The conversation this week is with Francis Brero. Francis is an aspiring hands-on T Rex obsessed with levering science to improve efficiency and effectiveness. He is a data scientist converted to sales along the way. Currently, he helps B2B SaaS companies get the most out of their inbound pipelines by automating the high-cost low leverage work of researching, qualifying, and engaging leads. He's the co-founder and CTO at MadKudu. MadKudu helps SaaS companies increase conversion rates, upsell, and eight and customer retention. It analyzes customer behavior in your app and enriches leads with relevant data to find out what truly makes people engage. It then predicts, which is a word that we like here on the podcast, which customers you should engage with and tells you why and when finally, it recommends actionable next steps to increase conversion and engagement.
If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events!
Resources and Topics Mentioned in this Episode
Francis Brero 0:00
One of the biggest challenges I feel that we have as an industry around AI, is you know how we design interactions around AI products. And there's at least in the world of b2b SaaS, and especially in the go to market side at the frontier between marketing and sales. There's these two AI personalities that are almost like constantly conflicting, right? There's the police AI and the buddy AI. And the idea is that a police AI is something that is designed to figure out what the right action is, regardless of what you're trying to do. And a buddy AI is rather an AI that's going to make recommendations. So it's telling you what to do it suggesting it and this is just a kind of design and UX perspective. But what's really interesting is if you take lead scoring, right, so marketing is say, hey, sales like these are good leads, you should go off to them. But from a sales perspective, the last thing they want is a UX with an AI behind it, that's telling them what to do. If you're pushing a police AI onto a sales team, you're very likely not going to get any adoption. That day, you might not might as well not have an AI if no one's going to use it, then it's not serving any purpose.
AI Announcer 1:06
Welcome to the conversations on applied AI podcast where Justin grumman's and the team at emerging technologies know of talk with experts in the fields of artificial intelligence and deep learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real world problems today. We hope that you find this episode educational and applicable to your industry and connect with us to learn more about our organization at applied ai.mn. Enjoy.
Justin Grammens 1:38
Welcome everyone to the conversations on applied AI Podcast. Today we'll be talking with Francis Brero. Francis is an aspiring hands on T Rex obsessed with levering science to improve efficiency and effectiveness. He is a data scientist converted to sales along the way. Currently, he helps B2B SaaS companies get the most out of their inbound pipelines by automating the high cost low leverage work of researching qualifying and engaging leads. He's the co founder and CTO at Mad kudu mad kudu helps Sass companies increase conversion rates, upsell and eight and customer retention. It analyzes customer behavior in your app and enriches leads with relevant data to find out what truly makes people engage. It then predicts, which is a word that we like here on the podcast, which customers you should engage with and tells you why and when finally, it recommends actionable next steps to increase conversion and engagement. Sounds like a super exciting product. And our listeners are definitely looking forward to learning more about this during the podcast. Thanks for being on the show, Francis.
Francis Brero 2:33
Yeah, absolutely. And thanks for having me. I'm excited to be here.
Justin Grammens 2:36
I really excited to dive into your product. But maybe you can give us a little bit of background kind of kind of how you got to where you're at.
Francis Brero 2:42
I started my academic background in fundamental mathematics and kind of along the way, started discovering the beauty of applied math and statistics with a lot of like operations research and started to kind of dive into into data science. And fortunately enough, I was able to find a job in the US because I come originally from from France, that job led me to really discover applied data science in a business context. And to dive more into the complexity of the third V in the three V's of big data, right, everyone talks about velocity and volume, like the kind of problematics that Facebook Twitter have to deal with, like massive data coming in at incredible speed and having to build predictions really quickly on top of but very seldom do we talk about the third view, which is variety. And that is the one that enterprise companies struggle the most with, because they have this like, wide variety of data. It's structured, but it's very sparse. There's a lot of missing information and a lot of what you learned in school of like applying all these like, cool mathematical concepts. So unnecessarily work on top of those enterprise datasets, you're really confronted to having to run data science and spending a lot of time on what I call the janitorial work of data sites are really cleaning the data set, increasing signal with future generation, and not just like pushing everything into some kind of neural net or whatever, like magical algorithm. And that was really my first kind of, I would say interesting computation with the real world, right, discovering that the datasets that we were given in grad school were Yeah, they were tailored to teach you to build SBD's or whatever kind of algorithms. But in the real world, at least in the world of enterprise, to datasets are much harder to extract signal from and there's a lot of business acumen that needs to come into play to be able to build out so anyway, that was kind of the big discovery during that those few years that company called Agile one, so, we're fortunate enough to move to the Bay Area raise like our Series A with Sequoia, raise a ton of capital and grow the company and then kind of decided to go my own way and start math cool where the goal is to help companies be able to Uh, as you mentioned, right, like leverage their data and essentially have a tool that is almost the data scientists for the go to market team. So really helping democratize data science in an area of the business that typically doesn't have a data scientist, right? You don't want your marketing team to have to hire a data scientist just to be able to understand what's going on in the business.
