Hi everyone, welcome to our webinar on voice monitoring. We’re just going to give people a few moments to join and then we’ll get started. All right, thank you guys for joining and welcome to our webinar on rethinking voice monitoring. My name is Yael, I’m the Director of Product Marketing at Shield. Today we hope to provide you with an insightful and informative session looking at how leading institutions are rethinking their voice monitoring. So our webinar today will cover a brief look at the regulatory landscape and how it differs across different jurisdictions. We’ll then move into what the key pillars are for building an effective voice program. And of course, no webinar nowadays would be complete without exploring how AI is influencing the landscape. And finally, we’re going to wrap up with how to strategically leverage voice in your compliance approach. Okay, so before we begin the session, I’m going to take you through some quick housekeeping pointers. The webinar is being recorded and it will be shared to view on demand afterwards or to share with colleagues. Please note that all attendees are hidden and muted during the webinar but you can ask questions at any point by typing them straight into the chat and we are available throughout to answer the questions. We will try to leave time at the end for live questions however if we don’t get around to your question during the session, we will follow-up with you afterwards by email. It’s important to note that we’re not going to be sharing any promotional content in this webinar, and all speakers are providing opinions that are personal to them. If at any point you do feel like you want to contact Shield to understand more about our voice offering, do feel free to click on the link that’s been pinned in the chat. So without further ado, I’m going to hand over to Alex, our esteemed moderator for today and Shields Director of Surveillance and Governance Strategy, and he will move forward and introduce our panelists. Over to you, Alex. Great. Thanks, Al, and hello everybody. Really happy to be here. I love these discussions. I find voice especially interesting because it’s one that every time we have this discussion the details change a bit. It’s one of the areas where we continue to see sort of advancement in where the field’s going. Now, before we jump into the agenda proper, I always like people to just introduce themselves. You don’t need me to say who you are. Tell the audience, who it is you are. So Raj, Mike, Aaron, in that order, if you could just tell us a little bit about yourself and then we’ll jump right into the discussion itself. Thanks, Alex. My name is Raj Davi. I’m the global head of surveillance, with Natixis. Natixis is a French investment bank, part of the BPCE Group. I’ve been in the surveillance space since, approximately two thousand and eight. Started in bank holding company, RegW Surveillance, and moved into, e com voice trade and insider dealing surveillance in twenty ten and have been in the space, ever since. Hi, everyone. I’m Mike Cowell. I’m a senior manager with Deloitte Financial Advisory Services. My core at Deloitte historically was digital forensics and e discovery specifically focused on mobile device and cell phone forensics around twenty eighteen. I started to see how that overlap worked really well with e communications capture and monitoring from mobile devices as we moved into a world where phones were being used more and more for general business. And since then, I’ve been working with clients across the board, FSI and beyond, to help them develop compliance programs, select tools, and ultimately work through the implementation process with them. And hi, I’m Aaron Erman. While I’m an American, you can’t tell from my accent. I’ve been living in London for seventeen years. My previous background contributing to today’s seminar or webinar is I spent six years at PwC in London building our surveillance and regulation practice, culminating in energy and commodities, but also in financial services. I then moved on to TPI Cap, where I led the compliance technology function for six years. I do not represent TPI Cap. Everything I say on today’s call is my own personal views. Currently, I’m the CEO and founder of RegTrail. We’re a reg intelligence platform focused on specifically horizon scanning for the energy and commodity trading space in the compliance and legal community. Looking forward to sharing insights. Amazing. And I’m Alex Dilsson. I’m the director of surveillance and governance strategy here at Shield. So just before we start, you know, I think a lot of these voice discussions, you know, one thing just kind of thinking about it before, you know, as I was prepping for this webinar is a lot of times these conversations, you know, cover a pretty well worn path, right? We always start with the journey from nothing. Right? No one’s monitoring any foot thing and voice to random samples to transcription. We parse word error rates. And there’s this across voice, there’s always been a general making the best of it kind of vibe that that always seems to lag e comms, but it’s better than nothing. Right? And I I think kind of focusing on on on and I I think it two things. One is this, I think it misses something crucial about the moment we’re in. Right? Which is that today’s Gen AI transcription tools show significantly lower error rates. They’re multilingual. They can be optimized for finance. In a lot of ways, the futures or sort of gaps we’ve long pointed to across surveillance have been solved, right, by a lot of the technology we’re seeing today. But we’re still seeing different sort of rates of adoption and certain elements holding firms back. And so what I wanted to focus on today, the technology, the operating models, the risk strategy, and what I’d like to do is just take a step back and sort of understand, you know, how it is we pull all these elements together. So we’ll go kind of section by section as Yael laid out, and we’ll start with the regulatory landscape. So Mike, obviously you’re in the business of advising people, right, and kind of knowing what the landscape looks like. So let’s start off with a snapshot of today. I think we understand a lot of the regulations that have driven voice to date, right? But it is somewhat diffuse and with divergence being top of mind sort of across global regulators and we see people sort of taking different tacks. What’s guiding a capture and by extension monitoring a voice across regulators today? Yeah, so first, you pointed out I’m in the advisory space, and as a consultant, I do need to say that I’m not a lawyer and my statements are not to be construed as legal advice. Always have to give that disclaimer every call I’m Of course. But with US