Episode 44: Why Sportsbooks Are Buying Not Building Pricing Expertise

[Music] Hello and welcome to episode 44 of Zero Lazy, the podcast from Ilas and Cry Check gaming. I'm your host Brad Allen and we are brought to you today by radar, the new Geo Location Provider for the US gaming industry. My guest today is Tom Daniel. Tom is the SVP of trading at Hattle. A formerly of Cambie very recent 17 years so we are going to talk some trading today and I trading when we had Trin Halon. We had Shipper on back in the day. There were some of our best listened episodes so we thought we'd go back to trading some automation stuff like that. So Tom let's just start with your journey from Cambie to Hattle. Cambie you know 700 traders I think, Hattle 60 staff in totals. So you know what what's the change in the light? Different culture going from a big organization to much more than one. Yeah I mean Hattle we're not quite a startup but we're not far off so we very much have that culture. I don't necessarily know everyone's job title but everyone gets involved in everything and there's that enthusiasm and momentum that you can have when it's a small organization. The large organizations as they grow you have to subdivide things. You have to break things up and then you have silos, you have a bit of bureaucracy. It's the nature of it and having stepped out of a large organization I'm really enjoying working in that sort of quasi-starter environment, the pace we're doing things mind blowing to me and yeah we focus on a small niche but perhaps we do it to a higher standard and I think the sort of intellectual challenge of that is really really exciting and that's what gets me a better morning. So very much enjoying it. You've been in B2B then I guess nearly 20 years so what are you seeing change recently about what operators want what customers want from a B2B provider? That's a nice thing. So Camby was kind of B2B by accident. The sort of genesis of Camby is that somebody came to uni but I said can we have your sports book? So Camby as a concept didn't really exist before the first customer existed and at that time it was like okay well this is the product it's defined by this customer who wants it let's go out and see who wants it and there was a demand for it and that's that's how it grew. But I think as that nature of suppliers grew and there was more of them and that ecosystem developed what customers really wanted started to change. The kind of monolithic fully managed solution here it is take it or leave it. That didn't really fit the demands of the bigger operators in particular and that was very much because the customer wanted control they wanted to be managing and in charge of that relationship and it's a single supplier, supplies everything to you there is an imbalance there and then the large operators didn't like it they want to be able to configure their various suppliers, have more control over it and then we've seen the kind of modularization that's happened over the last few years and that's only going to continue as the big operators get bigger and bigger and the tier two, the tier three is really struggled to take market share and presumably it's much easier to create a modular B2B solution from scratch or rather than reconfiguring a one stop shop platform. Absolutely if you have an existing solution that wasn't built to be modular it's very very difficult you essentially have to go into the code artificially create separation that's a very time consuming product and process, the flow of the data needs to change, process needs to change everything needs to be reconfigured so that you can subdivide it whereas if you're building something with a view of making it watch out on the start that's a much more straightforward process you always have that in mind and so the end solution is a lot better once you build it with modularization. This episode of Zero Lensy is sponsored by Radar the next generation of geo compliance is finally here. Discover why four thinking operators and vendors choose Radar for cost-effective geolocation compliance and location-based app experiences. With Radar you get unparalleled pricing performance ease of integration and support. Learn more at radar.com/gaming and thank you Radar for sponsoring this episode. So we've been talking before and what we've seen in the industry in recent years is big tier one operators you know that product everyone wants is player perhaps is S-E-P is in place it's my market but we've seen operators in the US specifically look to buy this in for bad and accurate form landstrum more recently and was it sports IQ over-drugings so why is it that the smaller more start-up e companies are able to build this and operators maybe maybe not come but just prefer to just buy them up rather than do it themselves love. Why is it such a challenge? I think you have to take a very long term view so the biggest challenge is designing the data architecture and building the data architecture which isn't very sexy and exciting but that's the biggest challenge of creating the modern trading solutions. The amount of data we process you know is vast the trader hits sim it's got to be published within a second or two seconds you know and that is difficult that takes time and understanding and making a choice on what technology you need today for your solution to be market-leading in three years time is non-trivial and is the element of risk associated with that and large