Justin Grammens 5:21
I think it's interesting, you know, you're you're actually talking to somebody who actually majored in math and applied math and specific in college. And for me, I joked that I kind of didn't really see the value in solving for x, I really wanted to apply it to something, right? So when I saw that mathematical equation, like what does that actually mean, in the real world, and I think you're right, there's a lot of contrived problems that happen in school, but actually getting out and getting your hands dirty, is one of the the janitorial work, that's what you're saying is this sort of get stuff all set up.
Francis Brero 5:48
And there's definitely a big difference between what you know, like, I have a bunch of friends from my program, who ended up at Facebook, Twitter, and things like that, like running data science teams there. And definitely, that's where like, understanding how to run Stochastic gradient descent, and like, these kinds of things are really critical, and important, because of the size of the data sets. But for, you know, 99% of the companies out there, the struggle isn't in picking a computationally powerful algorithm, it's really about knowing how to look at the data to extract signal from it to be able to, you know, solve for x
Justin Grammens 6:25
How big as your company and give us a timeline with regards to when you did, you know, move to the Bay Area and sort of get things going when you saw the need the market?
Francis Brero 6:33
Yeah, so I moved to the Bay Area. And 10 years ago, we started a company five years ago, or 30 employees suspend few years, like on the r&d side of trying to figure out what are the big challenges that b2b companies have when it comes to their data on their go to market side? And how do we package a solution that both does a good job at predicting and is also intelligible? Because one of the things that was really interesting was, in b2b organizations go to market teams are very reluctant to adopt something that's a black box, right? They, they want understand why a given recommendation was made, why given prediction was made, especially when you're in the realm of b2b where every single conversion is a pretty significant order. It's not like b2c where you might have like a lot of transactions, and it really becomes a numbers game. So I tend to compare it to the difference between low and high frequency trading in the world of high frequency trading, really, algorithms are the thing that are going to make the difference rates really about like, can your algorithm like predict faster and like predict more minor increments, and that's going to yield a ton of value, because you're making so many predictions. But when you're looking at longer term and lower frequency trading, if you look at how people do that, there's a lot more business sense that comes into it, like how you analyze the data. And how you do that is very different. So that was like a big part of trying to figure out what it is that we can do to analyze the data properly, to then package it into a product. And so we kind of came out of I would say, like stealth mode, about a year and a half ago with now a product and a Data Studio, that our customers can go in and actually build the segmentations. And predictions directly in there with an interface that is, like easier to understand for people that don't necessarily have, you know, a math degree and haven't necessarily built a regression in their life,
Justin Grammens 8:27
For sure. Now, are you guys needing to get into the world of like deep learning and machine learning? Or does can most your stuff be done with, like you say, sort of standard gradient descent?
Francis Brero 8:37
Yeah, it's a great question. So deep learning, we stay away from anything that is going to introduce these kinds of like hidden layers, so anything like a neural net, or things like that, where it's very, very hard to explain, like, yeah, you need an explanatory model to explain your result, it creates a ton of overhead. And again, like the volumes of data are not big enough towards going down that path, we've actually done a ton of iterations like, and one of the things that we found to work really well is like small random forests, or even like, think of one type of decision tree, where Decision Trees are very easy to understand, right? Because like these binary decisions that lead you down one path, and then you create a homogeneous group, and you can look at what is the typical outcome for people that fall into this decision? It's easy to understand, okay, if we get a new input, our prediction is basically seeing where would this new data point fall into and what was the historical performance and error and we can kind of assume that, you know, their performance should be similar. So it's a very intuitive algorithm, but it is also very powerful SR iterations. That's like one of the, for example, like one of the most frequent styles of algorithms that we use.