regulation, what we see is CFTC and Dob Frank have specific requirements or mandates for swap dealers and for trade reconstruction. But a lot of other e comms regulation in the US, and this is sort of how I look at it, vague on purpose. A lot of the regulators expect that firms are making good risk based decisions to meet the framework that’s laid out in the regulation, but they’re not giving specifics on necessarily what needs to be done at each step in the process. And with that, what we often see is large multinational financial firms are capturing and surveilling their voice, especially if they do have a swap dealer because they already have the technology in place. But once you move away from those organizations that are really focused on complying globally with compliance, maybe more US focused firms or smaller firms that are really getting their feet wet because it’s not detailed out what needs to happen with voice. We often see that it gets captured, but not necessarily monitored. And I think we’ll come to this a little bit later as we move towards how Gen AI is really gonna make this space sort of easier for everyone. As we get to that point and we see regulators getting that data more often, it’s my belief that the regulators are then going to expect that data in all circumstances, sort of flushing out that vagueness to now include it. However, we’re not necessarily at that point yet, but I see it very close on the horizon. Okay. And so, like, if I could parse that, like so I guess in your view, in your personal view, is there some wiggle room around the interpretive is there some wiggle room around the interpretability of the application of monitoring to voice, right? Or is it more just kind of a of a de facto way, like an organic, we’re just kind of following organic adoption and it isn’t that people necessarily made decisions, they just happen to be at different decision points, but on the same journey. Like is there a distinction there in your opinion? Yeah, I do think that there is an interpretation factor that goes on. The other thing that we see US regulators lean on is once a consent order has been issued to one firm, the regulators expect all of their firms to read that and then apply it. So as we see more and more voice come up in those consent orders, that’s when it will be the gray area will become no longer gray because we have these orders driving future adoption. Right. And then Aaron over to you. So we know you’ve been living in London a long time, right? And that means by extension the UK and EMEA have become much more part of your purview. So, you know, what does the regulatory landscape look like on your side of the pond? What are regulators saying about voice capture, voice monitoring in the space as you see it? Sure. I’ll keep it short. Hopefully everyone on the calls probably, if they’re sitting in seat, understand the regulatory landscape. But primarily, it’s the market abuse regulation. It’s remit for wholesale energy and products. And then it’s MiFID. MiFID and REMIT require voice recording for certain derivative transactions. MAR is your more broad risk assessment landscape. I think what is clear is voice recording under MiFID and Remit. People are doing that. Just to build on Michael’s point, the transcription and surveillance is not prescribed in regulation. That being said, what’s very clear is that now that technology is coming out to be able to transcribe voice, third party vendors are engaging with regulators and showing their kit to the regulators. The regulators are actually using it for their surveillance. Their expectations indirectly or via bilateral conversations are dictating and questioning why they’re not using the technology. I think everybody has their own risk assessment. They have to decide a voice is a risk. I presume if you are making transactions and you are using voice for that, that would fall under that. And then your defense for not having voice transcription or surveillance, you’d have to have a compelling argument why you’re not doing it if you’re a regulated firm. Unregulated, it’s a little bit more What would be an argument for not having one? And I’m just kind of wondering out loud. What would defense look like for a regulated firm not I thought you say voice isn’t you’re only using it for market color, for example. You’re not transacting. That’s a simple example. Unregulated firms may say that we don’t use voice to transact. It’s policed out. So there’s different mechanisms for firms to say While you might use voice for certain components of the pre trade transaction, you wouldn’t use voice to actually transact. And that’s where you can draw the line. Thread the needle. No, it’s fine. It’s fine. I think also just you say regulators. I’d go one step further to say that while the regulators are having conversations and asking firms and trying to understand, I would argue that a lot of the investment firms already are doing voice proactively. The exchanges are also involved. You will see in some enforcement decisions where they request not just the trade information but also the communications. That includes voice and e comms. And so firms are having to produce that information, usually recording. But that has now the expectation is starting to shift as well to say, you’re recording voice and you have the transcription, we want to see that too. Sure. I think one thing though, correct me if I’m wrong, I think is safe to say is the difference between things in Europe, especially the UK and the US, is under MiFID especially, you tend to see more groups of people being voice recorded. Whereas under Dodd Frank, you just tend to see individuals being voice recorded, right? And that’s kind of one of the complexities around navigating these regulatory differences between, you know, how we see voice recorded and monitored in the US versus the UK, which Raj, I think that’s a great point to kind of bring in, you know, ahead of surveillance, right? So here you sit right. I think we always we generally want our programs to be flat right from an operational perspective, because they’re easier to maintain, right? The insights you can get get out of them. You don’t have to justify one side over the other. But how do you balance these different regulatory obligations that are saying, you know, record this guy but not this guy over here and over here, you know, giving you whole populations you need to voice record and monitor while leaving others out. Right? So so what does that decision look like and that thought and and and how do you sort of get comfortable around having a good, voice solution in place? Yeah. It’s it’s a great question. And and I think I’ll start with I also, I’ve just recently I didn’t mention this in the intro, but I recently