How F1 Processes 1.1 MILLION Data Points Per Second — Ruth Buscombe, Motorsport Engineer and Presenter, F1 TV

0:00Raoul-Gabriel Urma:All the best strategic advancements have come from crushing errors. They're the ones that are the best teachers and I think as a leader, Toto Wolff said it best, where they won everything that year, everything went wrong that could go wrong in this. They said, you know, this is the day that our competitors should fear us the most because it's the days that we fail, it's the days that we get better. It's a lot harder to learn from your mistakes when you're covered in champagne.

0:25Ruth Buscombe:Welcome to Data and AI Mastery, the podcast where we bring you cutting-edge insights, practical advice, and inspiring stories from the leaders shaping the future of data and AI across the globe. I'm your host, Raoul Gabriel-Urmer, founder of Cambridge Spark, the leader in transformational data and AI upskilling, career development, and progression. In each episode, I will be diving into real-world case studies of companies harnessing the power of AI to drive innovation, reduce costs, and create new business opportunities. So whether you are an aspiring data scientist, AI engineer or seasoned executive, this show is designed to give you the tools and knowledge to stay ahead in a world where data is transforming every aspect of business.

1:09Ruth Buscombe:Stay ahead, stay inspired, stay masterful. Welcome to Data and AI Mastery. Hello Ruf, how are you?

1:19Raoul-Gabriel Urma:Fantastic, how are you? Old friend?

1:22Ruth Buscombe:Yeah, great to see you. It's been a little while. I was trying to remember, it's been over a decade since the Cambridge days.

1:29Raoul-Gabriel Urma:Yeah, it feels like a long time ago, but also not that long. You know, well, I'd think we're still young.

1:33Ruth Buscombe:That's true. Time flies.

1:39Ruth Buscombe:I'm so excited to talk to you today for a few reasons. One, I think your journey has been so inspiring, starting with engineering in Cambridge and getting to F1 as a race strategist, for Ferrari, Sober, Haas, and now in F1 TV. So just an amazing journey, so I can't wait to talk about it. And the second reason, I'm a big F1 geek myself, and I think there's a lot we can learn from what's happening in F1 for businesses around the data and AI. So can't wait to talk to you, Ruth.

2:09Raoul-Gabriel Urma:I'm looking forward to it. Let's get into it.

2:11Ruth Buscombe:All right. Well, look, to kick us off, maybe to paint the picture a little bit, But can you tell us about the scale of the data world in F1? Like, you know, what's the scale, the velocity, what's the variety of data that you have to deal with on a day to day?

2:27Raoul-Gabriel Urma:So Formula One, you know, we started from humble beginnings when it came to data. 75 years old this year. Obviously, we initially our data point was a stopwatch and a lap time in a muddy Silverstone field around 75 years ago in our first race. And like many industries, you know, we initially started off purely mechanical cars designed with pencil and paper. And in the 90s was kind of our data revolution. And we went from pretty much having not enough data to having too much very rapidly. So the first time a Formula One car ever had any kind of monitoring or live telemetry on it was in the 1980s, 1986.

3:03Raoul-Gabriel Urma:It was a McLaren. It wasn't actually even a Formula One car. was an Indy car in the States and it had 14 channels on it you could only get the data after the race when it was downloaded it was McLaren the first started having a proper live data on a Formula One car and for want of a better word it was terrible so 1986 the Australian Grand Prix told Alan Prost he'd run out of fuel like every good Aussie he ignored the instructions carried on to win the race but from those humble beginnings we've got a lot better and now you I don't know, our issue isn't not enough data or not correct data. It's actually trying to understand the data that we need within the sea that we have.

