[Music] thank you very much for having me Uh my aim in the next 20 minutes is to make you all F1 engineers Uh if you're using AI you might even beat HS on the grid And I can say that because I don't work in Formula 1 anymore Uh so a little bit about my background So I uh had a bit of a weird sort of start started working in healthcare I started at Great Orman Street Hospital had an operation there Didn't quite go to plan My uncle said "Let's sue them." He's American Uh and I said "No no I wanted to work with them." So I ended up working with the hospital on how to get patients connected I spent a summer in there as a kid I was pumped full of morphine and ketamine My best friends are off at festivals doing the same drugs just different different set of uh different set of circumstances Uh I loved working in healthcare It was really good fun So I worked with Cisco when I was about 17 years old to get kids Wi-Fi in hospitals Uh stop those horrible over the bed arms with $20 a day Uh moved everyone to getting iPads and Netflix and Disney and and you know things like replacing a premed with an iPad Who would have thought it that was really fun And if I got good at Wi-Fi I then got the opportunity to do that in retail So I worked with the likes of Harrods John Lewis Westfield Was never interested in giving you a Wi-Fi connection as a shopper but actually we could do things like understand your dwell time where you were going in the building And so I fell in love with data and insights And after a little bit of that I thought well who does data and insights really well but Formula 1 and McLaren And so that's where I sort of started my journey Uh this was home Uh the car park never looked that empty Uh and this is where uh the Formula 1 cars start their life So 90% of the Formula 1 car's life will spend its time uh here at the factory I should say my role at McLaren was to look after all of the technology I'm just not very good at it inside the car So Salesforce fan apps websites wind tunnels networks cyber security everything you see on the Formula 1 track I would look after uh if it goes inside the car There was really good and really smart people did that just not me So 90% of the car spends its life virtually They design it in CAD They look at photos of sharks cheetahs lions They get inspired by wildlife They copy each other on the grid Um and they design the cars virtually So 90% of the car's life will spend its time in CAD Uh in the back here is where we ran out of space and we put a mobile data center or a mobile high performance compute unit Uh and that was about $12 million worth of compute Into this virtual wind tunnel you analyze every single cubic centimeter square of what we call the wet surface So if you were to spray a car with water that's the wet surface you can see And so every single cubic centimeter you would analyze on your virtual part And if things look good you would then get your permission to start manufacturing and testing that part physically Now all of that virtual simulation is all highly regulated The memory to core ratio is regulated The FIA come inspect you every year uh it is a really really secretive uh and highly uh performance you know enhancer to the team um and so the cars would go through this virtual wind tunnel and if they look good we then 3D print it and that's where AI starts to start to play a role in McLaren so you've got results from the wind tunnel you generate about nine pabytes a week you delete a lot of it uh you then compare that to a wind tunnel so you would then 3D print the part in resin you do that because you can recycle them you print them at about 60% scale you put them into the wind tunnel and then you run wind over them to see how the cars will perform Uh anything over 8 meters a second from the wind tunnel blades is regulated again So you're regulated how much time and how much money you can spend on your um high performance compute your CFD you're regulated how much time is the wind tunnel But if the two of them start to correlate and match you can start to make some efficiencies So if you were to put a floor through the virtual wind tunnel a floor of an F1 car that takes about a million dollars worth of people's resource time effort and energy to manufacture to to come up with it Then takes a further six weeks to produce it So the team are racing this weekend in Canada It wouldn't be until uh Belgium that they'd be able to get that floor onto this car And that could be you know that could be serious points in championship But if your virtual wind tunnel and your real life wind tunnel start to correlate and the data looks good you can start to skip steps So you can actually start the production of parts by just putting through the virtual wind tunnel skipping the real wind tunnel which lives in here and actually get them onto the car much quicker So you put them through the manufacturing process Uh in any given season there's about 80,000 components that will change on the Formula 1 car That's from the start of the season uh which is typically in uh Bahrain or or Melbourne Uh all the way through to the end of the season 24 races later in Abu Dhabi 80,000 components on the car will change in time frames that's about once every 17 minutes a new part will come out of the factory There's a huge amount of data just in the manufacturing process that you have to get right understanding a manufacturing process How do you carbon laminate how do you do CNC milling what do these machines do all of that is is a real area where we were starting to look at AI as well And I'll get on to the kind of approach rather than just saying AI I'll