relocated to Paris. So I’ve been in Paris now for a year. Prior was in New York. So trying to find Alex, to your point, that middle ground between the different regulatory arbitrage regimes in Europe and in the US. So as someone who’s trying to build a global program, one hundred percent agree is we want to be able to build something that’s as consistently applied as possible across the entities, in which in the jurisdictions in which we operate. But I’m learning quite quickly that that’s not always possible and it’s outside of your control. For example, here in Europe, privacy is extremely, important topic that needs to be taken into consideration. I think to Aaron’s point as well, who you can who you can cover who you can’t cover. For example, in France, there are very strong union rules, prohibiting certain types of voice recording. You need permission from certain union groups to voice record individuals. There are opt in clauses that have to be provided as part of employment agreements as well too. So really starting to learn, that it’s just more than thinking about, you know, here’s the people that I want to record and here’s the infrastructure that I need. So really have been spending a lot more time with our privacy office, with our technology partners, our legal teams. There’s a piece of legislation here, the DORA regime essentially, which I think was talking about on third party material outsourcing. There are prescriptive rules in DORA which are far more intense than the US. I’ll give you a simple example. The right to audit clause in the US generally speaking, you have those clauses in those contracts, But in internal audit group, it’s never in their mandate or their charter to actually go visit on-site a third party vendor. They’ll rely on a third party firm, Big Four firm as an example to perform a review and attestation. Versus Indora, in my interpretation, there is an obligation that’s placed on internal audit groups to actually be able to do their own reviews and not simply rely on third parties. So these are just simple examples when you’re trying to negotiate a contract, when you’re trying to apply controls to a select group, a team, individuals, all of these things need to be taken into consideration. So with that being said, we try to harmonize and find opportunities where, you know, we can apply, the intent of the rule and apply it in a way that makes sense for the firm. It’s risk based. We really need to be thinking about a risk based type of methodology. We spend a lot of time on risk assessments of teams and individuals based on their activities who really needs to be in scope. There are some challenges as you guys know as well with the record keeping and retention. It’s a big cost to the organization as well to maintain records for seven, ten, thirty five years. My understanding is with certain swab products as well. So, you know, these are all things that have to be taken into consideration. But it starts with the basics. Maybe I’ll stop there. You know, really, every firm has its own policy or procedure around global communications, which usually are combined for voice and e com. It starts from the beginning on how you define a business communication. As an example, I’ve been at some firms where a calendar invite is considered to be a business communication, right? Even if you don’t even have a calendar event with a client, it’s considered to be a business communication. Now where does it start with voice? I mean if you call your wife or you call a personal contact, you know, it’s under the recording privilege. Do you need to keep it, maintain it, retain it still, or can you destroy it? So all of these questions are still out there that have to be thought through when you’re trying to implement a global program. So hopefully that helps Alex. No, helps a lot. One thing I’m going to stick a pin in and we’ll get back to it, and then I want to stick with you and ask you a few other questions. One thing to stick a pin in is, you know, one thing with some of the Gen AI tools we have now is their ability to sort of understand context more broadly and potentially suppress a lot of those personal communications and private communications, you know, that have always made it a little, you know, not difficult, but been a hurdle or a speed bump, you know, as comms, e comms of voice functions look to, But as these tools get richer and better, right, you know, they’re sort of better equipped to say, well, is a personal communication. Someone’s coming home late, they’re telling their partner they want a pizza and, you know, they’re gonna hit the gym before they get home or whatever it is, right? And we have the ability to suppress these. But we’ll get to that. I wanna talk first about just moving from, you know, what the move, you know, from regulatory compliance and sort of taking all these inputs in from from different regulators and how that becomes a capability, How do we kind of turn that into a program? So Raj, we’ll start with you just because you’re managing a program, but it’s really just for anyone here to kind of jump in. So what are the most important pillars for setting up a voice program? Happy to start us off and then open it up. I think more importantly, the most important part is one, you need to understand your businesses. You need to understand what type of regulatory construct or framework you’re under, whether you be a broker dealer or swap dealer or securities based swap dealer, an FCM, how you’re set up in different jurisdictional entities. So it starts with the regulatory, I think, regime in which you’re under. Because at the end of the day, you want to ensure that you’re, meeting and managing your regulatory risk. So once you’re able to do that and come up with a harmonized view of who the population should look like for voice recording as well as then voice surveillance. I think your next objective is to be able to develop internal solutions or in many cases find technological solutions that will help you solve for some of those problems. Because usually, like in our case, we have hundreds and hundreds of people that are part of the monitored community. And if you think about the amount of data that’s being generated, it’s in the terabytes at a minimum. So how do we get our hands around that volume of data? How do we make sure that it’s recorded? I mean, think one thing we jumped to surveillance, but I always use the chain analogy. You know, we’re at the end of the chain. If the records to begin with are not being captured adequately, if we don’t have all the COBOL devices recorded, maintained, we don’t have mobile carriers providing us the requisite information, we don’t have all the channels that people are speaking on, Teams as an example ingested