organizations generally don't like risk. What they should do is find two or three different start-ups and pick the winner but they don't want to do that because then they've backed two loses out that three and it looks bad on the balance sheet so instead it start-ups much much happier with risk you're taking the chances of this they can take a longer term view they're not distracted by the day-to-day operational difficulties of running a bit of sports put it down. After where about the compliance with regulatory stuff which always comes in and just demolishes your roadmap you know however however detailed you plan that over the next 18 months some regulatory three will come on take up your top three slots in your roadmap because you need to deliver that quickly that's not a challenge that the small start have they can concentrate on a long term view they can have a small team working that you know that doesn't have to deal with the bureaucracy doesn't have to overcome the silos of large organizations they can move quickly they can iterate quickly which is perhaps the most important thing and that's why they're now producing some of these more exotic more complicated products that the big operators have really struggled to build in house so I don't think it's necessarily surprising to see how the likes of simple ban act and angstrom have been cobbled up I think from our perspective though what makes it really interesting for us is this is now created a gap in the market we are supplying a similar product to some of the successful suppliers who've been acquired acquisition is not our strategy you know you look at the deal we've had with radar we can distribute our odds to their unified odds feet that means that customers can have our odds with absolute minimal integration difficulty they're already integrated to radar so we can supply our products now to existing customers of these suppliers and you know we're aiming to to grow our business accordingly and I think that yeah with the radar deal and what's coming up over the next few months for us I think our footprint is like to expand fairly quickly so not pursuing an acquisition strategy which I assume is the case until you know someone makes sure not for being enough there's there's there's always a limit no I think for a price I think I think when you are a growing supplier like we are you're just focused on building the best product you can that's possible that's what that's what customers want that's what someone looking to acquire you may be but I think the opportunity for us now is you know wide-scale distribution of our products and I think that is the pathway for us at the moment that we're looking to pursue if someone dangles a big check you know of course you never say never but who is there out there who's going to do it and I think maybe people are pausing seeing okay it's been a lot of acquisition but how are the operators embedding these new companies that they're requiring how does that translate to an improvement in their product you know how are they able to retain the key staff in that there's lots of challenges associated with that you know and I think people are perhaps pausing seeing how that's going to play out as well and also you know we are getting online with more and more customers people looking to see how that pans out so you know it's a rapidly changing industry M&A seems to go in fits and starts it's a bit of a fashion for a time and it quietens down maybe this is the peak of the way but maybe the way the bucksation is going to continue and that's probably well above my pay grade so Seas' buying zero flux was another acquisition in this area we can mention but from we have seen you know we have seen an angstrom product start to roll out and the MGM we did some product testing on that that eyelers and it seemed to be a pretty clear upgrade and obviously the the band-ac stuff on points bet and fanatics now has been is scored well in our testing as well. So I think there is some evidence that this kind of thing works. Obviously, you need to go into market share at some point. Right, now one of the reasons I wanted to talk to you today was you publish this week a kind of a white paper in article, automation and human interaction in sports betting. And I'm a sport by sport analysis. So from the top, will there ever be a day when sports trading, sports betting is completely automated, there's no need for a human intervention. No, because sports betting is inherently subjective and subjectivity leaves room for a degree of human input. You know, we're trying to predict the likelihood of things happening in the future based on limited data sets, limited sample size, you know, just purely looking at the analytics, will take you a long way. But, you know, a human being who can extrapolate from limited data sets, join the dots, you know, will be able to refine the output of analytics, I believe, always. And I think we as an ethos always want to have experts adding input, you know, not just managing, you know, overseeing the output of our analytics, but actively contributing towards it. So I remember, obviously, Huddle has this history with we met, Matthew, that with now Edmitha, and I'm reading like an ESPN article, and it was like a day with, what was it called before Huddle? What was that project? Project Prism, that was it. Yeah, you know, a day at that Prism, that they traded NFL Sunday. And they were set down with Davide