get on to the actual like how did we do it in a little bit So you spent six weeks in the virtual world You've got some data points You've got some correlation Things are looking good You then manufacture it You go through the the whole manufacturing process You've got people uh you know 3D scanning everything You're trying to correlate your your actual you know virtual design to the physical design Does it match up have you been able to ship it and get it out to the track in time and if all of that works and goes well you end up with this And this is the car in Silverstone last year Now this car on it I got better slide There we go 24 races 19 countries 120 person um trackside team over 60 tons of equipment will make its way to the track That should say 33 because I've left uh 33 people in IT Uh and $140 million cost cap Now the cost cap is like I think really good It's like constraint in any business is a really good thing So if your budget's getting slashed if your people are getting constrained it's a good thing right like I don't think this is is bad news That is where the innovation comes from in our sport Now um $140 million per year on OPEX the driver salaries thank God uh the top three earners travel marketing and a couple of other little bits including engines are all excluded from that $140 million In truth you're probably looking about 280 million to go racing on a year including your marketing costs as well Um you have 40 million in capex to invest over a 4-year period So unfortunately what it does is it moves things like the cloud becomes quite unattainable for the team because you don't want to burn your opex on on cloud resources Um but there's this also this really good little phrase inside the cost cap which says fair market value So we've got someone from Dell in the room Yeah there we go So uh if we said to and I'm at Cisco and well tell the Cisco joke but if we said to Michael Dell hey what if we made that Dell logo really big could you give me a CAD laptop rather than at $5,000 could you give it to me for a dollar if Michael and the team at Dell said yeah sure no problem Make our logo really big and you'll never pay for a laptop for the rest of the year There's a clause called fair market value and that means you have to recognize the fair market value of that laptop that you're consuming into your business So even if we did that even if we made the logo really really big and Michael and and the team at Dell or if we went to Carrie and Chuck at Cisco and said "Hey can I have a brand new network free of charge?" It would not matter You have to recognize the fair market value of the kit you're being given So if I worked for Quickfit or if I worked for Specs Savers and I went to go and buy the same thing that's what you have to recognize as fair market value What that did to us as a technology team was meant we had to pick technology because it was the best Not because we've been given money not because they're giving us you know sponsorship revenue We picked technology because it was the best for us or they had the best people or they had the best insights It was no longer about who can write the biggest check It really became about picking technology for the right reasons The team have been collecting data now for over 40 years So I can go back through libraries when I worked there and I could pick out I could do a Lewis Hamilton versus Sen lab I could get you know Oscar driving a simulator and comparing him to Mikah Hackenan Data has been pervasive through the team for over 40 years Now I spoke about those wind tunnels those virtual wind tunnels the racetrack Well when the cars go out racing we put 300 sensors on the cars So those cars will have air pressure sensors input sensors temperature uh pressures you name it We'll be tracking both um logical inputs and then also the the environmentals around the car The floor of the car alone has 34 air pressure sensors They would collect data at 80 hertz That's 80,000 times a second we're collecting data just from one of those floor sensors The way we get that back is as the cars go around the track at 200 odd miles an hour Uh there's an electronic control unit on board It makes some sensible edge decisions So it's got a little bit of um edge processing and it will decide what to send back to us in the garage Um throughout the course of the weekend we received about 250 million data points from two free hours on Sat or Friday and hours practice Saturday qualifying and the race 250 million data points uh from each car that's sent back around the garage using an adaptation of the Wi-Fi standard called Wi-Fi Max which our sister company McLaren Applied developed in the early 2000s And so it does super fast handoff Um but all of this is governed and all of this is regulated by the FIA So it's not like we can turn it up or turn it off Um lots of people sit there going "Why don't you use 5G?" It's a real ball to negotiate 24 different 5G spectrums around the world I tried it once I'm not doing it again But that's how the car goes racing Every single time we'd rock up at a racetrack this was my my brains of the operation This is our mobile data center on wheels Um it's one of them It weighs 990 kg DHL charged me $250 a kilo And it flies with the cars to each racetrack So in here you've got all the radio gear So this is how we talk to the drivers Remember that because I'll come back to in a second Uh all the coms go to the FIA So if you ever listen on the TV and you hear Max Vstappen swearing at JP that's because they have to share their coms So do we Uh you then have all the routters So this is how every single racetrack