A (3:23)

Thank you. That's awesome. We've had a lot of conversations recently, but probably the last year and a half or so about leveling the playing field. And that's exactly what you're doing, is bringing machine learning tool set on a budget for all people, all businesses, small and large alike, to level that playing field. Field. But I like you, dove in a little bit deeper into that is really helping the customer understand the business process mechanism behind analytics and those KPIs or key performance indicators. So I appreciate that. And Zohar Pecon, AI's predictive modeling copilot enables business intelligence analysts to build and train machine learning models independently. How does this tool bridge the gap between complex data science and practical business applications?

B (4:08)

So if you think about it, in many organizations there are data folks, bi people, data analysts, people that know the data very well, people that understand the business, but that don't necessarily have machine learning or data science experience. Those individuals are situated extremely well to bring business value by leveraging the data. They only lack the experience and the know how of actually building those predictive models. The idea behind our copilot is that we've done two main things. One, we've created guidance. We've created a very structured and guided user journey that walks those individuals via a conversational chat around all of the different steps that you have to make as a data practitioner if you want to build a predictive model. So there's the copilot in the sense of guiding you and generating code for you and helping you define the different aspects of the model. The other element sounds a bit technical, but is crucial when you think about machine learning, AI and predictive modeling. And it is the part about getting the data prepared, the data set that is eventually the data set that you will be using as a data practitioner for creating predictions. Getting to that data set is actually the hardest, most complicated aspect of building predictive models. Our copilot builds the data set for you. It connects to your data set, it mines the dataset, it canvases your data sources and it will then, given your definitions of what is the business problem you're trying to solve, will curate and build the relevant specific ML data set for you. Then it will obviously also train a model and reach predictions.

A (6:11)

That's awesome. Thank you. I like how you say really the people that are business minded or the people on the business side of things can do really well using the copilot. Because the copilot will help guide them to the right questions, so they ask or input the information they need to build those predictive models. So I really love that AI is getting very advanced at this point. So Zohar, with your dual PhDs in Philosophy and computational neuroscience, how have these disciplines influenced your approach to developing AI solutions that are both theoretically robust and practically applicable?

B (6:47)

So the computational neuroscience discipline is extremely, extremely statistically legal. It has all the elements of making sure you draw inferences based on statistics, machine learning, AI, et cetera, from different modalities in a very, very accurate manner. Obviously, you can't do anything meaningful scientifically, especially when it comes to a complex system as the brain, without being extremely, extremely careful and robust. I think we've tried to implement some of that scientific methodology into our R and D here at Pecan. The philosophy aspect is probably a little bit less straightforward. I think if I have to distill one major lesson that I took from my philosophical studies is that things are never 0 or 1 or black or white. There are always some shade of gray in between. And I think it also comes in day to day business AI practice. And I'll try to explain a little bit what I mean by that. In many cases, when you build a predictive model, especially if you're a classic data science practitioner, you would go ahead and try to do everything you can to optimize the model's accuracy, because that's the statistical dimension you're operating on. How can I get to the highest degree of accuracy in predicting, for example, whether a customer is going to churn next month? In reality though, when it comes to business value, there's a very strong Pareto rule when it comes to identifying predictive signal in your data. You can get to say 90% of the signal by investing 40% of the effort in many cases. And then you need to invest far, far greater resources in terms of additional data sets and additional tests and additional refinement to get to the extreme accuracy level, that is a potential accuracy level of the model. One of the realizations we had here in Peaken is that from a business perspective, in many cases that additional few points of accuracy do not justify the time, money and effort that are required to achieve them. And so we've built the system so it is not optimizing on the statistical KPIs on the statistical indicators, but rather on the business goal. If your goal is to reduce churn in the fastest way possible, then it's going to derive a little bit of a different modeling approach than if your goal is Just to reach a very high level of accuracy at the laboratory.

A (9:37)

Thank you so much. I appreciate that. And I like how you went in and kind of talked about the difference in philosophy versus computational neuroscience or really the hard figures and statistics you did. Also expand upon. Is the focus on the business goal over having a hundred percent accuracy as we all know, to get to that point does take a lot of resources, so I appreciate the share on that. And Zohar, Last question of the day. Beyond your role at Peak on AI, you've contributed to Forbes, Technology Council and other publications. What emerging trends in AI and predictive analytics are you most excited about and how do you see them evolving in the near future?

B (10:16)

I think there are several trends that are to some degree interrelated. First and foremost, and this is obviously one of my biggest passions, I think AI and data in general and data products are going to be democratized very aggressively. I foresee a short term future where organizations are driven by AI decision making on many fronts, but then coming with it would be explainability and insight discovery. What I mean by that is that it's not enough to have a model or an AI that tells you A versus B you'd like as a business executive or as a business user to understand why it's A rather than B. Today, the AI ability to explain their own decision making processes is limited. And I think there's a lot of research going both here at Pecan and obviously in other companies as well, into how can we understand better what drove a specific prediction or a decision coming out of an AI machine. In addition, I think the game, relatedly I think the algorithms are going to be far more oriented towards action recommendation. I think it's not enough to know what is going to happen or to have some kind of a estimation or likelihood of certain events. I think people would like to know what they should be doing so that a certain outcome either is going to be achieved or preempted. So getting both the explanation from the system and a recommendation for an action based on that is going to be crucial as well. Last but not least, Brian, I think simulations are going to be a big component of what AI will bring forward. So think of kind of what if type of an engagement. Think about asking the model, hey, if I were to change my price, what would be the effect on my revenue? If I'm going to change my branding, what is going to be the effect on my website traffic and so on and so forth. Being able to ask simulatory questions and getting back answers, predictions, explanations and action recommendations is going to be, in my mind, a huge chunk of the development of AI for business in the next few years.

A (12:55)

Thank you. I really appreciate you breaking some of that down. The explainability and inside discovery you first talked about is AI. It's sometimes it's hard. How did it come up with that prediction? And really, if we can delve into that more and as you mentioned, that recommendation for action, how did you get there? Explain that. And then what is the recommendation for action? So I really appreciate that, Zohar. It was such a pleasure having you on today, and I look forward to speaking with you real soon.

B (13:22)

Thank you so much, Brian. The pleasure is all mine.

A (13:24)

Bye for now.