Justin Grammens 9:47
Gotcha, for sure. I mean, yeah. Why? Why bring in something that is, like you said, more of a black box, you don't leave and understand. And there's so many different dials and knobs you can tweak when you're talking about neural nets, if it's not really needed. Do you guys kind of provide at a library that I can add to my app, is this more of a mobile based solution? Or are you sort of tracking clicks through through people's websites Tell me a little bit about how somebody would integrate your product, I guess, into into their existing offering,
Francis Brero 10:12
We essentially, that's what I was saying, we're a Data Studio for good market teams. So the idea is that we sit on top of whatever data you have. So either on top of like your snowflake, data warehouse plus your CRM, your marketing automation platform, connect all of that data into one place and allow the go to market teams to go in and build these predictions or segmentations on top of that data, and propagate the results back into, you know, the data warehouse CRM, marketing automation platform and the website. So really, the idea is to be to be sitting there. So we don't really track any data ourselves. We leverage the data that companies have, and we we have third party providers that help us like enhance the data to reduce data sparsity and increase kind of the the number of dimensions that we have.
Justin Grammens 10:57
Cool. Yeah, cuz oftentimes, people they'll, so it's up to them to collect their data. And then you guys, like you said, they'll start, you know, you, you help provide these tools to allow them to get a lot more insight into how they should be leveraging that data in the best way possible.
Francis Brero 11:09
Exactly. And there's a big component around there that I think is one of the biggest challenges I feel that we have as an industry around AI, is, you know, how we design interactions around AI products. And one of the big components that we've been pushing a lot on is, there's at least in the world of b2b SaaS, and especially in the go to market side at the frontier between marketing and sales. There's these two AI personalities that are almost constantly conflicting, right? There's the police, AI and the buddy AI. And the idea is that a police AI is something that is designed to figure out what the right action is, regardless of what you're trying to do. It's kind of cold is telling you, this is what you should do, because based on historical data, we recommend you do this to maximize your outcome. And a buddy AI is rather an AI that's going to make recommendations. So it's potentially doing the exact same thing in the background. But instead of telling you what to do, it's suggesting it and this is just a kind of design and UX perspective. But what's really interesting is if you take a simple example of lead scoring, right, so marketing is say, hey, sales like these are good leads, you should go off to them. Marketing wants to perceive this AI and intelligence as a police AI, they want to be able to tell sales, these are the leads or accounts should go after these are the ones you should ignore it. But from a sales perspective, the last thing they want is a UX with an AI behind it, that's telling them what to do, they want to have an AI does giving them a superpower, say, Hey, this is actually probably a better use of your time than this one, because XY and Z. And that is going to drive more adoption from their perspective than you know, just telling them what to do. And I feel like that is something where there's still a lot of work and a lot of research to be done in the AI community to make sure that we have clearly documented kind of UX concepts and AI personalities for different types of interactions. We have some of that in the b2c space, but I feel in the b2b side, it's not as clear. So it is something that we spend a lot of time on. Because if you're pushing a police AI onto a sales team, you're very likely not going to get any adoption. And at the end of day, you might not you might as well not have an AI if no one's going to use it, then it's not serving any purpose.
Justin Grammens 13:21
That's a great point. I mean, kind of use the AI where it can provide the most value to people, which oftentimes is maybe in the mundane area of things, but not actually override them, I guess what they want to do. And you're right, maybe maybe as consumers, we're getting a little bit more used to this. Maybe having having Alexa or Google or whatever, Siri, you can sort of ask it questions, it probably feels a little more buddy ish. But maybe in the b2b world, again, I'm just sort of riffing here, maybe in the b2b world definitely needs to be explained a little bit different and probably applied in a lot of different ways I had I've never heard of it sort of thought about that way, but makes a ton of sense.
Francis Brero 13:55
And then there's definitely different levels of, I guess, affinity to AI in different contexts. I think, even if we look at, you know, self driving cars, but people have very, very mixed opinions around that. And some of them want to have the car potentially like assists, and it's like a drive assist, it's going to tell you, Oh, you know, you're going over the lane, I'm just gonna give you a little warning. So you're still in control. And then there's like, people are strong believers, I just want the car to drive, you know, in my stead, and I believe the car is going to make better decisions. So really, that kind of police AI. What's interesting is that very often, what leads to that affinity to like the full on police AI is going to be your assumption as to how good your decision making is, and not, you know, say anything bad about salespeople. But one of the very common things like salespeople very often think they know their customers better than their marketers not saying if it's true or not, in some cases it is in some cases it isn't. But that's why they're gonna have a much harder time adopting something that's coming in to say, hey, this AI knows the customer better than you do. That's like very it's a very hard pill to swallow. And from a design perspective, if that's how you're designing your AI, you're just like running against a wall like nobody is going to like, or at least that customer is not going to adopt your product. So thinking of the design and how we drive adoption based on the customer is something that we absolutely need to think about. I feel like a lot of the AI products have been built by very strong engineering background, people who kind of see it really, as the police know that AI is the right solution, it's better, but they're not realizing well, you still have to think about your end customer, the end user, and what is the incentive for them to adopt it. And if they don't adopt it again, you might as well not build it. And that's something that I feel in the AI community is not discussed enough on how we can explain why a given, you know, recommendation was made, and all that kind of stuff. So there's something around that that really needs to be thought through a lot more in the community.