So this is where the sort of Cisco journey kicks in Um but whilst you've got 120 people at track it's hot it's humid you've got fans screaming at you It's really difficult So what we do is we have a room back in woking at the mission that big circular building that all that water outside keeps the building nice and cool by the way So you have a room of 34 engineers where they can in a calm air conditioned environment they can call the race strategy So as those cars are going around the track collecting data at 2 megabits per second that comes back into this stack here and I've got all my telemetry servers that will then replay them all the way back to Woking in England in about in Amsterdam it will be 15 millconds latency in Melbourne 284 milliseconds latency Um and so we build that into all the equations all the factoring as well So all of that data then arrives back in woking and actually the decision to pit the cars if you watch the race this weekend in Canada the decision when to come into the pit lane in Montreal is made all the way back in woking in England So that connectivity is absolutely critical Uh in terms of uh in terms of the speed of that connectivity it's 100 megaps I paid $1.2 million for it I have a one minute SLA with a company that provides it But your home broadband is probably less than a percent of the cost and probably about 10 times as quick Um so it's really difficult Uh this picture I took in Singapore last year Uh it was uh 38 degrees Celsius in the garage It was really sweaty It was 92% humidity Technology should not work in those environments It does uh you get it air conditioned Um but everything else in the garage we ship around the world So all the pit walls everything else you see that all get shipped That's cheaper Um and then this this rig we take with us in here uh is a huge amount of compute and that's really important because although you're collecting lots of data and telemetry at the track the edge this could be your factory your branch your stadium you name it Um we do all of the kind of big crunch work in the middle back at the MTC and then burst into the cloud uh when we need to Uh there's some stuff about Cisco garage ah pitull So I said to you earlier you can hear the drivers to the engineers talking on the radio This is the first use case of AI that I worked on two and a half years ago So if you can hear the drivers talking right and I can you can hear Lando and Oscar if you work at McLaren but you can also hear the other uh 18 drivers down the grid you can start to listen in and gain insights So just imagine if if you will if you were like in Las Vegas say like lots of people work in tech So Las Vegas is pretty much second home right if you're doing 200 miles an hour down the back of the strip and you put the brakes on you just imagine four times your own body weight pushing against the side of your head Don't try it on the road Um but you imagine four times your body weight pushing on the side of your head and your engineer by the way only one engineer talks to the drivers to the race They get on the radio and they say "Hey Lando tell me about the tires." Right okay But everyone else is listening to you So then they're listening to what you're saying and they're telling you know you're basically broadcasting how well your tires are doing Or I might say to you "Hey tell me about the tires or tell me how the tires are doing." One of those commands is for fake news So constantly through the race you're then saying "Hey tell me about the tires." Under 4G of pressure on your head you're then remembering "Oh that's a fake command Oh my tires are dreadful They're really really bad." Meanwhile you're setting the fastest lap time because everyone's listening to each other you're trying to trick each other and play out this game of where you're trying to get insights So how do you counteract that well the first thing you do is you use all your people back in Woking and mission control on a Sunday So I'd get in 20 volunteers uh 10 volunteers sorry post room procurement finance HR groundskeepers you name it I'd sit them down I'd put a headset on them I'd put one driver in one ear one driver in another ear and you would transcribe Now if Lewis gets on the radio and starts moaning about his tires I timed it It's 8 seconds So 8 seconds Lewis gets on and goes "My tires are shot They're really bad." 8 seconds later we've heard that message in England We've typed it into a system We've got a language model running so we know what the kind of the good words and bad words are And 8 seconds later it gets to this guy here Randy on the pit wall Well in 8 seconds the cars go over 80 meters So in 8 seconds I'm close to makes a difference 100 meters I'm through the next corner and I'm making decisions where cars were not where they are on track So that was the first use of AI We put it through WebEx calls We put it through Teams calls We put it through Pixel phones I put it through everything you can imagine We started to use AI to do speech to text And that's kind of like that's a good first step right that goes from 8 seconds to 2 seconds Those are the margins you're talking about in Formula 1 So AI really useful there It went from 8 seconds down to 2 seconds And that's what we called an efficiency project Um an efficiency project was like where we want to really save time And that was the kind of how we looked at AI um an efficiency project would um anything that we we'd do that would just be like a manual process that wasn't really smart like you know doing speech to text isn't new you can buy tools out there that do it um so it took us from 8 seconds