Transcript
both on and off the pitch, but who will be the winners and losers? And can AI really predict the path to the final? I'm really super stoked to be joined by two superstars here, Steve Palmer and Nick Magnusson, and together we're going to be talking about the ups and downs of football, data and AI for the next 20 minutes or so. But first, I have to say that the views and opinions you're going to hear today are purely those based on the contributors, ourselves. They don't represent Qlik. All the predictions we're going to make, including those from Qlik products, are really based on all the available data that we've got. It can be inherently uncertain, and they're not guaranteed of future outcomes. So nothing that we provide here can constitute as betting, financial or professional advice. And Qlik is not liable for any decisions made based on this content. So with that out of the way, I can really introduce our special guests here. Steve Palmer, who's the VP of Data Solutions in the Premier League. Steve is a former professional footballer with over 18 years of experience and across his career. He's probably played well over 600 games under his belt, and that's across Ipswich Town, Watford and QPR. And uniquely, what I love about Steve is that he is a Cambridge University graduate, so he can talk up data better than I can, which is awesome. And because of that, he's got both passions together, leading the data solutions at Premier League, where they and his team use Qlik to power analytics for the league, its clubs and the key stakeholders. And then Nick, who is our head of AI at Qlik, he's my AI guru that I go to. He leads our AI strategy and product development since the acquisition of BigSquid, where he was formerly the CEO. And he brings well over 25 years of hands-on experience in machine learning and predictive analytics. And I believe, Nick, you also once a pro footballer as well. So again, you can bring the best of both worlds into this conversation. And then you got me, your emcee, Adam Mayer. I've been in Qlik around the global products team. So always excited to talk about data and products and how our customers are using them. When it comes to my football knowledge, I'm lucky if I know my offside from my elbow. So I'm really glad to have you two here. So let's get the conversation. Welcome guys. So we thought we'd have a bit of fun here with an app that we built. Obviously, we're a data analytics company. So we've got an awesome demo team in Qlik. So in this app, you can actually choose your teams, choose the grouping placements, the playoffs all the way through to the final. So it's a great way to kind of predict, use data to predict the potential outcomes, but also create a little bit of friendly competition as well. And we'll be sharing this app out. We're going to walk through it in a minute in terms of how you can use it. And you can see the results that I've done. I've purely gone on it from using our predictive models that we put in. What is used well over thousands of matches since 2010 to train the data and we'll delve a little bit more deeper into the models behind there. But as you can see here, I've built up the playoffs. I pretty much, most of the time used the predictive elements that we got from the model all the way through until I introduced my own little biases here coming from the UK. Obviously, I've put England to win because they will do, right? But you can have your own fun here. You can build your own brackets. We're going to get Steve and Nick to do that as well. We'll share it out. And we have a nice little kind of competition running where near the end, we go for about three different episodes, probably throughout the championship. And at the end of it all, once the final whistle's blown, we'll go through and pick the top three winners of the closest predictions based on your human intelligence as well as artificial intelligence. And the winners will get a team jersey of their choice. So that's something to play for. So let's dive in and just show you kind of how you can build this yourself. And we'll talk a little bit around the AI models in play and obviously how that can work in the real world scenarios as well. So maybe if we just pick the first group in Wadi and we can show folks that are watching this, just a quick walkthrough on how you can start building out your team. We won't go through all of them, but you can do it manually here as you can see Wadi doing. And you can also look in and start seeing the stats that we pulled on all that kind of training data to see what are the most likely kind of outcomes for those teams. And we also can go into a fair bit of detail around the explainability using Qlik solutions here. All of this is Qlik solutions and we're just using public data, a whole host of different type kind of training data elements that we put in to bring all this data to life. And that helps you to kind of bring that human intelligence piece in as well and start playing around with the data and choose your own outcomes. And it's a nice little kind of what-if analysis that we can do in Qlik on the daily. And we put that into here so you can have a play with it. But that's a quick overview of the app. Maybe we can dive a little bit deeper in there in terms of, if I bring in Nick and Steve into the conversation now, what do you think in terms of the whole AI piece? Can it really just rule the