into our vault system. We’re only as good as the data that’s available to us upstream. So it all falls into this chain of governance that needs to be put in place. So we need to have the right parties upstream from the various respective areas involved in the process end to end. And I think that’s where a lot of people fail as they think let’s just put a new tool in to solve the problem at the end of the chain, but they fail to realize that they have so many gaps upstream. And no matter what tool you put in place, you’re still going to have those gaps. You’re still going to have the false positives, the data quality issues, all the things that people generally suffer with. So really thinking holistically end to end about how you want to maximize the surveillance piece is another thing that I think it’s very important to think about early on when you’re thinking about what direction you want to go in. And then the last thing I would just, emphasize is, you know, keeping in mind, you know, it’s a great point Mike raised around expectations. There are, I have never been part of an exam quite honestly, where it was not expected that you had a voice surveillance program. I think in today’s world, if you have data that you’re recording and you’re record keeping, you’re storing, there’s kind of an expectation that you’re monitoring. You’re doing some form of monitoring on it. So I I think you don’t even have a defense. I’ve never even tried. That’s a good I’ll try it. I would try it, but I would never even try it. Okay. We’ll have another very short webinar where you come back and tell us how that went. Mike, I see a nod in your head. So, you know, what are, you know, we have some pillars, but I don’t know if we have a full structure yet. So if you could kind of fill it out for us. Yeah. So Raj touched on the concept of understanding your channels and understanding where the data is actually coming from. And Whenever I’m given a soapbox I love to get on it on this topic because it’s the hardest piece for me to get some of my clients to fully understand and especially when it comes to a cell phone or a mobile device is there isn’t a one magic solution to capture every channel. A lot of people think of a cell phone as a single endpoint and if you put something on that phone you’re going to capture everything. But if you think about today’s landscape of communication, you can call me on my phone directly. You can call my team’s phone number to get it routed to my phone. You can call me on teams, which doesn’t even go over the phone. However, it still comes to my cell phone. And what I often see as far as a blind spot goes is we solved for the cell phone actual answer was solved for one application on the cell phone and every other application needs to be solved for individually. I often tell people think about a phone as a box of channels and you have to solve for every single app individually. You might have a solution that covers multiple apps, but you can’t assume that. And that tends to be one of the pitfalls that I see people fall into is thinking, oh, we bought this great product, we installed it on the phone, we’re getting the text messages, we’re getting the native phone calls, and then an investigation comes around and you find that everyone’s been calling on WhatsApp because it’s easier. And that’s Or you’re not capturing voice notes or or you’re capturing voice notes, but not transcribing them. Yep. Right. Exactly. Cool. Aaron, anything, else you would add? Yeah, I would. I mean, I’m coming at it from a slightly different role. I think it depends on where people are on their journey. But if you are just starting, I think just the headline, voice is a cross functional change transformation in an organization. There’s a number of different stakeholders. And creating a steering committee with executive sponsorship is critical. And I’ll just throw out the people that would be involved. It would be compliance. It would be IT. It would be voice infrastructure. It would be HR as well. And so getting that cross functional working group together is critical to start the journey because there’s going to be different data inputs that come from different teams. I mentioned HR because when we talk about traders or brokers that are operating jurisdictions or they might be multihadding in different regulatory regimes, that HR data and where that comes from your enterprise HR system is fundamental for how you actually define the lineage of who’s going to be monitored and what gets turned on. I guess the compliance, Raj, you’ve already touched on it. The channels, Michael, you’ve touched on completeness. I think the recording capabilities is fundamental too and the metadata behind it. You just assume that you could turn on the switch and you’re going to get really good data. And so you have to really map out each channel, the data that you need, and make sure that it actually is recording correctly and build controls around it even before you think about going out to find a third party or building in house for a surveillance capability or transcription capability. You’ve to have everything lined up first to then ingest into whatever surveillance system. And then the surveillance system itself, obviously, then I’m sure we’ll get into that in a little bit more detail. But there’s a lot that goes into making sure the pipes work and what the output looks like. And I can add Alex, can I add real quick? Sorry. Just to add to point. I can tell you just from my experience here that this is maybe the third or fourth bank that I’ve worked out in a surveillance area. I have never found an institution that could tell me who the accountable executive is for something like voice. Okay. The problem is I think it’s a shared ownership to Aaron’s point. And what ends up happening is people assume that IT is the owner, but the reality of it is this is business data. This is business ownership that needs to be participating in this as well too. So, most times people rely on the IT teams and they assume that all of these regulatory requirements that the IT teams understand what the business requirements are to develop a holistic program and they just don’t without your participation, without your interpretation of rule and expectations, these things fall apart. So I think Aaron’s point is spot on that it’s really a cross collaboration. But at the end of the day, you need to find someone who’s going to be the accountable executive. And I can tell you from experience extremely difficult to find one person or who will say I own this. Own this program end to end. That that that’s a really great point. If if I go back to, you know, when I was in in in a surveillance role, right, if I think of who owned the voice day to day, there