3:41Raoul-Gabriel Urma:So our data challenge is quite unique. A Formula One car produces around 1.1 million data points per second. We run 300 sensors on each of Formula One cars. Each Formula One team has two cars, unless one of them is broken down. and that data is being beamed live from a racetrack at around 200 miles an hour and a lot of the data insights aka which of those 1.1 million data points are actually useful for us in terms of trying to make the car go faster trying to react to a strategy decision try to optimize the driving and all of a lot of that work is done not locally at a racetrack so for example we've just come back from Singapore or we go racing in Melbourne but up to a 10 ,000 mile trip away with a lot of the teams or all the teams are based in europe um so we actually beam the data so being there's 1.1 million data points for two times of our cars uh from a race car that's traveling um it's about 10 you know 20 000 mile round trip when we're going racing in melbourne um running on teams you know we're various different kind of cloud partners ai systems uh to generate insights and back to the pit wall back to the engineers that are going to communicate the insights so the actual trained models output in less than a second so there's there's a lot in there right so as a formula one team it's about you know reliability so data good data centricity

5:07Ruth Buscombe:good reliability wow that's fascinating because you've got a combination of a high volume but also low latency and you made an interesting point clearly it's a lot of data collected lots of sensors can you give a perspective around uh how do you decide what data is worth collecting or not because i guess there's several school of thought one is let's just collect everything and we'll figure out later it might be handy but often the business world is you know you want to be precise about what you collect kind of aligning it with some business outcome so i guess how do you make this decision of we should collect this data on this sensor because we'll definitely benefit from it versus let's just collect it and figure out later it's a really good

5:48Raoul-Gabriel Urma:That's a really good question. And actually, I think historically, a bit like many industries, you know, we weren't very good at it. You know, we just try and collect absolutely everything. But Formula One is a pressure cooker for marginal gains. So if we take the 2024 season, the average gap in terms of absolute performance between first and last is 1.3 percent. So the leader in the field is only 1.3 percent on average better than the absolute loser. and if taking it even more so pole for example in the first six races of the season so the first place in qualifying uh was set by less than a tenth the japanese grand prix for example where max verstappen from red bull put it on pole uh by 12 000 is the equivalent of 76 centimeters so around a uh over five kilometer track it's about as far away between myself and yourself yeah um in terms of of how pole is defined so when you are driven by a sport where the difference is between success and failure are such small margins.

6:47Raoul-Gabriel Urma:I think it's Emily Bozerup's theorem, you know, necessity is the mother of invention. Actually extracting the key insights that allow you to go faster than your competitor, allow you to recover from, you know, mistakes or errors, allow you to predict, you know, a weather pattern or be able to predict a tyre performance marginally better than your competitor. And the better job you can do, extracting those insights, you know, effectively you know communicating the key changes that you need to do the better you are so it is really driven by necessity so just like every other industry in the world we were really hit

7:22Ruth Buscombe:hard as a sport by the financial crisis you know prior to the financial crisis we kind of had a

7:28Raoul-Gabriel Urma:sport where we had the haves and had nots a lot of the big manufacturer teams and a lot of money a lot of big sponsorship uh that you know all went away um and then we had a situation where you know kind of prior to when the sport was sold in 2017 uh to a company called liberty media um the only team that was turning a profit was scooter or ferrari every other team was in the red and we were losing teams um every year so now formula one is a cost cap so unless you're cheating um we have to make your formula one car for the same price as everybody else so that becomes an efficiency formula which is why actually the data race is a real race within the story so So AI, for example, being able to augment what you do and being able to do more with less, being able to free up the resource that graduates like myself did manually or like, you know, semi-manually 15 years ago.

8:18Raoul-Gabriel Urma:Actually being able to do that programmatically by having kind of modular AI applications to answer the same questions that you were using, you know, manpower for 15 years ago allows you to get that tiny little extra marginal gain in performance. So we really are driven by the level of competition. And the biggest feedback loop in Formula One happens 24 times a year. You know, we don't get quarterly reports.

8:46Ruth Buscombe:I'd love to hear your thoughts around, you know, once we've got the data collected, a big challenge that a lot of business faces, how do you translate data into actionable insight that moved the needle? How does it work in the F1 world? maybe you know either on the racetrack or off the racetrack you know you've collect all of this data it's amazing what happens next like how do you use it to like inform like improvements

9:12Raoul-Gabriel Urma:it's a really good uh it's a really good question again so broadly the way that we use data as engineers goes down into kind of three main buckets so one is improving our own performance um so finding the key kpis based on you know the wealth of experience that we've had before that are going to make a car go marginally faster around a racetrack, for example, or be slightly better in terms of the way that it's using the tyres, for example. So the integral of performance along a stint is slightly better or improve the drivability. That means that the driver is able to extract more from the package that you have.

