Best Bike Split Founder Ryan Cooper

SHOW TRANSCRIPT: Marginal Gains Podcast - Best Bike Split with Ryan Cooper

Hottie: The Marginal Gains Podcast is brought to you by SILCA, creators of the finest pumps, 3D-printed parts, and drivetrain efficiency products for bicycles.

I did two things, two maintenance items, that is, to our bikes before I left for Leadville: freshened up the sealant and waxed the chains using Ultimate Sealant and Secret Chain Blend. More than eight hours of riding on Leadville's rough trails and dusty to normal conditions: no flats and no squeaky chains. In fact, my tire pressure never dropped during the whole trip, and our chains never needed topping off. All we had to do was ride. Imagine that.

Whether you're going to Leadville or not, you want to keep air in your tires and lube on your chain. To see how SILCA can help with that, head to silca.cc. I'm Michael Hotten, better known as Hottie.

We are also sponsored by Shimano. When we talk helmets on the show, it is mostly in the context of aero performance. But the primary job of that shell is to protect your head, and we do recommend you wear head protection. I just recently bought a new helmet from Lazer. Lazer is part of the Shimano family. Lazer makes great head protection, some of which has proved itself in a wind tunnel, and other models do great on a hot day when cooling is at a premium.

The great thing about Lazer is you don't have to go to the top of their line to find one of their finest technologies. For $80, I was able to get the Tonic, a road helmet that comes with Lazer's class-leading KinetiCore anti-rotational impact system. I love that helmet. Even though it is neither particularly aero nor filled with dozens of vents, it does a great job without all that stuff and still has the most important element in a helmet, and that is protection. The entire Lazer lineup can be found at lasersport.us. And thanks, Shimano, for supporting Marginal Gains.

Hottie (01:57): Best Bike Split has come up on a number of occasions on this show. It is a terrific online marginal gains tool that can help riders project times on anything from a short TT course to a 200-mile trip over the Flint Hills of Kansas. Behind BBS is a lot of math and a man named Ryan Cooper.

Ryan first started working on what would become Best Bike Split in 2013. He was working on his PhD in operations research and optimization when a Tour de France time trial got him thinking: could he predict times of specific riders? Turns out he and his math could. Now that story is somewhat known, but Josh and I wanted to know more. So we got Ryan Cooper on mic to explain what is under the hood and what he says are the new frontiers of what is truly a piece of marginal gains history.

Hottie (02:59): All right, like we said, Ryan Cooper, co-founder and chief scientist at Best Bike Split  he is our guest. We've been fascinated with Best Bike Split for a long time on this show, Josh. You've referenced it a number of times when we've talked about optimization and helping people go faster. So, Ryan Cooper, we want to thank you for coming on and sharing your story and explaining Best Bike Split to us.

Ryan: No, thank you guys for having me on. This is going to be awesome.

Hottie: Do people just say BBS? I just stumbled over "Best Bike Split" because it's a little rough to get through. BBS  does that work too?

Ryan: It is. I think most people refer to it as BBS. It's one of those things where when you're trying to name stuff, it's so hard to name things and even trying to find domain names. And so it was, well, what are we trying to achieve? We're trying to get your best splits possible. So yeah, that's the etymology of the name, not overly complicated.

Hottie: It's probably worked out well for you in SEO, I'd imagine. Sometimes it's those straightforward names where you inadvertently were more clever than you thought.

Ryan (04:02): No, that's true. Yeah, and it is funny because a lot of people will search  if you do a Google search or whatever  for "the best bike split" of this or different races, especially around triathlon times and things like that.

Hottie: We always love to ask guests about their bike origin story. How did it start for you?

Ryan: I was actually kind of a latecomer to cycling. I was in college and a friend of mine was racing for the Texas A&M club team. He was doing a lot of mountain biking. I was home for my first summer  I went to the University of Texas  and my best friend and I went to opposite schools. We were both home for the summer, and he's like, let's get some training and go ride.

