speaker-0: Hi everyone and welcome again to Machine Dreams. This is the show where we get behind the curtains to see exactly what's working in AI today, what is not working and where the industry is headed next. I'm Collaoulai Samuel. I'm a technology analyst covering AI, cybersecurity and enterprise technology. And as always, of course, I'm joined by my remarkable co-host, Dheera. Hi, Dheera. Doing very well. How are you?

speaker-1: Hi Sam, how you doing? Thank you. I'm Leah Stern, a communication strategist and venture capitalist. Today we're talking about how software companies are changing as AI becomes central to what they do and whether all this AI talk is actually new or just old ideas with a fresh coat of paint. Our guest is Hafiz Jacobson, CEO of Datra Holdings, which turns interaction data like conversations, voice and text into behavioral insights companies can actually use. Datra's platform powers everything from risk prediction and financial institutions to hiring decisions in HR, all by analyzing how people communicate. VoiceSense, which does voice-based behavioral analytics, is one of Datra's core brands. Hafiz, thanks so much for joining us.

speaker-2: It's a pleasure. It's a great subject to kind of look underneath in terms of are things just being repackaged under AI or is it really something new? So it's a really interesting subject.

speaker-0: Thank you so much, Afiz, for joining us. So let's just start right at the core of the conversation today with something that you're really focused on at Datra. And that's the idea that software companies today are moving from selling software to identifying themselves around the old conversation about data. And what I'd like you to do is to walk us through what that actually means, what's actually changing in the software space in the AI era.

speaker-2: So it's a great place to start. And I like from the intro really the concept of what is changing, what is actual packaging versus what is really new. ⁓ So look on the software side, the way I see it, and again, kind of cutting through kind of what it means in terms of the delivery of it is I see software as a platform. And before it was, the importance was stressed in terms of You know, how good is your platform and how easy is it for people to actually interact with your platform? And let's say put data into your platform and maybe a good example of this, which is a, you know, a big product would be something like Salesforce where, you know, had a software package and was really about the ease of using that across different segments of, of, of a business and where it could add value for. individuals effectively to plug in information. I think when it comes to kind of all the discussions and you can see common volatility in the stock market around software companies, it's really about then whether that is the platform SaaS model. Is that the route that is going to move forward or is it really about how do these very good software platforms adapt to AI? and really looking at what is the output and what is that data showing and how do you use it? Because what's really changed is you don't necessarily need to now input the data the same way you did before. The data now becomes automated and it's about really the ease and the worth of that data. Because again, it's that point of we all talk about is, you know, now you're going to hear data as a service. And we use that as well. But the real question is, is how do you actually then use that? And how does an organization actually get value from using more data?

speaker-1: And in that point, it would lead to the next question of, so when you save data as a service, you're talking about turning conversations like voice calls and text exchanges into something a bank or an HR team can act on. How does that work? What are you pulling out of those conversations that really matters?

speaker-2: So look, on this case, it's what, and I'll get to the core of what VoiceSense does in terms of the technology that we have itself. We are, through our technology, listening to kind of the behavioral aspects of how we speak. There's a footprint in terms of individually, in terms of how we speak, and the music, let's say, to put it in lay terms in terms of how we speak. And we've been able to tie this up to kind of processing and looking at the machine learning of this over years and years of data points to then see, we pull common threads in terms of to link how we speak to let's say risk management in terms of is someone ⁓ inherently taking more risks than other individuals and then can tie into am I someone that should receive a personal loan from a bank or not? And it really stems to kind of the fundamentals of can we use something that was untapped in terms of my voice footprint to allow banks in this instance, risk management to have another piece of information to help them make assessments of your character. I would say, you know, looking at this, we can do this at scale, but again, always tying back to a simple view, let's say 30 years ago in banking, you would sit in front of your bank manager, They would sit there and assess who you are as an individual. We just do that at scale and we can do it more rapidly.

speaker-1: That's what I do when I log into my banking here in the UK ⁓ at Santander. My voice is my password. So, Merry Christmas.

speaker-2: Yeah, look, that's a security element of it in terms of what is your actual bio metric in terms of your voice. We are looking not just at the security component, but we're looking at through this conversation. What is my, what is my behavior? What is my personality profile? And can that then help individuals just like we're speaking to each other now, we're all making different views on, you know, somebody's view of who they are as a personality. We can do that through natural speech and do it at very rapid pace and do it at scale. And particularly where this can work really for the benefit and not to sound too marketing, how we can help for the good is there are many areas, let's say in emerging markets where before there was no real data point for individuals in terms of how they could get some kind of credit. because there's no credit history. This is a way that we can work with financial institutions to work with, let's say, the unbanked, the populations that could not have ⁓ access to credit in the past.