Justin Grammens 15:57
Absolutely, yeah. Yeah, you know, and in some ways, it's just it, the AI is just a tool. And depending on how we use it, it can be adopted or not adopted. And I'm reading this book called 1000 brains. It's interesting, this was this was the guy I've referred to this on prior podcast, but I'm kind of my mind's deep into it right now. But it's written by the guy that started palm computing. And it was funny in the book, he shared a story about him at Intel in the early 90s, talking about handheld computers, obviously precursor to the palm. And he had he had started the company, but he, they were by no means known by anybody at all. And it got a lot of pushback from management at Intel saying, Well, what are we going to use these handheld computers for? Because at in the early 90s, it was all just word processing. Right? That's that was the only concept that people had was like, Well, why would I write on a little tablet computer? You know, but problem was, was like people were viewing the problem, from the existing viewpoint of what are our current problems today? And so they weren't thinking about the whole app. I mean, the internet hadn't been involved, you know, there was no photos, you know, there was there was no camera on a phone, you know what I mean? There was no social media there. There, there wasn't all this other stuff there. So it's like, you know, as technology evolves, then the problem or the solution that you're trying to solve, you know, you're actually bringing in the wrong tool sets. So as you were talking, I was thinking about this, like we might actually be, I believe we're in the way, infancy of this. And we kind of have this tool that we're thinking about current world problems that we have today. But the reality is, in the next five to 10 years, there's gonna be a huge explosion of all these other things that we didn't even know were coming down the line. That makes sense.
Francis Brero 17:30
Absolutely. And that's true. And in a scary way, also, right? Because even if we look at the digitalization of our lives, right, 1020 years ago, like we had very little digital trace, and now there's so much information everywhere about what we do, like, you know, it's funny, in the kind of conspiracy about the vaccines where people are worried about, like, you know, trackers being injected with the vaccine, you're like, Well, come to realize your phone is tracking you like we don't need another tracker, we already have one. It's called your phone, rally. Google has all that data about where you are, what you're like, if they wanted to know what you're saying they could like all of that already exists, right? So there definitely is something around that that, to some extent, can be scary, right? If it falls under the wrong hands. But there's absolutely something to be said about how it is going to be transformational. It's hard to think in, like disruptive innovation, we're right, built the thinking incremental. But there are definitely like big things that are likely to happen that will dramatically change how we do business, how we interact with ai na, that's funny, like, even like the whole Metaverse thing. It's still very incremental, right? It's still like, Okay, same thing, but now in a digital world, but there's something that can go way beyond that and transcend it. I mean, there's a question to be met, or something to be said about. If an AI is able to analyze every single one of your meetings, it could also start learning how you think like, what are your affinities? What are the things that you're likely to agree or disagree with? And then you could imagine this AI, basically running 10 meetings in parallel in your stead, because it kind of knows for a lot of the decisions, what it is you're going to say or what it is you're going to agree with or disagree with. And then we could almost compress all of our like week meetings into 30 minutes of an AI running that meeting was the eyes of the other people. And we can focus on something else, right? That's like completely disruptive and it's not like decriminalize Oh yeah. Let's do meetings in a Metaverse living great. Like, where are you doing meetings right now on like a video call that that there's something about that, I think is it's hard to project because when we'll see it arrive, it will be so obvious, but until it arrives, it's hard to imagine but even just this rite of thinking, we could finally do the Tim Ferriss Four Hour Workweek, for sure
Justin Grammens 19:56
For sure. Yep, just offload everything off to the machines and we We can enjoy doing the high value stuff, you know, the stuff that we really enjoy doing. You touched on that a little bit, I thought just you mentioned my notes here that, you know, having it being built by people with an engineering background. So I think you're right, I think it's these, it's these technical things that people are thinking about, hey, I'll apply machines here and apply machines there. But I think once we get into the more creativity side of this, it's going to be very, very interesting guys, it's going to it's going to transform all the uses of how we can apply this technology to not just the very simple technology, or engineering based things, I guess, right?