was a voice team, but they were managing desk phones. This was like an old school team that understood analog technology. Right? And all the kinks that went with it. They had nothing to do with mobile capture. Right? They had nothing to do with, you know, once the SEC and CFTC’s fine started coming out with WhatsApp and SMS capture, which was a whole other team that was spun up just to meet that need. Right? And then you have your sort of compliance tech teams that’ll sit across a scaffold into that, but they don’t here’s another thing that would come up is, like, things like the quality of the recording. Now this is another one of these things that that is a little more moot with with Gen AI transcription, but with kind of more legacy surveillance platforms that are using older methods of transcription. If the quality of the recording, so that the rate it’s compressed at, right, is too high, then it’s actually hard to get a good transcription out of it. But the reason those compression rates are low is because it costs less, and you assume you’re just storing the communications. And it isn’t until you have to transcribe them that you bump up against this thing that you can’t actually make good transcripts because the audio quality is bad. So who has to approve that cost? Right? Because the tech guys overseeing those landlines want nothing to do with it. Right? Their day to day is fixed. Right? And now it’s got to go up to, you know, this sort of nebulous executive sponsor. So I think that’s a really great point, Raj. I think, Erin, back to you, I think one of the things, know, one of the really good points you talked about when we were first prepping for this call was that a lot of education comes with voice too, right? That it’s this opportunity, right, to I think kind of bring some of these stakeholders together. And I think maybe let’s use that as a segue. We’ll talk about kind of what that education, what the elements of that education are, and I think we’ve touched on some, and that’ll kind of help us switch into kind of what a good program actually looks like once you’ve kind of brought these people together and kind of educated them. So what does that education piece look like Aaron? Well I think just getting a current state view of what’s happening in the organization and just being able to articulate that to explain the exposure the firm has and the risk of the swearing. And you’d be surprised when you lift the hood up, you go into a corner and there’s a desk doing something and you just find out. So that road show of finding out the inventory does take time. And that’s the starting point. That in of itself is education to see who’s using what. The second education point is helping people understand what voice monitoring is as a capability, why it’s important, and influencing them why there’s a business case to invest in this. And that’s more at the exec level than it is the people in compliance or IT that know that this is a risk already. But there’s a storytelling behind that. And again, people will say this is a cost. Why is this getting on the books? How do we minimize this cost? That’s just standard business operating procedure. But you have the role and responsibility to educate in that. And then the third thing is trying to help them understand the value that it brings to the organization beyond just the compliance. And there’s a business partnering element. There’s a conduct and culture element. There’s front office potential opportunities with analytics or using the data for other use cases. So that is all culminating in the business case that helps you understand it’s not just a point in time, click a button, start transcribing and monitoring, but there’s a lot of opportunities that come with it. Cool. And I find that we often end up teaching people a lot about, you know, something that we’re all familiar with, right? E comms is so established, but with voice, people do have consumer experience to reference. Like, so even my mother, you know, she’ll she’ll she’ll watch like a a message I leave her, you know, get transcribed and that kind of thing. So I find that also there’s this sort of education piece that goes into explaining the tech. But this consumer experience where you can actually close the gap if you can make it relatable. Mike, let’s kind of transition into kind of the programs, right? What surveillance across voice looks like in production. Are some of the traditional elements of those have been random samples, there’s shops that have transcription in place, there’s lexicon. What are the component pieces of a program and then probably open it up again, but also what makes a good program? So first, on the topic of a good program, I wanna echo what was just spoken about, where you have this cross functional team. What I’ve seen, and I think has been expressed already on this call, is the most successful firms and businesses out there understand that you can’t just have compliance legal lobbying things over the fence to IT and IT just magically figures it out and everyone’s happy and moves forward. So it’s really having that team that comes together, Business representatives are discussing what they need. IT is opining on what is technically possible. And then you’ve got legal HR and compliance in the room to really talk about what is required so that they can find that happy medium. Because I think, Raj and Aaron have probably seen this as well, if it’s just thrown to IT and it works and IT doesn’t actually understand the rationale behind it or they don’t have a touch point on a regular basis with legal and compliance teams to actually talk about what is needed versus what the business is saying they want to use, you end up in really sort of odd situations or in some cases everything is captured well, but you end up with thirty different tools each doing a micro piece of what could be done and we run into sort of data format compression all different issues where it’s not consistently coming through. Plus you’re spending a lot of money on many tools when potentially there is a way to tie that together if you have these teams communicating on a regular basis and then owning the process, not just sort of in the consulting world filling out a ticket and hoping someone down the line does the right thing. But let me ask something just because we have kind of kind of drifted back to this topic and clearly it kind of resonates with the group here. So a question for you, Mike, and then a question for Raj. So Mike, how executive should this committee be? Generally, I see relatively high up. So heads of business or their representative, if they choose one, generally somebody fairly senior and legal who can make decisions on the spot about interpreting the