down, and the saint he was trading, which is Saint to us, is the Broncos as someone. Saint Steincourt, that went down, and I remember, Davideau, in this article saying, I think Trevor Simion is better than the market. And so he was massively one sided. And this is a color we were talking about, that, you know, the model has an opinion, but a trader can theoretically override it if he thinks his opinion is better. But I suppose the trick is, how do you know long term that your trader is better than the model? Can you track that? Yeah, absolutely. So we look at every time a trader changes a parameter, and we can look at what's the link between how he's performing in terms of uptime, it's a margin, and how he's interacting with the tool. Which parameters are he changing? How frequently is he doing this? How does that compare to other traders and what his overall performance is? A lot of these instances, like the example you just gave, are very, very specific and very contextualized. So in context, you have to always consider, you know, what was the specific circumstances of that interaction. But there's a clear link very often between the actions that you're doing and the overall performance. A lot of times you should kind of see this evolution of traders where they start off not really interacting too much with the model, because they're someone experienced, they tend to interact more with it, and actually sometimes their performance can decrease a little bit, because they're overplaying it, and then that normalizes, and then they gradually start, you know, really adding value in key moments, which is what we want them to do. It's not really, you know, a lot of people experience with models is, okay, traders are interacting with the model, but what they're doing is compensating for known flaws in the model. Okay, in this situation, we know the model isn't doing this, late in the basketball game during garbage time, you know, our scoring rate is way off, we need to fight against the model. So it comes with a fairly mechanical process, where in this situation, the trader will always do this, and you're very clear before and after. Before the trades were doing this, this was the overall performance. After performance improved, okay, we want the traders to be doing that. Our philosophy is, okay, we'll just adapt the model. - We'll get the product. - So we'll get the product. - Yeah, but to do that, you need really tight feedback loops between traders and the quants and the devs, you know, and the product guys, you know. It, it, we want to have a process whereby we get feedback from the traders, based in Vegas, and the next day is ready for them, ready to go for the next game, and because we're a small organisation, we can do that. I mean, I've genuinely blown away by the frequency which we iterate that quickly. It's certainly not what my experience has been. I think what most people in training organisations experience is, and that's one of the benefits of biggest small organisation. We can be that flexible. You know, we are not beholden to a roadmap that is this monoliths of cars, this stone that we can never change, because that's what everybody has committed to. - I think one of the things you wrote about in your article was, the pursuit of more uptime, you know, trying to get close to 100% in NFL say, and that's in, in moments where the sport will be traditionally suspended or markets suspended, say a thumble review or, you know, sporting a fourth down, your traders can take over and toggle some dollars. So could you just kind of explain to me literally what that looks like when? - I mean, it's not super complicated. So we have obviously the data feed that is giving us events day, we are driving the pricing based on that event state, and if there is a thumble and it's pretty clear which team has got it, but the scout needs to wait for the on-field referees to make that confirmation, our traders can override the game state as they write. We know this team has it, we know they're going to have it on this, this yard line. Okay, let's go, let's carry on trading and get the game up before the scout confirms that. I'm sure most organisations have that as well. Our traders are watching the low latency feeds, so you know, it's happening during reviews, it's happening during challenges, and it's that concept of marginal games that happens relatively frequently during games. Each of your instance may already add up to 10, 15 seconds, but over the course of a game, you can add up to several minutes of overtime, and it's important for us because that's one of our key metrics that we are challenging our traders, you know, we want to have minimal time, we want to have the best user experience. We have what we would argue one of the best football models in the world built by Matt and Ed, let's use it, let's trust it, let's get the prices up there, and let's give the best user experience we can. Do you have an industry best standard, like do you look at and think we want to get closer to standard, or whatever, or do you think you're already top? I wouldn't say we're top, I'd say we have an aspiration to be top, we are comparing ourselves to the big operating out there as well. It's a working progress for us, but I think if you look at historically from the deck prison days as well, you know, high up time was the key indicator, and getting up before everyone else on watching the market comes to us is something that we take