Francis Brero 20:31
That's the exciting part of the time we're living in, because we are at the point where a lot of the fundamental blocks are being put together to make it easy for everyone to run AI, right. So in the world of math career, like we are packaging, data science and machine learning for non technical people to be able to run predictive models in the scope of like go to market. But that's like one kind of use case. But if you think of what GPT three is, right? That is an amazing project, they managed to build these like neural nets that are trained, I mean, on natural language processing, and package them in a way that now people can use all this intelligence and build apps on top of it. So I was testing this app where you can create a song in the style of Kanye West, if you want to just like say, hey, I want to Kanye West song. And I wanted to talk about kudos. And it just like, spits out this whole song in his style. And I think the applications are amazing, because someone without a technical background, which potentially just was a design backgrounds, think of like, this is a problem that I see in the world. And I would like to help people solve it can now skip the technical part of having to build all this AI and actually can build with the existing blocks. And I think that is going to unleash a ton of innovation, because now the barrier is no longer going to be, you know, access to computer science major. And you can actually leverage all that packaged intelligence to actually build stuff on top of so we really are putting together the building blocks. And I think we're gonna see a huge amount of new products coming out around that
Justin Grammens 22:04
It really comes down to Yeah, I guess in some ways, just democratizing, you know, the, the accessibility to all this, my background has been a lot on the internet of things over the past 10 years or so. And I think when I go and speak to people, and I said this five, six years ago, because when like the Raspberry Pi Zero came out for like five bucks, like anybody could smoke a sensor up to a thing and run it. And the internet becoming more and more ubiquitous, you know, everywhere. So now you have people all over the world that for, you know, between five and $10, can essentially start bringing products online and you know, sensing data in the physical world and translating it into digital, it's yeah, it's just it's very, very powerful to bring those tools to the masses.
Francis Brero 22:44
Justin Grammens 25:03
Totally man have, as you were speaking about design, there's a book by Daniel Pink. And if you're familiar with him, he has a book called A Whole New Mind. It really talks about sort of right brain thinking, right, and that we're all sort of brought up in this in this logical left brain, you know, thing where it's like, if you're an engineer, you know, you have to learn math, and you have to learn stem STEM based careers, and all that type of stuff. And it's very, very structured. And what he argues in the book is, is that all worked for, you know, basically 20th century, but in the 21st century, it's not the that's not what you're going to want. And in fact, that's not where the value comes, comes in. And kind of back to what we were saying earlier, it's like, you know, if you're just a regimented person doing tasks over and over again, your job is the first thing that's going to be actually automated. It's the creative side of it, you know, that is, it's actually was where you're going to bring the value to the market. And that's where you'll be doing the four hour workweek is on those types of tasks.
Francis Brero 25:55
And this something that I'm actually I had been pitching at my engineering school back in in France was, I wish there was a hacking class that was taught, because I think we're, yes, we had, like, you know, like, classic like C Plus Plus, and like compiler code, and like understanding all of that, which is great, like you understand, like, how it works in the background. But there's something to be said about, pick a problem and figure out how quickly you can build a solution based on all the tools that exist, right? And that's the thing that that's the job that's not going to get automated in the future, as you said, right? Building compiler code. Yes, there are going to be folks and I have friends who work on the compiler code at Facebook, their job is secure. But that's like three people in the world rewriting how like PHP gets compiled at Facebook, because of the massive scale. But for the rest of us, there's something to be said about how do I take all the blocks that exists and solve a problem for someone validate that this problem is worth solving, and then figure out how I scale that. And I love the Raspberry Pi for this, because it allows you to build really quick prototypes to solve for things and like patch things together, this whole concept of hacking and having hackathons where you have a 24 hour limit, and within 24 hours, you have to ship something that's functional, that proves your hypothesis and is, you know, like tests, if there's value in this, I think that is a class that absolutely needs to be taught in every single engineering school to force people out of the mentality of it's just like writing the right code with, you know, the right constructors, and all that stuff is still helpful to understand in the background, but probably going to become less and less important in the future. And that, to me, that was the best experience, or I was very fortunate to be able to go to the Stanford design school. And that was my first exposure to really design thinking and, you know, thinking like user first and try to put together like, very basic prototypes, just to test that. I think it really opened my mind to going like beyond just like the technical aspect of, you know, just building and optimizing for the execution of whatever you're building.