regulation, not somebody who’s going to have to go determine and come back. And then with IT, this is expensive. That’s what we’ve been talking about to this This balloons in cost very, very quickly as you want to bring on more channels. You want to meet your clients where they are, which means you’re using WhatsApp. If you’re working in Central America, you might be using WeChat in China that getting all of that and that knowledge together along with the spend does require a pretty high seat in the organization. I wouldn’t necessarily think you have you know C suite on this panel but you’re getting pretty close to the decision makers in the organization because as quickly as technology moves, you need these decisions and the spend to be able to be actioned pretty rapidly. Okay, and then Raj quickly to you, is there a role and a vertical, you know, within FSIs that would be the ideal sponsor? Yeah, you know, my experience is I we have our legal team, so to Mike’s point, but the CISO, right, the chief data officer at the end of the day, someone who’s at that level who understands what the requirements are, what the challenges are, I think also as an important ally to have on a program like this. If I could just spend a second because I think the world is changing quite honestly. We were just having this conversation around show me the return on my investment. Right? This is not something in compliance I’ve ever had to answer. I mean, for the last fifteen years, you pull up the Wall Street Journal and show people an article and say, hey, this could be us and let’s just spend money, money, money. And the world is kinda shifting. The regulatory changes are coming. It’s becoming somewhat less regulated in certain jurisdictions where people are more hands off. So with that being said, people are challenging much more the investments we’re making on these compliance tools. So you’re being forced to kind of think about what is the return on this investment that I’m making? What are the other use cases possibly to Mike’s point? Can I converge different groups that are doing similar functions? Can I can they all use this tool collectively? Show me that you can make improvements with AI on your headcount numbers. Can you flat line your team rather than keep hiring people to generate the review false alerts and false positives? So these are all questions that I’m getting now as we’re thinking about investments and spending money, and the old adage of, you know, it’s in the Wall Street Journal doesn’t really fly anymore. You know, you really have to put strong business cases. Now, of the other things that we unfortunately have measured ourselves against historically are the number of store filings or suspicious activity report filings. And if you look historically from my experience, they’re always coming from the trade side, They’re not really coming from e comm invoice programs. So people are looking for that metric quite honestly, as a return on the investment. Show me you’re finding that smoking gun. Show me you’re finding something different that I couldn’t have found with a simpler lexicon based solution. Right? So these are the kinds of discussions that are happening now and you need to do I would actually turn that on its head slightly because one thing I’m seeing right, is so everything you’re saying totally resonates, right? And I promise not to kind of bring AI up sort of at each kind of inflection point in our conversation. But what I I’m actually hearing is is is less the can we put in a lexicon to solve for this? It’s more if we’re going to if we’re gonna do this, can we figure out if there’s a way that AI can do it that brings us the efficiencies we want and keeps the cost flat? That’s sort of the change I’ve seen at more conferences, from more customers, from more prospective customers, right? In, you know, and it’s pretty much jurisdiction agnostic even though we’re seeing these sort of different flavors of the US sort of diverging this way to sort of a looser, more business friendly approach, Europe kind of hardening, UK somewhere in the middle, right? But the message across each of them has been, okay, if we’re gonna do this and you say there’s AI here, right, is that gonna make us more efficient, right, and can we keep costs flat? And less of the like, hey, are we sure you actually need us to do anything, right, it’s kind of how do we solve for everything at once? Yeah, I agree. And it’s FOMO, Alex. I mean, I could add, like, you know, I think the last five years people, they read about AI at the surface, the periphery, they know that AI help them in certain areas. So, it’s a fear of missing out. People maybe really didn’t think to your point around what the use cases were and how we’re going to benefit from this. It was more of a fear of missing out. And I think it’s kind of, I agree with you one hundred It’s less around now, especially with tier one banks. I work at more of a tier two investment bank. So, it’s a little bit interesting as well too, because tier one banks are operating globally, almost every jurisdiction, almost every regulator. I think at one point I counted three hundred regulators at one of the banks I worked for that were asking, you know, that we were under the purview of. So it’s less regulatory. I think it’s more now becoming, you know, show me that my investment is going to yield, some value. And just saying that you won’t be in the newspaper or you’re going to avoid a fine, this is just some of the stuff that historically It’s not as resonant. I think, you know, so maybe just maybe let’s just fast forward because I think, you know, we’ve been having a great discussion, right? So we’re kind of covering, I think a lot of stuff at Eclipse. So instead of being a kind of a little too sort of formulaic here, want to maybe just fast forward to some some of these AI capabilities then. And and Mike, I know when we first discussed talked, you know, it was something you kinda jumped right into. Right? So with with kind of what today’s day to day voice programs look like, can you talk about sort of the delta between where programs are at, where the AI is at and kind of, and we could start to talk about how we kind of bridge that gap. Yeah. So prior to Gen AI, I’m only gonna take one step back because we don’t have an infinite amount of time here. But if we go back one step to machine learning models and natural language processing, you often had to pick a language, you had to pick sometimes a specialty and do training yourself or pick a vendor who has trained their model in the language that you’re believing people are going to speak, talking about the topics that make sense to their services, to what they’re selling. And as soon as your representative on the phone stepped