down to 2 seconds Now what's smart is if you work with lots of the cloud providers you can then do sentiment analysis So if a driver's on the radio swearing about the tires really complaining really giving it some passion it starts to feel quite real if they're on the radio and going "Oh my tires are really bad and I don't really then you know that's not not quite true." And so we do sentiment analysis to start to weight their responses and say "Okay Max is on the radio complaining but it doesn't sound like he's really that upset or that angry about it." Therefore we wait his his truthfulness score as quite low Well that was quite an easy AI project that was around efficiency We worked with lots of uh cloud providers We worked with Google They were working with Burger King at the time Um like they want to remove engine noise from drive-thru We got really good at removing engine noise We worked on that project with Google cloud and they did a really good job there But you know we ended up buying a model online and running it ourselves The way we would think about AI broadly is in four buckets and the first is as an IT like leader is I want to save my team time So use AI coming from all of your vendors to save time as an as a function So that could be AI in security running the network running the storage running Salesforce producing emails whatever it was We used AI to help save time And that's because we knew our business really well as the IT department And if we could save a if we could save ourselves time then we could work on the next three buckets where we'd add performance The first one was efficiency right and I've just described the radio uh scenario The next was scale So efficiency is I know what I'm doing I can I can save some time But scale was there is too much data to look at How do I condense it down and provide some smarts now as the cars go around the track uh the the tires on them that's like a really good area of insight to see how tires are behaving So what we did was we we took some cameras we timed them up with GPS signals and we would fire uh images down the pit wall as cars passed us coming down the pit straight You're only allowed to shoot at eight frames a second So that was regulated So we'd get inspected by the FIA to check what we were doing But at eight frames a second I can capture every single car that goes past me Now first thing you do is you use um that sort of AI challenge You go well that's a Ferrari That's an Aston Martin That's not too difficult Um so then you use AI models You just strip them away and off you go with them But then from there you start to identify right is that blister is that a hard spot what does that look like so you take over 100,000 images during a race and you whittle them down to sub 400 and you can then share those with the the tire team or Hashi So the model is called Auto Hashi Um I would then share it with Hashi and he would then be able to look at those tire images and say "Okay I can see the Ferraris are graining faster I can see that Gaston Martins are actually looking after their tires really well So that was a big scale um uh model challenge What was funny in that one was we started that with a graduate called Joe and Joe phoned me up and said "Hey I've got this idea Do you think we could work on it together?" I was like "Yep no problem What do you need?" He said "Here's my compute requirements." Well we didn't have that compute left in the building We were running at really lean levels Uh it was supply chain challenges So we couldn't get a hold of what we needed really quickly So we we burst into the cloud So I said "Well you know how do we get set up?" So we reached out to the team at Google Cloud We started working with them We text TK We said "Can we have some credits?" He said "No problem." And I got this like 12digit code on WhatsApp and it said $50,000 of credits I Oh this is fun So working with Joe I built Joe an account I gave him the 50,000 of credits I said "Right there you go Off you go Joe No problem." And a day and a half later he phones me and said "Oh it's broken." I like "What what what do you mean?" He said "Well I've analyzed all the last two years worth of tire images and it's just stopped." I like "Hm." So I log into Google Cloud take a look trying to see what's going wrong and he burnt through 50,000 of credits in two days but he'd come to an answer and a conclusion which would have taken us with a graduate a year to produce so I turned around to Laura our CFO I was like right we've we figured it out we we've got this time model can I have some more money and she went what do you mean I said well I got the credits from TK last time can I have some and you know by the way those credits went against our cap so although I didn't have any like money going out the building I start to recognize the cost of it she was like no you can't not not if you're going to spend 50 grand in two days So that became this real interesting conundrum for us So that's why we then pivoted invested into AI clusters on prem Uh and we started to invest Our first cluster was about 650,000 We got a call from the team at Dell one day and they said "We've had an order canceled Do you want some?" We said "Yes please The supply chain's rubbish." Uh yes please We'll take it now And so we built this AI cluster on prem that started to us allow us to do all these really good innovative projects The challenge we then came against was then when Cisco stepped in was you know if you've got this amazing GPU cluster and Jensen's been kind enough to sell some to you and you've managed to find some stock which isn't a problem anymore