Justin Grammens 28:01
Yeah, awesome. I think design thinking is, is really, really great approach. You guys bring that into your company, and a lot of these aspects as well
Francis Brero 28:10
We bring a lot of the concepts. So twice a month, we have a half day on the product team, where we just like build things and prototype them. And we have some of the taxonomy that I took back from at least from design thinking of like, are you doing a critical function prototype for critical experience for it's I, it's funny how often having words to put around the concepts help people kind of think that way. If you're thinking I'm building a critical experience prototype, it doesn't really matter how you're building, it's just like, does it test the experience or saying, telling people this is a Wizard of Oz prototype, so it doesn't matter if it works, even if it's someone like doing things in the background, and when I click on a button, there's actually someone typing the answer that should be displayed, it's fine. But it's a good way to test it if it's valuable or not. And it surprisingly, frees people's minds. When they have a name to it, they're okay with it not actually working and just being a Wizard of Oz prototype. And so I found that alone, to be incredibly useful to allow people to think freely and to go crazy. And then there's a lot of things around, you know, how you do user research and having user research at the core of everything you build. I think that elements is a big component that I brought back from, you know, the overall methodology of design thinking.
Justin Grammens 29:25
That's phenomenal. Yeah, yeah, for sure. I mean, I liked the Wizard of Oz approach. I guess I hadn't really heard that. But it's kind of ignore the man behind the curtain right for forget about the implementation details, how I'm actually going to use this thing as the end user, and you can sort of gloss over, because we'll we'll solve the technology problem in the back end. You know, I think the the main question is, is does this experience provide value to me as the end user, which you want to get to as soon as possible,
Francis Brero 29:48
Right, because there's nothing more frustrating or sadder, I guess, then to actually build the backend implementation, only to realize it's not valuable mean that the experience is not great. And they're like, Well, great. We just wasted a bunch of time. And we could have tested this super easily with the Wizard of Oz prototype. And I see a lot of companies struggle with this. And a lot of engineers struggle with it where, like, the first like reflex is like, let me go into the code and like build something, instead of figuring out a way to actually mimic the experience to test its value before actually building anything.
Justin Grammens 30:20
Sure. We talked about some specific, you know, one things I like to think about is a little bit of like narrow AI versus sort of this general artificial intelligence, right. And so it feels to me like, I would like to get your opinion on this. But it feels like yes, we have created some very interesting problems that we've been able to solve right? Very narrow data sets to allow us to write code to allow us to write music, actually, I think about grammerly. And even just some of the auto completion in Gmail. That's pretty cool, right? It's making me a better writer. And it's picking up all sorts of stuff beyond simple punctuation. It's actually Finishing Sentences for me, so that's awesome. But I mean, it still feels like it's there, specifically, purpose built AIS, there's no, you know, to us sort of like general computing standpoint, there is no general processor out there to sort of do all of this, do you think we'll get to that someday? Have you? You've seen, you know, glimmers of that or even like, like, what would that mean to, to us as humans? That kind of thinking, kind of, like high level here?
Francis Brero 31:14
I don't think we ever will. The level of or at least when we look at AI. It's an even as we have it today and how we designed it. It's excellent, like beyond even what human capabilities in one dimension. And even on the computing side, right? Like, there is a reason why we have CPUs and GPUs, like GPUs are great at matrix operations. And like, CPUs are not that great at it. But then GPUs are terrible for, like reading. So the same way that you know, we have these like, competing powers that are optimized for different things. I do think AIs are going to keep on being built on that because there's still an element of a you could say, arguably like there, you don't have to necessarily train it specifically can be unsupervised can learn some tactics and it can become creative. I think. I don't think AI is always only going to mimic what or recreate what's what it has seen. I think we saw that with was a deep mind. I think I know that deep mind. The name of the was a deep go, what was the name of the algorithm that played go?
Justin Grammens 32:13
I think you're right, thank you. What was the VM? Was that the IBM one? No, no, sir. Google one. Oh, the Google one.
Francis Brero 32:19
So I'd highly recommend if anyone has, you know, should have time to waste. I mean, there's like one moment when it was playing against the go, Master. And the algorithm played the most awkward play ever made, like, nobody understood is so funny, because I mean, it was like a big event, then you had like, like this TV cast. And like there was this, like experts was like, moving the little gold pieces to kind of show what was going on in the on the board. And when the algorithm played, you see the, you know, the expert there take the piece and look at the board is like, wait, no, this doesn't make sense, right? Like what the algorithm did was never seen before made absolutely no sense. And that's incredible, because it means that he was able to create a new strategy that it hadn't seen before. And so I do think we'll see creation, but it's still like something that, you know, specific to go in that case, it's optimized for something very, very specific. I think, having an AI that can do and think almost like a human brain. I don't think we'll get there. But to some extent, I feel like that's, you know, it's part of believing in something greater than just like, know, the electric wiring in our brains. And there's something more to who we are than that.