particular pride in, and we place a lot of importance in as well. It's everyone can wait to a 360's five-go up and then just say, okay, we use that as a baseline, but we want to be able quickly, and we want to be a part currently. How long does that last, like, I know you guys have written blogs saying we were, the market came to us 112 times out, 106th-deal, whatever this, you know, in this game or season. How long does that last, like if you're consistent, you're right, surely eventually the market just copies you. Yeah, yeah, I mean, I think copying, you know, they're using user reference, if you are up and you're way out, you know, it's going to recognize that and everyone's going to be okay, well, maybe we won't place too much trust in them as well. People tend to look at this certain market making books out of their reputation. I don't think anyone's going to change their opinion quickly, but are, you know, long-term objective, it is to be one of the sources of truth for the industry, particularly for, for, particularly, you know, for the big games. And you have this idea in your post-trading is a weak link system, which may have made me think of like an NFL offensive line life. There's one hole, there's one, you know, there's one flawed player in that line. That's where the pressure comes from. That's where the system breaks. And so, trading's like that as well. So, talking through this idea and then how that affects how you trade. So, you have to know where your weak, okay, our model's not great at pricing. This market or this situation, okay, in those situations, do we suspend, do we increase the whole percentage, do we reduce limits, you know, that's sort of ongoing, trading consideration that everyone's managing. And that is a, you know, it's a war, continually against the sharp guys who will find a weakness, exploit it, okay, we need to adapt, until we can adapt the pricing, how we're managing the sort of risk management aspects of it. Yeah, I mean, you need to be aware of where you're strong and where you're weak. But I think how it relates is kind of this chain link concept of, if you're going to move a chain link fence forward, you can't just move one part of it, you need to move all of it forward in one go. And this applies when you're building a new modern trading system, you need to have data architecture, you need to have excellent quench, you need to have the main experts who know the sports, you need to have quick development with quick iteration as well. And to the point we were talking about earlier, this is where small organizations, I think, are much better suited to building these systems, because you know, if you are a legacy operator with legacy systems, it's all well-in-good saying, okay, we're going to be AI first, we're going to have AI strategy, we're going to have a little lows of machine learning models. Okay, but can your architecture handle it? You know, how long does it take you to run a simulation? Okay, you can't just have some guys churning out models in Python, part of it, you need the architecture, you need the guys turning out the models as well, but you need to collect the right data. And so the legacy operator is finding really, really hard to move all aspects of what you need to build a modern trade system forward as one, and that's where they tend to struggle with. There's small organizations, you don't have any legacies, you start from a completely blank piece of paper, you say, okay, this is where we need to be in three years time. We have a plan today we can commit to where we're going to be there, and that's what's happened over the last three or four years with the switches, for like, use gameplays of the world, and that's where they're really starting to sort of reap the benefit of that, and that's why they're able to produce some really, really interesting products. As some of the systems, you have some of the hands-reetraiding, you can see with these guys, it's very, very impressive, you know, like I say, the depth, the sophistication of our football and our basketball models really amazes me. Some of the data points, I still argue with traders, how can that be significant? How can that piece of information be significant and relevant, and they got you, but it's two or three percent, but with these other things as well, it all adds up, and it's all relevant, and I'm not going to argue with them, they're the experts, but there's remarkable. There's many examples, many examples of super niche inputs in the day. I mean, we consider the average air density in each baseball ground. I mean, it's pitching in Denver, it's different from pitching in San Diego. The conditions are different, that's not revolutionary, I could tell you off the record a few of our data points for basketball that you will challenge and not believe, but I can show you the dates for it. This is a lot of, you know, Matt and Ed's legacy and they're there, remarkable attention to detail. It's amazing. I remember an article on ESPN a couple years back about the decline in NBA home field advantage, because they reckon players were using Tinder more than just going out and getting drunk. I wonder if that's in the model somewhere. So one of my thoughts and concerns is if things keep getting more and more automated, then I feel like there's going to be more and more holes for people playing the different game. You know, sort of people eyeballing things, watching sports, who, you