The only bike I had was my stepdad's old steel-frame, downtube-shifters bike. I just remember getting on that, making homemade Gatorade, salt, and orange juice, and water, and then going out for what was supposed to be a 20-mile ride. We ended up getting lost, and I think it ended up being about a three-and-a-half-hour ride to get back.

After that, I was in love with it. I thought this is the most awesome thing. I quickly adjusted bike selection and nutrition. But yeah, I was hooked after that first time on it.

Josh (05:32): Love it. We're big fans of getting lost. I always harp on my guys we all have computers and use them and love them and I'm like, you need to not ride with it sometimes and just get lost. That's the fun, right? Which is kind of ironically the complete opposite of what you do now not the fun part, but you're all about the data. How did you connect those dots from getting lost to every inch of the course and every bit of the data?

Ryan: I was an electrical engineering undergrad. I worked in the aerospace industry for a long time. Always very engineering in the weeds. I went back to do a master's program in operations research  applied mathematics large-scale optimization modeling.

Typically this is used for inventory management or scheduling. I had already been bit by the endurance sports bug. I was racing bikes. I was racing triathlon, racing Ironman, trying to work full-time and do a master's and a PhD. It just wasn't feasible to juggle everything.

My first foray into it was a little startup a buddy and I created called Optimize Training. It was adaptive training think RUNNA or some of these apps that are out now back in 2010. We thought it was going to take off. It didn't. People didn't trust data. They didn't trust "the AI bot," or whatever you want to call it. At the time, it was more like constraint programming. Coaches hated it.

I took it through the USA Triathlon coaching certification process. They have a test at the end, and I let the algorithm do all the work didn't tell them that. It got a B-plus. So it wasn't horrible. It planned out athletes' progression and prioritization from starting date to goal date. But it never took off. We didn't have marketing or market appeal, and we were competing against coaches.

In 2013, I watched a symposium called "Math in Cycling" at the University of Utah. Speakers talked about data analytics in cycling. Somebody said, "We can do time-trial pacing optimization, but you have to ride the course, then do data analysis, then run it through a model, and 24 to 36 hours later, you get how you should have ridden the course."

I thought there's got to be a better way without pre-riding, with automation, and it should run fast. All my study work then was about making models run fast.

The aha moment came watching the 2013 Tour de France time trial. Tony Martin was at the top of his game. I built a model based on the course, broke it down, and checked if I could get close to predicted times. I was within a couple of seconds of Tony Martin's time and the GC riders' times. I said, maybe this has legs if we can make it usable for riders to plan strategy.

Afterwards, I saw Tony Martin's cheat sheet for that race. It had every part of the course broken into 200-meter segments and exactly what he would do at each one all written by hand. That's what I wanted for anybody.

Josh: That's cool. I was working with Tony Martin at the time, my last year at Zipp. We had something similar. Pre-ride GPS into a 3D model, overlay wind mapping. At first, we used constant power, then realized we needed a different strategy. It was the most glorified spreadsheet ever. Before then, guys were just like, "go until you get tunnel vision." 2012 was the beginning of "do this here, do that there."

And nobody was thinking about things like, do you go harder into the headwind or with the tailwind? Do I pedal downhill or tuck?

Ryan: Exactly. A little full circle there.

Josh: Can you share what's happening under the hood?

Ryan: It's similar to what you described just much faster and at scale. We take a GPX or ride file, segment by direction and elevation change, overlay wind data and (soon) road-surface data, then build segments.

The optimization looks at power, time, and min VI (variability index). Some athletes can absorb punchiness think Cancellara, Wout, or Pogacar where they can go to the well over and over. Others need steady state. Those athlete-specific traits are built in. We also look at cornering ability and max descent speeds. For off-road, we model "skill" via momentum retention through turns.

Hottie: So, can Best Bike Split look at a rider's history and infer skill?

Ryan: Yes. You can pull in previous data, and we analyze it. Some of that isn't public yet but is coming soon. There's always a balance between usability and building purely for the pointy end. Ideally, it's for everyone, but it's catered to front-pack racers. Many athletes and coaches also use it for cutoff management in races like Leadville.