out of that box, your transcription just could rapidly go downhill. You changed language. Well, maybe you have a call where you’re speaking both English and Spanish, and you might have a model for both English and Spanish, but your call was already set to English at the start, so now the Spanish model might not even be picking it up in some instances. Or Or you’re transcribing it twice and you have two calls Exactly. Just to be able to generate those two transcriptions. And that’s forgetting if someone has a thick accent or, you know, you have like someone in Scotland talking to someone in Japan, right, or market differentiations and so on, right? Yeah, and I always, I hate to pick on Switzerland, but they’re the easiest example here. Back when we were relying on machine learning exclusively or machine learning models, I should split this because AI is such a big umbrella, when we’re relying on those machine learning models, the dialects and the specific languages spoken in Switzerland are really just spoken in Switzerland. So you didn’t have a huge market of vendors, of model developers really focusing on those languages. Also my understanding, I’m not an expert on Swiss law, voice recording isn’t a high priority or the transcription of this isn’t a high priority. So there wasn’t an ability to really even implement consistent transcription. Gen AI really takes us away from that. It’s the foundational models are trained on so much and they’ve seen every language, they scrape the internet for it, so that concern really gets sidelined because you can switch languages. Underline that quickly, right? You know, it’s a Gen AI capabilities are inherently multilingual. Yeah, right. Multilingual they know all the acronyms whereas traditionally if you had specific acronyms your products, you would have to train on those for in the machine learning models. So the Gen AI makes that upfront hurdle a bit easier. It also lets you, like you were saying, train on multiple languages. You can switch language five or six times in a single conversation and you can get a transcript that’s all in English, right? It’ll translate on the fly as transcribing. You might have two transcriptions, one that switches between the languages and an English one, but it’s able to handle all of that sort of natively. The other big piece that we see when we’re talking about Gen AI, and I’ll switch hats a little bit into my e discovery role and how we implement voice transcription when it comes to e discovery reviews for obviously legal matters or other reasons you’d be going through a discovery process is what previous iterations of AI didn’t have that Gen AI does have is the ability to make a summary of the transcription that was made. And there’s a famous example of traders talking about Chinese food prices, it really doesn’t make sense because they’re trying to hide that they’re talking about stock prices. And when you have your summary feature, your transcription might be talking about the price of Chinese food. But in the summary, the Gen AI model might go, well that’s, why was that being discussed at eight o’clock in the morning and why was the price of one meal three hundred dollars That doesn’t make any sense. They’re talking about something else and your summary might capture that where the literal transcription does not. And that’s a feature that we now have. And in discovery, that really helps you get at the heart of what if people are trying to hide things. And obviously that is a direct translation to what could be done these voice call transcriptions and monitoring is not just understanding word for word what is being said, but why is that being said and what does it mean in the larger context of this conversation is a leap that we didn’t have before. Right. The summarization, I’d even forgotten to talk about that. That was, that’s actually, you could take that a step further, right, and go beyond summarizing, right? And there’s a lot of instruct, you know, are instruction based LLMs, there’s a lot you can bake into it to you know make it understand those calls through the lens of a compliance officer to identify the particular risk, right? To go beyond just summarization because sometimes with a long call a summarization can miss kind of what the risk is or what the, you know, what the purpose is of particular purview of people who use our systems, right. So whether it’s someone in compliance or surveillance, an HR person or that kind of thing. So that summarization can be taken a step further and really configured for the purview specific to, you know, certain risks. Aaron, anything, you know, like for just, you know, you helped oversee a global tech function and with the analysis across kind of a voice around solution. Anything else you would kind of factor in around just the performance of the AI or how AI advances voice surveillance? Yeah. I’ll take it back to the grassroots of onboarding a voice transcription vendor. I think everyone assumes that AI and Gen AI is going to deliver almost perfect transcription. While accuracy rates are significantly high, especially in mainstream languages, You can’t assume that. And so a takeaway I leave for everybody here is if you have time and bake time into your voice transcription, you need to test the transcription provider for accuracy. Ground truth takes time. That’s the term ground truth where you record and you actually transcribe one hundred percent accuracy. And then you send that exact voice note to the vendor. They look at it. They give you their transcription. And then you mark their homework. Or they can mark it on their behalf using third party tools. But it’s about accuracy. So everyone will say that accuracy is accurate. And typically, it’s accurate. But you should make sure that you test drive before you buy. It’s really important because while I agree with Michael that Gen AI can take conversations, can take acronyms, I would also state that depending on what you trade, how you trade it, how it’s spoken, the transcription might not get it accurate. And there is tuning that is required. And so you need to make sure that you have the right supplier vendor that’s going to work with you on tuning your models because they may not be as accurate as you need them to be. And if they aren’t accurate, guess what? Your whole operating model of getting efficiencies with AI goes down the rabbit hole. You increase your risk to the organization because you have more alerts to review. And so if you get it wrong, you’ve already turned the lights on and now it’s discoverable in court for your transcription that’s being recorded that you have to review. So I would just don’t go live until you’ve done a lot of the homework and made sure as you turn on the You have to do the