speaker-0: I think there's a lot of great conversation to have, especially as AI starts to change how our companies are operating at the very core. There is the term which I heard a couple of weeks ago about ⁓ called SaaSpocalypse, which is essentially AI agents changing the way traditional SaaS companies work. I think that data of the service, as you've put it, is quite an interesting thing. What I really want to get at is a concrete example of maybe a company that came to you thinking they had a software problem, but it really turned out that they actually had a data problem.

speaker-2: Yeah, look, it's, it's, think one of the things with, with AI and, and with the use of data, I think it's, it's, and again, I come from a banking, ⁓ profession as well. And we used to have the phrase, you know, bad data in bad results out. And it's the same now with, with AI as well. If, if you're pulling data, that's not really relevant to the decisions that you need to make, then, you know, regardless of how good you are, processing that data, it doesn't really add any value. And I think, you know, we're working with, you know, large, ⁓ let's say brand organizations, and they use us to really help them with ⁓ customer profiling, you know, and obviously they use other forms. These are huge enterprise companies. They're using other sources of pulling data of, you know, what is the customer? Why do they buy a certain brand? We go in there and work with their data to see if we can add another layer and effectively, you know, target that in a way that adds value and they get better sales. Because ultimately it's not just falling in love with my technology or any other AI technology. If you cannot use that, that actually has a return for a business, then it's not really viable for a business to put in place.

speaker-1: In regards to the technology, were there any components that are interesting of how you came about? Was it based on maybe something learned in the military or something specific, just out of curiosity? Because it's just so cool.

speaker-2: Well, it's interesting. So it did come from kind of trying to assess how individuals from an HR perspective would work in high stressful situations. So, and it started looking at, again, from a government perspective, how could that be used to see how individuals would work when they were put under heavy strain? And what we started to learn from that is not only could we understand kind of that momentary change in terms of how I might deal with stress, if you ask me a very difficult question, you how do I respond to that? But actually looking at, could we predict how somebody would in the future likely behave in certain circumstances? And again, all these are statistical driven in terms of you know, how good can we get to predicting how someone's going to deal with very stressful situations? And that's where it really developed ⁓ initially.

speaker-1: Is it similar to how a lie detector test works?

speaker-2: So, great question. We actually like to stay away from the view of lie detection. You know, people can have different views without getting controversial. Does lie detection work? We're not really trying to take whether someone is lying. What we are analyzing is through natural cadence in terms of how we speak. What is my personality type? What kind of character really am I? Am I someone as a character that might be more apt to be involved in fraud or not, but not necessarily saying, if I tell you that ⁓ it's raining right now and it's not, can you detect whether I'm lying? So there's a nuance in terms of that.

speaker-1: You can really tell ⁓ how a person behaves or their character just by the way that they speak.

speaker-2: Well, that's the core of what VoiceSense does in terms of the natural way that we speak. It's linked to the signal processing, but then very importantly, it's actually linked to results that we've seen over time. And I think one of the things that is always discussed here is machine learning versus AI. And so what we have basically done is seen those certain kind of very undiscernible nuances to our ear. but then we pick up, run it through our models, but we then had linked it to millions of results over time to see that those patterns actually then statistically prove out that there's a high probability, let's say 85 % probability that something actually will take place in terms of those types of operations.

speaker-0: think it's interesting that you just spoke about results, how you've been able to do what you do given the results you've seen so far. what I want to ask is how you deal with bias, because I know that when it comes to detection or some type of predictive capabilities or some AI driven prediction outcomes, there's always the issue of bias, right? How do you deal with bias?

speaker-2: Look, it's a great question because also bias is kind of what gets people concerned about using different technologies. And again, forgive me if I sound like I'm marketing our product on this item, but it's a really important ⁓ aspect in terms of bias. Because we are not looking at the words that you say or the education or am I using Oxford English versus something else. We're just looking at the natural way that we speak. We believe that we take the biases out in terms of what you might perceive of me if I speak in a very educated manner versus if I wasn't educated, but just looking at the core of what kind of individual that I am. And we have done this. now work in, I think we work in every major language in the world and our model, our base model because we're not looking at the words, can cut through those different elements. And we believe take out those biases that we might have all as individuals, whether that is gender-based, whether that is age-based ⁓ or education-based.

speaker-1: And you work with companies across financial services, HR, mental health, these different industries with different use cases. What's the pattern that you see throughout? Do companies building or buying AI systems consistently get wrong? What are some of the things they're doing wrong?