Hottie: What separates the pointy end use case from the century rider?

Ryan: It's great for both. Closer to the pointy end, you care about smaller tweaks. I talked to Grace Brown she used BBS for her TT world title. She adjusted advanced settings to her style: able to surge then settle into diesel power. She was crystal clear on the course: every turn, surface, and what to do if it's wet. There were many crashes that day  Chloe crashed and still got bronze but Grace looked like it was a day in the park. I think she beat a third of the men's times. Visualization and prep matter.

At that level, BBS shines. You also have better, cleaner data. With everyday riders and older left-only power meters, accuracy is the limiter. For modeling, you want accuracy.

Josh (19:27): It's the perfect tool for the person who will actually use it. If you're engaged, it's great. It's not great for the rider who doesn't know their gear and has an ancient, uncalibrated power meter.

Ryan: We've had a ton of bikepackers using it for long journeys and RAAM participants using it for overall planning and 200-mile chunks.

Hottie: What variables wobble the most and can throw things off?

Ryan (21:07): Weather and drag. Group rides add drafting complexity. We typically model from the peloton lead, with an estimated peloton CdA, to get expected finishing times.

Then there's our Time Analysis Tool: it shows where increasing power or decreasing drag yields the most time gains. It helps identify sections to attack or defend. For example, a long crosswind section that turns into a tailwind: a surge before the turn can create separation that's hard to close once you hit the tailwind. We've seen this in gravel, including Unbound.

Elevation data is another error source. We can correct elevation, and are rolling out bridge and tunnel detection. Barometric drift at altitude can cause elevation to grow or shrink over time. So wind and elevation are the most error-prone.

Hottie (25:15): Can you fix things on the fly, if conditions change?

Ryan: To a degree. We're looking at a Hammerhead app. Building robust apps directly on head units is limiting. If everyone mounted a phone, you could do real-time pulling and adjustments. Within Garmin/Wahoo confines, it's tougher.

Josh: What can you do on current head units?

Ryan: We put the pacing plan on the unit. Upload the course to Garmin Connect and use our IQ data field. It shows power targets based on lat/long. A new data field is coming with an easy color gauge to show if you're over- or under-cooking it.

It's a guide; feel on the day matters. With pro teams, we'd set a day-of baseline and simple adjustments by situation: gradient category, wind type, plus/minus watts from baseline individualized to the rider. If your heart rate spikes, adjust.

Hottie: You used the heart rate word! We still like a governor on big climbs.

Ryan: Jan Frodeno stuck to a power plan, but heart rate was the limiter the do-not-exceed.

Josh (29:27): New features: sensors and AI?

Ryan: We released an AI workout builder about a year and a half ago. Not full training plans, but key workouts specific to your race plan. We auto-push to Wahoo, Garmin, Zwift. Select a race, give prompts like time available or platform, and it builds structured workouts.

We also have an AI race assistant to answer gearing, nutrition, and plan questions and to help us prioritize features based on what people ask.

Josh: Aero sensors real-time CdA?

Ryan: Lots of interest. The big issue is lack of standardization for wind data and CdA fields. Most wind sensors are pitot-style, measuring wind speed and combining with power, rolling resistance estimates, weight, etc., to compute CdA on the head unit or after. We also estimate CdA from ground speed, power, and expected wind. When data is clean, it's good. Pulling wind speed directly would help. Wahoo ACE has started building protocol support.

Josh: Standards in cycling, ha!

Ryan: I work with USA Cycling on data modeling for the track team. We've used sensors effectively for team pursuit and are looking at team sprint.

Hottie: What about a velodrome?

Ryan (35:09): We don't offer much commercially there small market. I wish track were bigger in the U.S. Through USA Cycling, we just won the Edelman Award for modeling team pursuit an applied analytics award up against Amazon, Lufthansa, Waste Management, etc. First time a sports project won.

Josh (37:13): Track strategy gets infinitely complicated: pull lengths, whether to finish with three, riding the black line.

Ryan: Exactly. Every board above the black line adds distance per lap; it adds up fast.