world and make the decisions like pretty much I've done on my piece here or there are certain elements people kind of need to be aware of, not only in the real world, obviously, but how can we, are there tips as well on using the app? Yeah, I'm happy to jump in there, Adam. Obviously, we've tried to collect the best data we can to help power the predictive model here that people have access to. It does include 14 different features that made the cut. I would categorize them into three different themes, if you will. One is the overall strength of the squad as it stands today or on match day. And as Wadi may demonstrate, there's the ability to play a little bit of what-if, if a player is injured and somebody else has to come into the squad, the starting 11, what does that do to the probability that they'll win the game or they'll draw, et cetera. Interestingly, this actually comes from Electronic Arts, which is a video game provider, and they do a pretty robust job of collecting individual player data across a variety of different metrics, eight different categories, as I understand it, 5 million plus data points on players across the globe. So it is actually a pretty robust understanding of the individual player. But then you pair that with two different other themes that are in the model. One is recent form, what I would characterize as momentum in the sporting world. So last 10 games, how have they done? What's the goal differential? And then overall strength would be the third theme. That's more of a durable over the past number of years, how these teams performed. So you're combining these things. I think it gives a pretty good overview of the elements that we can capture in data. And obviously, as we'll get to in further discussion here, there's a lot of things that can't be accounted for in data with regard to sport and football being no different. Yeah, absolutely. Thanks for that overview, Nick. That's a good little segue. I think you might want to be jumping in here, Steve. Like you're coming in both for your on the pitch experience, but all the data that you get to kind of use on the daily across the Premier League, there's a lot of variable points and areas as Nick mentions that you just can't capture in the data. So yeah. I'm a huge fan of AI as a tool to supplement the knowledge that's out there. It can, you know, it can compose vast amounts of data and provide predictions and provide insight and provide creative thoughts. And, you know, this application is no different. It's an amazing thing. But I really hope it's not a good predictor because I think that's the beauty of our game is its uncertainty. That's one of the key things about football is that uncertainty, which I hope the model can't determine because it's working off of facts that have been, you know, things that have happened. So I'm a big fan. I also think that, you know, people build in their brackets. If everybody just trusted the AI, everyone will come up with the same prediction and then we wouldn't have much of a game, would we? So I hope everybody gives it, you know, adds their own personal thoughts, their own human intelligence to back up and support. Yeah, by all means, use the AI as a guide, but then tailor it to your own thoughts and see where you are. And I'm sure we'll have, you know, talking through the, you know, how the groups have been formulated. You know, probably the group winners are easier to predict. If you're looking to be a bit different, look at positions two and three because that's the place where the model will give you some ideas. But there's much more uncertainty, I think, in positions two and three. So, yeah, that's a really good point in terms of, you know, AI being a guide and not having all the data in place. So, yeah, AI is obviously only as good as the data and the foundation that you build the model on. But we have got some interesting elements in this particular model. And I think, Nick, I can hand over to you, maybe just peel that one layer of the onion down in terms of some of the interesting factors out of all those variables that can be at play that we've captured here. Yeah, I think Steve brought up a good point with regard to, you know, using your human intuition as a complementary aspect to what AI may be giving you from an empirical basis. It's just worth noting, customers, we see this quite frequently where they can train a machine learning model and get from it empirical validation. Or, you know, frankly, in some cases, it's counter to what their intuition may have indicated. And so I think the two work well together. If I'm going to pull it into group A and we start focusing there, given that they're kicking off the tournament, you know, we did have a environmental factor that we brought into the model. It shows that historically, the further a team traveled to a match, the worse they do. I think that probably makes some sense, given travel and jet lag and such. But with regard to group A, there are some other environmental factors, I think, that maybe aren't captured in the model that we will want to account for, i.e., Azteca Stadium in Mexico City is at 7,000 vertical feet, 2,200 meters for our international audience. That's a non-trivial altitude to be playing a match at. And while some of these teams are being, I think, recognizing that and they're training at elevation, certainly Chechi is on the other end of that where they're