9:44Raoul-Gabriel Urma:The second kind of pillar is competitor analysis, which is our engineer speak for copy somebody else's homework. So we want to be always looking at what someone else is doing. We have a lot of game theory. So, for example, rear wing level, effectively in Formula One, we have a lot of racetracks, you're kind of like standard racetracks where you can put more downforce on so you can have a bigger rear wing and it will go faster over one lap. So your raw lap time in qualifying, for example, will be faster. You can take the same configuration of car, only change the rear wing, go for a skinny rear wing where the total lap time for over one time lap is slightly slower, but your top speed on the straights is higher.

10:21Raoul-Gabriel Urma:so you can take qualifying position versus race ability because if you have sitting on pole with your massive rear wing you start the race and you're six or seven kph down on the straights for example you'll just get overtaken by the entire field you'll be kind of a proverbial sitting duck so competitor analysis to try to understand the kpis that are going to expose you understand where your sensitivity is um is very important and then finally race strategy which is every good unreliable narrator uh like myself obviously as a person that spent their entire life in my strategy is clearly the most important of the three pillars um is about how do you actually go racing how do you predict the key elements that are going to be the difference between deciding whether or not a strategy is going to be a one-stop or a two-stop or predicting where the rain is going to hit and what tie you need to fit and all three of those pillars comes down to that feedback loop of what makes you faster.

11:17And the homework that we do is born into simulation

11:21Raoul-Gabriel Urma:to really understand our sensitivities. So we do, you know, when you go see a race, for example, that is the kind of the tip of the iceberg. And a lot of the homework that teams do in terms of preparation is, as you say, really trying to understand what are we most sensitive to.

11:36Ruth Buscombe:I hope you're enjoying today's conversation. If you're finding the insights useful, please do take a moment to subscribe to the Data and AI Mastery podcast and leave us a review on Apple Podcasts, Spotify or YouTube. Every new follow helps us reach more people and shed incredible work being done by today's Data and AI leader. All right, let's go back to the episode. And you talked about race strategy and it really resonates because in the business world, any good strategy needs a diagnosis. You know, what's your understanding of the market? what choices are you going to make so there's quite lots of parallel which takes me nicely to actually strategy at the race you know being an f1 gig I love your perspective because you know you're there at the race you know you've got all the data right you can see what's happening on the track your competitors you've got all your models and you gotta make some decisions right like are we gonna pay it we're not gonna pay it I'm gonna wait can you give us an example perhaps where you went against the data.

12:40Ruth Buscombe:Because there's one argument, if the data is so good, if you've got the best model in the world, why not let strategy be entirely driven by the data and the model? Why should we still have a human input into it?

12:53Raoul-Gabriel Urma:So I think there's a kind of twist on that, which is the human experience, when people often ask, how much of it is a data-driven decision and how much of it is a gut feel? As long as my audience is geeky enough, I just reply, well, actually, your gut feel is just an LLM of your own experience on a pit walk.

13:12Ruth Buscombe:That's fair.