Hottie: How about XC MTB or cyclocross?

Ryan: For MTB and CX, most use is course recon. Accel/decel and tactics dominate. For longer MTB where pacing matters, you can rate course difficulty and meter efforts accordingly.

Josh: I envision F1-style overlays someday: momentum retention scores, etc.

Ryan (40:26): The foundation is good data. LLMs are great at explanation but not inherently great at parsing cycling data relationships. Our approach: give AI access to physics-based tools we build; let the LLM orchestrate and explain results.

Hottie (42:15): Equipment: I love changing equipment or tire pressure and seeing new targets. Does BBS recognize brands?

Ryan: Not currently. Early on we partnered with Trek Factory Racing. Their internal model took hours to run; ours took under a second with the same parameters, enabling far more iterations. Trek has been a partner ever since (Lidl-Trek now).

Access to tunnel and rolling-resistance data helped us refine what matters: rider is biggest, but frames and cockpits matter some. We re-baseline with new tunnel data and sources like bicyclerollingresistance.com. A new gravel-surface protocol shows surface impact on Crr. Somewhat surprisingly, the deltas across surfaces are fairly tire-agnostic. Input your tire's measured rolling resistance and BBS will be spot on.

Josh: Was that the Irish group using mean square vertical profile?

Ryan (46:00): I know that work. This particular data was from a researcher in Georgia who is publishing soon.

Josh: We've joked there is no stiffness parameter. If stiffness mattered much, it would be in the model.

Ryan: Exactly. The big three in the model: CdA (nonlinear), Crr and weight (linear), and drivetrain efficiency (top-line haircut on power). Other stuff is often implicit via these.

Hottie (48:05): Is BBS for bikes only?

Ryan: Motorsports do a ton of optimization. I'm very interested in road-surface data. We want specifics: quality paved vs. concrete vs. asphalt, chip seal, cobbles. Chip seal is like mini cobbles; you lose power transfer because you're bouncing.

We're building a service using AI to take course data plus imagery (satellite and street-view sources) to estimate surface conditions segment by segment, adjust rolling resistance, and highlight surfaces. Example: "cobbles for 300 m ahead."

Josh: Mark road furniture and take it to the UCI, focus on safety.

Ryan (51:10): Exactly. Highlight sections that get slick in rain. That's what was needed for the Paris time trial.

Hottie: Could that info reach the athlete in real time?

Ryan: Yes. That's in the next iteration, alerts like points of interest: "gravel ahead," "potholes," "chip seal incoming."

Josh: UCI, buy a group license.

Ryan: The radio-in-the-jersey rule was interesting. People were putting bottles down the jersey for aero gains. We tested it. it works. At Worlds 2023, I think someone flipped the race radio vertically and taped it for protection and aero benefit.

Josh: Little radio baby.

Ryan: It was noticeably faster before action was taken.

Josh: If it were up to me, UCI would have almost no rules except aesthetic ones. Sock height is fine. Let everything else go and make the sport look cool.

Josh (56:56): New features time frame?

Ryan: Initial rollout of surface-data viewing is within weeks. It will be an advanced feature (on/off) showing surface types across a course in 78 categories. Then imagery-based specific course-surface data. We're working to keep compute costs down.

Josh: That's right in our listeners' wheelhouse.

Hottie: Ryan Cooper, Best Bike Split, awesome conversation. We'll get lots of comments and questions. Hopefully, we have you on again once people use these features. Thanks for joining us.

Ryan (58:33): Thank you. It's been awesome.

Hottie: Thanks again to Ryan Cooper, co-founder of Best Bike Split. We're looking forward to his tarmac analysis tool, and hopefully the UCI is open to it as well.

bestbikesplit.com is the website. Check out their YouTube channel for explanations on using BBS. You can comment on this or any episode at marginalgainspodcast.cc, where you'll see links to all the platforms for listening. The website is also where we go hunting for questions for Josh in our Ask Josh Anything shows.

We'll be back soon with more questions and answers from Josh. For now, thanks for listening to the Marginal Gains Podcast.