Justin Grammens 33:34
Yeah, for sure. I was Googling real quick, who was AlphaGo? AlphaGo? Thank you. Yeah, did this but but yeah, yeah, you're right. I mean, it's like a five year old, you know, understands, you know, how to bounce a ball and what's going to happen and a computer is very, very difficult to be able to just very simple concepts, right. And even recognizing patterns within within certain things. I think even you know, young kids can see the creativity in that where I think it's gonna be difficult for a computer to mimic as well. Well, yeah. So you know, I guess what are some other things you'd like to do outside of your professional life? Are there other hobbies, stuff you stuff you do?
Francis Brero 34:09
Yeah, absolutely. Big time, big time surfer. That's the beauty of living in or close to Santa Cruz. And it's a great sport to kind of, it's my own way of doing meditation, right? Like looking at the horizon, waiting for the waves to come trying to be in the right position, you have to be in the moment and be there, which kind of is a good way to I guess, meditate, right, which is about clearing your brain of anything else and focusing on what's happening. And if you don't, then he take the wave on the nose and he get like, dropped all the way back to the shore. So it's a there's an obligation to be in the moments. I really enjoy that. And then I also play this European sport called handball, Team handball at a like high competitive level. We're actually the US national champions with the San Francisco team. And we're fortunate enough to fly to Saudi Arabia this year. We're for the World Cup of clubs, where we got to play against, like, some of the top teams in Europe like Barcelona and, and all of those. So that takes a lot of time. A lot of my evenings are like handball practice.
Justin Grammens 35:12
That's awesome. And is that is that like inside of an enclosed area? Kind of like, like, I played a lot of racquetball in the past, right? So is it like that except with your hands?
Francis Brero 35:20
So it's not the one against the wall. It's more like imagine water polo without the water. I see. Okay, so you have to goalpost like the goal is on each side. And you passable, its size three. So it's yeah, just big enough that you can, it's a little bit smaller than waterpolo one.
Justin Grammens 35:37
I see. Okay. But you can palm it with your hands and you just throwing it back and forth. And kind of number of like how many players are on a side, it's kind of like soccer?
Francis Brero 35:46
So it's six on the court and one goalie and you have this kind of half circle area in front of the goal where players could not step into. So you kind of have this defense zone where like the six player in defense, wrap around the zone and in the offense is kind of tactically moving the ball, the ball side to side trying to create space, it's like, very, very high contact, high intensity and super fast game. It's, I'm still surprised it's not bigger in the USA feel like it. It combines the show of basketball. It's fast paced, just like basketball intensity is close to American football. And it's a really high scoring game. It's great on TV, but it is yeah, it historically, it's really huge. In Europe, it's big in South America, growing in in Asia, and in the US trying to grow it more. So we also coach a bunch of schools and PE teachers to have them introduce handball in their high schools and middle schools to kind of grow the sport here.
Justin Grammens 36:42
That's awesome. Well, good luck, and the national champions that would be that'd be great. When is that going to be
Francis Brero 36:48
this year is in May, and most likely in Detroit?
Justin Grammens 36:51
Okay, I don't think you want to go to Saudi Arabia in the summer.
Francis Brero 36:54
Well, you know, we were there in in October, and it was still it was like, I don't know, in Fahrenheit, unfortunately, it was 34 degrees Celsius all the time. Day and night, it doesn't change. It's just warm all the time.
Justin Grammens 37:06
Oh, man, crazy. Crazy. You know, the other thing I'd like to ask people about on the show is, you know, thinking about your career and and sort of advice, I guess, to people that are maybe coming out of school things that you have done along the way that have allowed you to sort of advance in just in technology in general AI, you know, whatever it is any words of wisdom, I guess?
Francis Brero 37:24
Yeah, I think one big thing is so either if you already have a job, and if you're at a company 100% Sure, there are data science problems that you can solve. And there's a way to just take all the side project, don't spend more than 10% of your time on it. But look for things that you could, you know, improve with data and could be in any area of the business could be like financial forecasting, it could be, you know, costs, it could be, you know, on the go to market size, you know, are we going to hit our numbers, all that kind of stuff and interacting with that data and trying to figure out how you would solve the problem is super helpful. And usually, because you're at the company, you have access to the data sets, pretty straightforward that you don't have a job yet. Or if you're a kind of like still a student, or even if you have a job, and you don't really have time to invest, to get access to data, I would say, meetups and hackathons, I think are an awesome way to discover the world of AI. Because often there's the mental barrier of not even knowing how to get started on the project. And the cool thing about a lot of the hackathons that are run is that you get you know, you can get paired with people, and you can get paired with someone who has already done this or like you know, join a team that has some experience, it's a really great way in a short period of time to have to solve a problem to see how they think about solving the problem. And what you'll see is after two or three, you become much better at doing these. And I would say once you've done one or two with people around you to help, there's a website that I love, which is driven data.org Essentially, it's like Kaggle competitions is a data science and machine learning competitions, but for for non to help nonprofit organizations with different problems. So they have this one that I recommend a lot of people start with, which is predicting if you have a list of water pumps in Africa, and you're trying to predict which ones are going to break down in the near future based on historical data based on information about the area, rainfall, all that kind of stuff. It's a good way to start thinking about how would you approach building that prediction and just like iterate, iterate, iterate to get to better and better predictions, and you can kind of compare to the historical level of predictions people got. So it's a really good way to get your hands dirty and to play around with it.