13:12Raoul-Gabriel Urma:And actually, you know, as a, you know, let's say someone that's more hands on keyboard, someone that's writing one of these simulations that's going to output the data set. You know, if your conclusions and if your, you know, approach is being challenged by somebody with more real world experience, you know, nine out of 10 times, that is actually a second order effect that you haven't modeled. for example you're missing something so there's potentially some kind of level in that that is is based on you know like an effect that you're not capturing you know let's take overtaking for example that actually you know to overtake one car versus to overtake a train of cars um is not you know the sum of its parts it's very very different it's different for a driver it's different in terms of the instabilities it's different in terms of when you're physically going offline so you know you need to add that to your modeling one out of 10 times when you're talking to somebody that's uh says there um it's the human error element um and actually as humans there's lots of medical papers uh written about it we have a huge anchor bias to extreme events in our memory so uh just like in business you know you don't remember the average day you remember really good moments and really bad moments or really good strategies and really bad strategies uh and often it's it's very very important to you know numerically challenge that anchor bias even my own experience where you know it's very easy to um you know you've been racing a track for 15 years now and you say ah you know the undercut here works really well like i remember we undercut five cars with kimmy raikkonen and kimmy in 2018 but actually for the other nine of them um it was actually faster to do something else so i think challenging why somebody has a different conclusion is really important and in formula one there's there's two parts there's there's making data driven decisions and then there's executing well and the two things are very separate it is always important in formula one to make a data driven decision and you know we're not making decisions on perfect data sets the best way that i've kind of described it is that it's like you're looking after somebody else's portfolio you know no head of strategy uh as of right now unfortunately including myself including has ever owned the formula one team we don't own our own um you know our own portfolio we are entrusted through our skills and our team's skills to be able to um come up with the best menu of options so the best number of you know stock options for example and to equate to the best of our abilities the risk versus reward profile of that and to kind of present that to the stakeholders which is usually your ceo that represents the the team owners for example and the drivers that are able to you know whether or not they can actually execute those plans and so our job is to say you know and and the answer can be different right if you have the fastest car with the fastest driver and you're starting from the front row you know you don't it's silly to take the level of overhead of risk that you don't need to if for example you are starting last with maybe one of the slower cars with a driver that's you know not a world champion you know actually there's a lot of um kind of like you know there's an opportunity cost to not taking risk because in formula one it's only the top 10 cars that score and actually if you don't take any risk uh you're pretty much guaranteed to fail so actually the safest option is usually a bit like when you're building a portfolio to take two different risky options with your two cars that actually overall are covering two ends of a spectrum and that gives you a slightly safer portfolio so you know whether or not that reality actually happens it's it's very important to make data driven decisions because whether they go right or whether they go wrong they allow you to improve the process does that make sense moving forward

17:00Ruth Buscombe:so clearly as a race strategist there's a set of skills that are really important to be successful to execute on the strategy and i can't help but relate that to the business world especially with scale-ups you know when you're growing fast uh you're also competing to to get business and it can be really stressful if you communicate your strategy really well to people in the organization so is there any sort of insights you've got from from your world that you know in terms of leadership skills or skill set to kind of like help with execution that might be useful for

17:34Raoul-Gabriel Urma:people in the scale-up world that's really good i think in formula one there's there's two main things i think when you look at successful individuals within formula one the kind of you know traits that you see over and over again you know from kind of the engineering side the first i think is actually finding that pressure is a privilege and i think if you talk to any any embrace the pressure embrace the pressure and embrace those moments and you know if you find a let's take strategist for example because if you find a formula one strategist that is not looking forward to the race day or a wet to dry qualifying um that's what you're doing it for you know that's why you know if you do every formula one race you're away from home for six months a year you know you're working you know long hours there's much more efficient ways to make money um than to be a formula one engineer you're doing it because you love that job and there's many roles even within formula one that have a very different balance of pressure or just you know everybody's under pressure but different times you know it's and actually if you are not enjoying and embracing that pressure and then i think secondly uh when it comes to to kind of the the traits it's it's creating a culture that encourages failing forward because in formula one you know so it's a weird industry because in most places if you spent a million pounds worth of development budget on something that gave you, you know, let's say 0.02 % improvement in your product that your competitor could copy within six races.

19:08Ruth Buscombe:Yeah.

19:09Raoul-Gabriel Urma:Or sorry, six weeks, for example. You'd probably, and you failed five times and you came up with one idea that added a small marginal again that could be copied almost immediately. You'd be in trouble. In Formula One, you'd probably get a promotion because that actually represents like a significant improvement. Um, and I think having kind of creating a culture where, you know, you are allowed and encouraged to fail and you're allowed to, and it's a NASA term we do. We take a lot of stuff from NASA, like pulling the strings of failure is the way that you make progress and is the way you go forward.

