defined user cases together, I think that will really benefit you as a business as you move forward. When it comes to ChatGPT, look, we're going to see this everywhere over the next few years. Just kind of keeping you back of the mind, the subjective and the objective usage of ChatGPT, the objective uses where you use maybe within Teams or someone like that to analyze the meeting, provide a meeting summary, provide bullet points and actions and things like that. Great. You're good to go on that. I don't think there's any real challenges there at all. On the most objective where you're getting a subjective analysis of a not clearly defined data set, then just think about RAVE, review, assess, validate and employ. If you can validate the output from ChatGPT against some other data set, just to ensure it's accurate, be confident and then be comfortable in that before you employ it, particularly if you're looking to change business processes when you go around doing that, it isn't infallible. Itself admit, it's not infallible. So just bear that in mind because I don't want to see anybody fall into the trap of just assuming anything that ChatGPT produces or larger language engines like ChatGPT, it's immediately going to be 100% accurate. I think that would be a dangerous trap to fall into. Beyond that just be specific about what you want to get out of it from an analytics platform.

Great. Okay, John. Well, thank you very much. I think we've all learned an awful lot about voice analytics and the benefits that it can bring to organizations in any industry. I think it was really fascinating a lot of great insights there. So thank you very much for all of that, John. And thank you to all our participants for joining us today. I hope you enjoyed it. Just a reminder that in the next few days, everyone who registered and participated in the webinar will receive an e-mail from us with a link to the recording, which you can listen to again, if you miss part of it, if you want to listen to all of it again, if you want to share it with colleagues who may be interested, please feel free to do so. And if you have any thoughts, comments, if you want to learn more about AudioCodes call recording, meeting productivity, voice analytics solutions, please feel free to contact us. There's a link in the Q&A section, I posted earlier, which you can use or you can contact us via our website, audiocodes.com or via your local AudioCodes representative. Any way you want to contact us, please feel free to do so, and we'll be happy to help you and explain and discuss your needs with you. And that's it, we reached the end of the session. Thank you very much again. Thank you, John. Thanks, everyone, for joining in, and we look forward to welcoming you to future AudioCodes events as well. Thank you very much, and have a great day. Bye-bye.