Justin Grammens 39:32
That's phenomenal. Awesome. Yeah, we'll definitely drop that for sure. In the notes. Yeah, as we kind of like wind down here. How can people get ahold of you, Francis? What's what's the best way?
Francis Brero 39:42
Usually fairly active on Twitter or on LinkedIn? So they can Yeah, feel free to hit me up there and connect more than happy to give advice and figure out how I can help in any way.
Justin Grammens 39:54
Yeah, I mean, I would definitely say amen to the meetups and, and hackathons, you know, as people are listening into this podcast. You know, we have a monthly meetup that we meet on the first Thursday of the month. But you know, it's all the they were in person before the pandemic hit. But now, you know, it's the beauty of I guess going virtual with everything is we can bring in speakers from all over. And we can have attendees come from all over. And my also also started an Internet of Things. hackday IoT hackday here in the Twin Cities started that in 2015, and brought together people and it's like, what could you build in 12 hours. And so it was a sort of a 7am to 7pm event where people came in, and they brought in Raspberry Pi's and pick processors or anything like that, that they really wanted, just just to try and try and have some fun together. And actually a couple, two, there were actually were two startups that came out of this event that we did that actually ended up moving on and actually getting funding and their products are now in the market. So yeah, it can turn into some really, really beautiful things. So I totally agree with you on that.
Francis Brero 40:52
In any case, it's always valuable. Like even if it doesn't turn into a company, there's something so rewarding about building anything like I, I love that feeling. And I think everyone does it like at the end like seeing your creation work is just incredible. I built a garage door opener with a Raspberry Pi. So I could on my phone, click on a button and it would open the garage door. And then I was flying back to Europe. And so from Europe, I called my wife I said, Hey, I'm gonna open the garage, or I opened it. And it's it's incredibly nerdy. But that feeling is is amazing. And that's one site I will recommend for people, especially if they're coming to your Meetup for the Internet of Things is instructables.com. That website has a ton of how tos on and they have a specific section for Raspberry Pi's where you have projects that will go from something that can be built in 3040 minutes all the way to something that's going to take a lot longer, we have to do maybe some soldiering and things like that, you can end up building your own tablet with a Raspberry Pi, and a touchscreen. And it's like it's not that hard, actually, it's so cool to do it. I rebuilt a Sonos, which is the bluetooth speaker system with Raspberry Pi's because now the Raspberry Pi's have built in Bluetooth. So you just have to configure it to be a speaker, plug in your whatever, like $5 speakers that you can buy. And now you just have your phone, connect to the Raspberry Pi and play some music. And it's again, it doesn't have to solve world hunger. But there's something about building this and flexing that muscle of creativity of building things and realizing it's actually simpler than it seems to build. And that makes it then easier to go and solve bigger problems because you know, you can build all the steps along the way into the bigger project.
Justin Grammens 42:36
Absolutely. Yeah, people need to get a couple of wins under their belts, right. And even if it's just a couple small steps, you right, it gives you that confidence to go ahead and move forward with it for sure. Well, great, Francis, I appreciate the time today. Really appreciate your your perspective and everything. And you know, it sounds like you guys have a really interesting product sounds you actually have a really interesting sort of culture that you built that mad kudu and you know, wish you guys nothing but the best and I'm sure you'll be evolving. I'm sure you went through throughout your career and applying AI and machine learning and all sorts of new ways in the future. So best of luck to you and thanks again.
Francis Brero 43:09
Yeah, thanks for having me. This is a blast.
AI Announcer 43:12
You've listened to another episode of the conversations on applied AI podcast. We hope you are eager to learn more about applying artificial intelligence and deep learning within your organization. You can visit us at applied ai.mn To keep up to date on our events and connect with our amazing community. Please don't hesitate to reach out to Justin at applied ai.mn If you are interested in participating in a future episode. Thank you for listening