19:45Raoul-Gabriel Urma:You know, like, you know, the whole, we go to the moon, not because it is easy, but because it is hard. And I think, you know, kind of having a culture where, you know, the only way you go forward is by failing, you know, like the whole Edison 83 times to make a light bulb. what would all be sitting in the dark that you kind of have to work your way through the nose like if i told you that if you find 20 different uh ways to develop a you know an element on the floor that's going to add performance in the first 19 would fail but by the time you tried the 20th idea you'll you'll get it yeah how quickly would you work your way through all of the all of the nose and all the failures um and and all the best you know all the best strategic advancements have come from like crushing errors um you know they're the ones that are the best best teachers and i think as a leader Toto Wolff said it best after this uh you know awful Hockenheim Grand Prix they did uh back in 2018 I believe it was um where they won everything that year everything went wrong that could go wrong in this and they said you know this is the day that our competitors should fear us the most um because it's the fate the days that we fail it's the days that we get better it's a lot hard harder to learn from your mistakes when you're covered in champagne I really love that

20:53Ruth Buscombe:that's really inspiring. So combination of passion and feeling forward, like you say, right? Like having this culture of we're trying stuff out and it might not work out, but you know what, we'll come out stronger if it doesn't work out because we'll learn from it and that's going to push the organization forward. I think every startup organization can learn from that. So thank you, Ruf. And I guess maybe like a final question to take you to the AI space. you know we clearly talked about some really interesting use cases um how do you see the future of f1 or even f1 broadcasting with ai what are the sort of things that you're excited about footfield so i think um you know we there's so many different things that are changing kind of

21:38Raoul-Gabriel Urma:every single week like we're coming up with notes i work with um aws we've got new insights where you know it's basically if you can believe it and you can imagine it they can do it which is amazing after you know working within a formula one team where we have slightly more limited resource the thing that i'm most excited about because we have such a deep repository of it um is the the kind of the ai applications to do with video databases um so amazon nova that launched this year kind of when you think we've got 75 years worth of broadcast video and the the ability to kind of you know like uh programmatically look at you know give me every single time that pia gasly was overtaken in the breaking and then you know you can have a go at that you know give me every single time that which you know lewis hamilton wore gold um through a language query exactly um so everything from everything from you know the broadcast but even you know the socials for example you know transcription and actually really you know bringing the story to the fans as well because you know it's very important when you're trying to you know grow a global sport like we are you know and especially you know we're growing very quickly and very rapidly in the states ever since drive to survive um you know 11 year-on-year growth this year alone in our half-term report uh that came out in august um you know when it came to there and our our audience is getting bigger younger and more diverse and it's really important that we are able to use and leverage the ai um solutions that we have to explain what's going on simultaneously to fans that have been watching it for 30 years you know to our incredibly informed fans also to be able to tell the data stories uh to you know a new generation of fans that maybe don't have as much information there's kind of boundless abilities to be able to let you know make it customer facing make it fan facing and actually make sure that you know we are able to capture and analyze what makes our sport great what makes our sport not so good so that we can as a as a as a sport make sure that we're telling the best stories we're not missing anything and keeping our fans entertained and giving them as far as we can what they want to see.

23:43Ruth Buscombe:That's great. I love it because it's really, you can have an opportunity to focus on the customer a lot more. And again, I guess back to business, we focus a lot on do more with less, cost efficiency, automation, which is one part of the story. The other is how do you delight your customer more? And the thing we're describing here is through AI, you have an opportunity to create new experience for the audience, get them closer to the action, feel part of the data and the videos themselves. So I think business can run from that too, right? Focusing a bit more on the customer and innovation.

24:21Ruth Buscombe:Super cool, Ruf. So I've got three very short quickfire round of questions. I love a quickfire. Well, I can't wait to find out the answers to this question. I think the first one is, let's be geeky a little bit. What is your favorite programming language?

24:34Raoul-Gabriel Urma:Oh, do you know what? I have to say Python because I was told I was a boomer. A boomer. I'm a millennial, by the way, not even close to a boomer. By some members of my team because I was taught MATLAB at university. We both went to the same university at Cambridge. And ahead of the pandemic shutdown, you know, I was told that I was a boomer for only using MATLAB and not for, you know, being... Bearing in mind I wasn't a hands-on keyboard anymore and hadn't been for about five years. But I kind of entertained my space in the Python environment and then came back and I was accepted by the Gen Zers.