training out of Fort Worth, Texas, which is only at a couple hundred feet. So again, maybe there's an element there that's not captured in the model where they're going to be a bit more susceptible to that altitude than some of the teams that are trying to acclimate by training at that elevation. You know, the other thing I would call out is, just historically speaking, the World Cup has been hosted in the Americas eight times, going back all the way to the very beginning, and seven of those eight times, it's won by South American teams. So there is sort of that home turf feel and advantage historically. We'll see if that plays out this time around. Yeah, absolutely. And yeah, not forgetting, just playing in your home turf as well, you've got the home crowd behind you. That can make all the difference, right, Steve? Yeah, I think it can work both ways. If you've got, you know, if you're expected to do very well and you don't, aren't doing particularly well, the home crowd can turn on you a little bit. I think, you know, Nick is clearly right, seven out of eight, you know, times as people, well, that means one out of eight times it wasn't won. So that was why, you know, European team has been successful. That adds an element of unpredictability in there. So, but I'm playing the devil's advocate a little bit here, which is my sort of role in that. And, you know, if we look at this group, and it was the point I was trying to make earlier that, you know, Mexico are the strongest team in that particular division. So regardless of where they're playing, what altitude you'd expect them to win. And in this particular group, I'd be looking at positions two and three, does he, who's going to be the automatic, the second automatic qualifier? And who's going to be third? Is it going to be Chechya? Is it going to be South Korea? You know, they're probably the two strongest of those two teams. And then you look at those two teams and think, well, the strongest parts of both those teams are their forward lines. Are they going to be fit? You know, are they available? You know, how are they going to get on against each other? How can they, if they are, you know, how are they going to get on against South Africa and Mexico? Are they going to do, you know, steady against Mexico and do very well against South Africa? So I'd be looking at all those sort of human elements to it, as well as the AI that, you know, Nika and the fantastic models that are backing this up. I think once we get to the, you know, the knockout stages, then the environmental issue factors, the travel factors become more apparent because people have been able to pick their training bases relatively close to where they're playing their qualifying games. So fascinating, all the various parts. And as I said at the start, I really hope it's too complicated for AI. I think we set ourselves a big challenge here. It's almost trying to predict the unpredictable with all those factors. But it's still quite a fun way to kind of bring data to life, to bring kind of click to life and get more folks using it. So just to recap, then we'll put the link in the description, wherever, whichever channel you're watching this on, you'll be able to get access to the app. We'll start doing some social shares as well. So a little challenge to Steve and Nick, you know, build your own bracket, your groupins, let's get it out there. Let's have a little friendly competition. You guys have got a lot more experience than me in terms of making more human intelligence decisions. I'm lucky if I'm barely human. So that's two out of one out of two ain't bad, right? But yeah, we'll share the link so you can get access to the app. You can have a play of yourself. Just to recap, we are doing a competition. So folks can, the top three folks I think it's going to be, will be able to win a jersey, a team jersey of their choice. So it's not necessarily tied to who you think is going to win versus who you want to win like me. So I think that's a good way to kind of wrap up. We're going to have another session where we'll probably near the end of the group stage. Once we're a bit further in, as you were saying there, Steve, that would be a good time to kind of regroup. We'll see how our predictions have fared. We'll have more discussions around the model AI and kind of everyday use, because I think if I can summarize what we've talked about today, it is really about using data and all of the data points that you can get to be able to build and have trust on the models that you're building. So you've got a really reliable foundation to have reliable AI, because AI is only as good as the data you're putting into it, right? But that all-important human intelligence really plays a part there. It's that kind of experience, the intuition, and all of those outside variables that aren't necessarily captured in the data that you've got to play with. And I think for the next episode, we'll delve a little bit deeper into that, what happens when the model goes wrong, maybe, and other things to kind of be aware of. So thank you for listening. Thank you, Steve and Nick. Real pleasure to have you on here. It's been a great discussion. Really looking forward to the next one. And I'm really looking forward to seeing all of those brackets shared out on your social platform of choice. So Steve, Nick, I'm looking at you. Thank you very much. And we'll see you next time.