know, tracking new stories and stuff. So is there a danger of over automation that sort of creates then some holes for the human betters? I think full automation is danger and the nature of automation you choose is danger. You want to build some sort of highly complicated multi-level neural network to solve a problem. Okay, that's all well and good, but you don't really understand what's going on in that model. And if the model starts a diverge from where you want it to be, you're going to have a hard time bringing it back on track because you're not going to know how to manage it and what you need to change. And a lot of these models, you know, they're based on ASX, which is very powerful and very useful. But machine learning isn't great at extrapolating from limited data sets. And we always have limited data sets. There's not that many football games in a season. They just aren't, you know, a rookie quarterback, yes, he's played at college, but he hasn't played that many games. So you don't have a huge amount of data, which means you need the humans who can extrapolate from that limited data because the edge cases happen all the time. With limited data, overfitting is a real, real problem. How do you know if this is genuinely a lumpy distribution like, you know, the points in football games are because of the key points, the trees, the sevens, or it's just a bad distribution. What you do in the context changes, you know, they change the past interference rule or the implications of that because we need to know before we have any data for the season starts. So you need that human with the insight and intuition, you could say, okay, this is the consequence. This is what's going to be, if it's snowing and you can't pass the ball anymore, okay, we start off with this game. People are going to watch it and they're going to want to bet on it. But what is the scoring rate going to be, and if a human who knows what he's doing has a best guess, you know, that's better than nothing, and you don't have any data. So pure automation gives you nothing. So more automation has produced, you know, better trading, more markets, uptime is much, much higher as well. But it's a tool. It's not a means to, you know, it's a means to an end. It's not the end itself. And, you know, you always want to ally the and pair it with genuine experts who really know what they're doing. That concept of man plus machine beats either man or machine is very, very true. I remember there was a Buffalo, New England Patriots game in the snow and about 40 miles an hour wind. And I think the models have no clue what to do with it. I think if it's 2006. And, you know, bet builder unders would have made everyone for shoots and I'm sort of sure a few people got away with them. Yeah, absolutely. And those outliers are, they're someone of the most fun times to bet. You know, when Germany scored four goals in the World Cup semi-final and David Luis is crying on the pitch, you know, everyone's like, well, how many Germany are going to score? Everyone's piling into the German overs. You've got to trade it. But there's no machine reading the body language that Brazilian players as much as you'd want this bit. So you need the person to go, okay, this is done as a contest. But let's try and have a decent projection for the goals now. Yeah, yeah, like that's the thing in the NFL playoffs as well where teams just give up all like all teams don't stop scoring because, you know, the other teams are out, dreams overseas instead. And the other team thought we're not letting it up because it's the playoffs. So it seemed like, you know, big fingers cover off, which again, we have to probably don't have playoff games for that to be. You have the perverse incentive as well, where teams tank deliberately because they want the draw pick. Okay. It doesn't happen, but it does happen. So what does what does that mean? Yeah, this sort of outside context is really important. If a coach is under pressure, he's going to go for it on for on seven when he shouldn't, when he had to exceed, he shouldn't, but he really, really needs to win over his last job so he doesn't care. You know, how do those factors play into it as well? And that's that's where the human steps in and says, okay, I can add more context than the data as I know what data goes into the model. And I know what data doesn't go into the model. And I can add another level of refinement into it because I'm aware of that outside context. Yeah. Last example, I can think of what was like things maybe not a model with it. It was better to can break the cover 60% against the spread for 20 years, whatever they did. And then now I'm the ret is to be covering six percent against the spread. And I thought the best example was the Super Bowl where it renew the playoff double overtime rules. And Shannon and they knew that they had all the time. They didn't, the marathon part right now. And to me, it was just like that, that was part of the reason he was such a good coach of coverage spread because he knew those little details that he just coached out coach Shannon and that, that tiny thing. Anyway, right. So I'm, I'll let you go. I very much appreciate your time. And good luck with everything over at Huddle and keep growing that business. Two, our list is if you would like to hear more from Eilers and Criter, please drop me a line. I linked in and I'll point you in the right direction of all our research. Thanks for listening.