But two years ago if you know you managed to get the GPUs you'd found enough space them you'd found enough power you found enough way to cool them well then the last bit of the problem was how do I get the data to those really expensive really thirsty really temperature-hungry GPUs well that was a network And so our old network back in the day and bearing in mind we hadn't invested in it for over 10 years was like drip feeding data into this tool right it was super slow And it wasn't just dripping data It was like giving it bad quality It was like latency There was jitter There was problems in the network And so the next big challenge for us was invest in a network So we went 100 gig to the access layer So that mean everyone's desktops they have 100 gig In the data center they had 100 gig Across the core of the building they had 100 gig And we started running the network very similar to how Cisco would do it for Goldman Sachs right this is banking level uh infrastructure where they cannot tolerate latency Our data scientists could not tolerate latency And so when you're looking at those 100,000 images or you're looking at 250 million data points you cannot tolerate latency when you're spending money on training those models And so the network for us became the bottleneck And if you're looking at starting this stuff in your business make sure you get that bit right because your data scientists otherwise train rubbish and it's like training with pixelated images and it's just it's rubbish Even if it's not images you just end up with garbage So put the money into it and make sure you get those foundations right The final project was then around and I'll shut up and take questions was around uh creativity Now everyone so far has spoken about creativity and like Adobe Firefly and I can create a new company logo and make it look like grass or I can make it look like bubbles or I can do really cool stuff in AI and I can deep fake our CEO which you should definitely do because it's really good fun Um and our drivers loved it because I could then do AI on their voice I mean they didn't have to do as much like media work So right do the do like the cool creative stuff But if your company does design work and you're like you know you've got CAD data and you've got all this engineering data what if that got creative so we dubbed a project silicon engineering and rather than with that 140 million trying to hire more people which would just drive salaries down we invested into AI and tried to see if we could rather than get creative in the marketing sense could we get creative in the design and engineering sense we had some early wins and early successes where you could say "Hey could you make the front wing a little bit lighter could you make the rear wing a little bit more flexible?" That's still within the regulations Um so you know you could then start to look at it Now that's an interesting conundrum right would I trust Boeing if they're out is anyone from Boeing good Um right Would I trust Boeing if they start to use AI to design parts of their plane maybe it might help them maybe it might not You know if I was designing trains going forwards what if I used AI to try and make the carriages lighter more efficient less uh less expensive to make so I think it gets really interesting when you look at AI and creativity in that sense The final part to all of that was like we were doing some really cutting edge and really interesting stuff but my sport used to be all about intellectual property The biggest challenge I had was stopping people putting all of that really clever design data into AI tools publicly So Randy the guy in the pit wall here would write a race report really detailed cover loads of insight give it to marketing they bump it into chat GPT Hey could you create a blog post for this great Well now we've just given away all of our strategy insights Now there's only other nine people in the country that are interested in it but that does get a bit sketchy So that's where we turn again to Cisco So um and that's one of the reasons and you can ask that the first question is why did I leave the sport um but that was one of the things we turned to those folk is to make sure that it's almost like DLP on what's going into these AI tools So suddenly you can use whatever you want Don't block it For goodness sake do not block AI in your business because everyone will just go and find you know the the most awful tool out there the deep sea the really vulnerable ones but embrace it right but put DLP over the top of it and then you can say you can pull whatever you want into any AI model that exists out there but just make sure it's not the strategy data and it's not the financial data and it's not this data That stuff becomes really important If an engineer walked out the building with a car design that's about two and a half terabytes of data I will spot you doing that I will spot you plugging in data I will spot you trying to upload it to Dropbox I will spot you you know pulling that across onto your laptop You you walk out with a model I've got no idea It's like two 300 meg worth of data the actual model that you're running Well that's also where you can then put like the AI security tools that we've been developing to make sure you know where those AI models are the efficacy of them how they got to be and everything else you're doing with them Um I'm aware I'm over time but I'm happy to take any questions I'm sticking around I've not got flights this evening I like going to the pub Um if you've got any questions then please ask And thank you very much I have a Okay [Applause] [Music]