testing. Think it goes without saying, and I think there is always some tuning. I think one other thing to insist of vendors that you know says me in the vendor seat but hey one other thing is that vendors should have their own benchmarking for their transcription against both ground truth and you know, I would say other transcription providers, right? Because that’s another thing that, you know, we sort of left it out. But a lot of times, you know, institutions we ask, well, why can’t we just use AWS, right? You know, what’s special about these guys? Put in aside the advantages of having everything in one platform, e comms next to voice and that kind of thing. You know, what makes kind of what’s in your vendor sauce special and specific to financial institutions or your purview? So I think those are important questions for people to ask of any voice vendor. Rod, one thing Alex, sorry to just jump in. One of the other things and Erin right to your point is right now we have about what six mainstream foundational models out there plus anyone can make their own LLM model on top of that. So Gen AI is not necessarily always equal to Gen AI. You don’t just wanna to see what it does on the front end, understanding which back end is being used. Is it a mainstream model? Is it a proprietary model? All of that comes into play because, and again I’ll go back to my eDiscovery hat, one of our Gen AI tools in eDiscovery is running four different foundational models in the background because one is really good at images, one is really good at audio, one is really good at interpreting text and we know that when we’re sending images we use one, when we’re sending audio we use another. And I don’t believe that it’s fully understood that every foundational model has a pro and a con. It does one thing really well, it does something else really bad. And I often see people advertising a model for voice and I’m like, I don’t know if that’s the model I would use. So sort of having that understanding as well on top of just is it accurate or how is it comparing to others also helps you drive what’s going to happen in the future because if you know one foundational model has been particularly interested in images, well, four generations of our transcription software down the line, does it make sense that that’s the model being used? Maybe it does, that’s something that we need to be on top And I would add revisiting those decisions sort of down the line too, right? So we see customers that have controls in place, you know, that, you know, ask of their vendors, right, to sort of say essentially like, is your solution still good? Right? You know, you know it falls on us to then validate that, right? So now we’re almost at time. I want to get to a question, but before I do that I wanted to to Raj, one thing when you and I were talking kind of before this, this would kind of was a really I think interesting perspective and a good one to kind of end on before we jump to a question is one of the things you talked about was around reframing voices as business intelligence and seeing voices more than just the surveillance obligation, but as a kind of asset, right? So could you talk about that and then we’ll see if we have time for a question. Yeah, no happy to do it. So I think maybe a dovetail off of Mike’s comment before. So if you think of voice as a strategic asset, there are multiple use cases, that could be explored with that data, with that information. And I know when we spoke and the way I think about this is, you know, what other groups in my firm or in this institution are relying on voice as an example to perform their day to day. So we have large retail banking operations as an example. They do quality of service reviews, metrics with their interactions of their customers, payment centers, customer contact, that may be going on. I know within our firm we have a large legal team, right? That’s looking to Mike’s point around, legal records transcriptions. I can tell you historically we used to have to call a third party law firm because we couldn’t rely on our own internal systems, and the transcription, when asked. So that’s another use case that probably could be explored. We have other teams that are listening to voice records, looking for voice trading activity, right? So there are a whole bunch of different use cases is the point in which, the solutions that we’re talking about, the technology, the generative AI technology, can be deployed. But I think it’s imperative of us as surveillance officers. This goes back to the whole ROI discussion around trying to explore those use cases, trying to see whether there are other stakeholders in the organization that could benefit from the tools and what we’re talking about here, I think is critical as well to get buy in, to get support, to get the funding you need, to implement the solutions. I think the more the merrier is the approach I take, you know, where you can start to get more teams leveraged onto single sources to be able to pull data from. So, you know, I think there’s a great opportunity, I think for many financial services firms to be able to use this data in a manner which is just beyond compliance, just beyond surveillance. Even, and Aaron may know this like voice bioprints, right? Mean, we have such a hard time identifying who’s speaking to who, what phone number is this? Biometrics. Biometrics. There’s so many interesting things that you can try to do. Voice is my password and then identifying. Cool. All right. So it sounds like we could totally talk like a whole other hour, right? It feels like we just got started. Think, you know, Raj, Aaron, I think just to that last point, I think, you know, I think a through line through this, right, but, you know, I think if I were to draw a through line from how we get to this sort of tech moment that we’ve arrived at, right, I think is sort of this community we bring together, right, to make the case for that change, right. And I think, Raj, one of the things you guys are outlining is that there are opportunities beyond just surveillance, right, that can help pad that case and also put firms on a future footing, right, as we continue to see sort of advancements in the tech across voice. I should say we’re out of time. So there were some questions coming in in in in the chat. And so we’ve been answering those as they’ve been coming in, and we’ll email any the email anyone that where we weren’t able to get to it. Other than that, I wanted to thank Raj Dave, Mike Cowell, Aaron Erman. This has been a really, really great discussion, and and so thank you everyone for for for joining rethinking voice monitoring. Have a great day. Thank you. Thanks, everybody. Thanks, Alex, for hosting. Goodbye. Thanks, everyone.