speaker-2: think the first thing that I think is very important is ⁓ how do we work with existing processes? Because again, it's individuals that are still making these decisions to put in any new technology, any new system. And I think kind of the main ⁓ item that I think either companies get wrong or we do that are selling AI to them is how do we work within your existing processes and how do we prove that we actually add value. So it's not just that we provide something that's really interesting and that is, you know, a technology that is, you know, gives very good statistical evidence that, that we are very accurate, but how do we put this into your processes that actually then allows whatever system, whether it's an automated system, whether it's a human element, how do you actually make better decisions for what you're using that technology for. And I think that really is kind of the core of both on both sides of the equation of what groups get wrong. They might come to an AI company and they say, okay, we want to now reinvent the wheel on this and groups will be gladly going in and say, okay, everything you've done to date is wrong. I think there's really a balance to that. And we just think the best way to get groups and individuals to adapt is try to figure out how do we work within the existing processes. and show that we actually add value to whatever that decision. As you say, we work in so many multi-use cases. How do we add value in that particular use case that we're in?

speaker-0: Fizz, just out of curiosity, you talked about the difference between AI and machine learning. And that's something that I've heard over and over again. And even though I've heard it so many times, it's still something that bedazzles me in a way. And I'm sure that our listeners would like to know what exactly is that difference ⁓ between AI and machine learning.

speaker-2: I think there's a reason, I think you can ask so many different people and you'll just like me, you'll get a different answer. I think the way I look at it is I think the core of what still takes place in what we call AI is really the engine is still machine learning. I think the AI component is how these systems now communicate with us. And for example, if you just use an open ⁓ LLM LLM model, like a chat GPT. The engine itself is machine learning. It's based on creating very large frameworks and it processes that at such speeds. But the way it interacts with you is really that AI element in terms of to feel that you are interacting with something that actually has intelligence. And it's a really interesting nuance, I still believe the core of it is still ⁓ ML, machine learning. And until we really get to that point, what you hear of generative AI, that's really a big distinction. And I think, again, from our perspective, where we like to try to look at where we fit into that, is we're trying to add, if you look at, let's say, AI being IQ, we like to try to position ourselves at providing that EQ, that emotional intelligence. to whatever systems that you're using.

speaker-1: If we're looking ahead and we're talking about the future of behavioral AI, where is it going? What sectors do you think it'll move into? How will it become much more mainstream in the future?

speaker-2: Well, I think, you know, in a component of it, know, groups are always looking for behavioral insight, whether again, that's on the consumer side, it's on the banking side, certainly in the mental health side. I see from our perspective, the mental health aspects of it is really a big growth area. ⁓ One, I think, you know, just overall mental health and us as individuals. You know, the discussion points and looking after mental health is kind of a global discussion point, particularly, you know, in today's world going through COVID and how we're so attached to, you know, our social apps that we have and isolation. I think the whole concept of how do we protect our mental health? How do we self monitor our mental health? And how do we work with mental health professionals or help centers to create that space? I see that being a big impact that we can have and we can do that again at scale. ⁓ And it goes then to like all technologies, can we actually provide good information at scale and provide efficiencies? Let's say in just to give an example, let's say remote patient monitoring instead of ⁓ having potentially high cost of doctors going out and seeing patients or trying to monitor those patients. over ⁓ extended periods of time, we can do that remotely and in a way that is not intrusive.

speaker-0: Just on the back of that question about bias, what you're doing analyzing voice data to predict behaviors is quite powerful, but it also raises questions about privacy. how do you handle privacy? What do you tell a company that is nervous about using this kind of technology?

speaker-2: Well, one, don't run away from the sensitive item on this as well. I don't try to sugar coat any item that, know, personal privacy on these issues, whether it's in risk management or mental health or HR, there are key aspects of each component. And so I just look at it in terms of saying, what are your, you know, when you have, let's say employees or you're dealing with people with sensitive mental health issues. One, you need to make sure that the data is protected. And a component of that is that we can deliver our system, let's say on-premise to use an old term or private cloud to make sure that we never, no third party is ever seeing the data that is provided. So we can provide our output onto a private server or their closed end system. And that way the information is protected within their space. And then it's really for them to make sure that they work with their customer, their base of people that say, are using this data for these reasons and make sure that they're comfortable in that process. So I think the most important part is not to run away from it or to dismiss it, but to, deal with that very human aspect that, we all have, including myself.

speaker-0: Thank you so much, Afiz, for a great conversation. I appreciate you being on the show. Thank you for your insights.

speaker-2: not at all. was my pleasure.

speaker-1: And thank you to everyone listening. This is Machine Dreams, where we explore the technology shaping our future and the people making sure they actually work in the real world. See you next time.