Strava may be the most popular fitness app in the world, with 135 million users and counting, 51 million logged activities per week, and 1 million communal clubs. But its new Athlete Intelligence shows Strava isn’t resting on its laurels, and it wants you to know that its AI isn’t “just an LLM,” even if you might assume that.
On Wednesday, I watched cheerful Strava employees celebrate the opening of their new San Francisco HQ after a brief speech from San Francisco Mayor Daniel Lurie. He called Strava the “tip of the spear” to convince people that “San Francisco is open for business” and “the greatest city in the world.”
Strava’s multi-floor office focuses more on celebrating the world of fitness if anything. I toured the office with CEO Mike Martin and Chief of Staff Shelby Rhodes, who showed me meeting rooms named after internationally beloved segments like Central Park or Alpe d’Huez. I also saw a live Strava Photo Map that highlights photos that their AI labels as “beautiful” from around the world, as well as employees’ bibs from their favorite races.
But my favorite exhibit focused on SF: a live, chronological map of the city that showed the live flyover progress of local Strava users’ activity that day. It’s an incredible effect, and I told the Strava execs that they should seriously consider upgrading its weekly heat maps to let users see the previous day’s activity, at least in the big cities. They don’t have any plans for it, but hopefully, I planted the seed for it to happen.
Add in the library of running books and wall art showing people’s GPS-made art, and it was nerd heaven for a runner like me. But as much as I enjoyed the tour, my main objective was to sit down with CPO Matt Salazar and CTO Rob Terrell to dive into Athlete Intelligence — and figure out where Strava is going next.
Unsurprisingly, they couldn’t say too much that’s new about their “future roadmap,” but I still got a sense of where Strava Athlete Intelligence could go next.
Athlete Intelligence has room to grow
Athlete Intelligence just left beta, and Strava claims “80+% of users say that Athlete Intelligence is very helpful” based on in-app feedback. I myself have noticed a definite improvement in the accuracy of the data, but I asked Salazar and Terrell what their vision of Athlete Intelligence is and how it benefits users.
Terrell explained that “it was important to me that Strava do something that was much more significant” than the “low-effort” LLMs out there that spit out vague answers and can only give generalized advice from Google.
Even if “from the outside, it looks like it’s just an LLM,” their goal was to “really synthesize the granular data that we have about the users’ activity and do deeper analysis with machine learning to understand the context of that data.”
Salazar, in turn, noted that “not all people are at the same experience level to interpret heart rate zones or their pace.” Their AI makes this data “much simpler and easier to understand” for beginners who might not understand the impact of anaerobic activity or who don’t want to check back at old activities to see how their pace has changed.
Strava’s AI primarily focuses on runs, hikes, walks, and cycling. I asked about their plans to expand to other activities, and Salazar explained that even as they “make sure it covers everything,” eventually, they want to ensure first that “when we put out something, it’s a sufficient quality,” and that other sports might need more internal testing before it’s “fully baked.”
I then noted that Athlete Intelligence is very responsive rather than proactive: It tells you how you did after the fact but doesn’t necessarily recommend a training plan or target pace for future runs as you get on Garmin watches. Will that change?
Salazar said that they’re not specifically looking for “the next AI thing we can do,” but rather “how do we help users achieve their goals?” Up until now, that’s been through incentives like challenge badges, monthly goal targets, and kudos from friends. But AI is one “particular tool that we’re looking at” that could come into play next.
Terrell agreed that Strava “historically has been a backward-looking product” that largely leaves planning in users’ hands and that while they’re “not announcing any products” on this front at this time, they are “interested in…being forward-looking.”
Even without any specifics, a more proactive form of Athlete Intelligence would make it more useful for beginners and serious athletes, so consider me hopeful that this evolution comes next.
Strava knows you’re mad about race distances
Image 1 of 4
The most passionate response I got from Terrell, Salazar, and even the CEO was when I jokingly mentioned the time I didn’t get credit for a half-marathon PR because my watch didn’t measure the full 13.1 miles. Apparently, a lot of users want Strava to “fix” this and aren’t at all amused about it.
I frequently test the GPS accuracy of smartwatches and know how fallible they can be. But plenty of Strava users don’t realize that — as Terrell put it — the app is “limited by the accuracy of devices that users are using” and that they want to be “as accurate as possible” rather than “fudge” the data.
Strava itself said in its Year in Review 2024 that 72% of Strava activities come from users’ phones, which aren’t necessarily as optimized as smartwatches when it comes to GPS accuracy.
Users can trim activities if they forget to pause before driving home, but letting users add distance to a run is a slippery slope. Yes, sometimes GPS inaccuracies give you the wrong data, but sometimes, it’s the course itself that’s short. Strava’s team wants to be “truth tellers…when folks are trying to compare themselves against each other, especially at a pro level,” Salazar explained.
“When we first joined, Matt and I were like, ‘Let’s just fix it. This is easy!’ And then we dug into it, and it’s like, oh yeah, there’s nothing—,” Terrell said, with Salazar jumping in with, “there’s a lot of steps, yeah.”
“We understand the user pain here, and we’re looking for ways to solve it, but there’s not an easy answer,” Terrell concluded.
Strava could help you ‘celebrate’ your race triumphs better
I also asked if Strava would ever consider using its segment leaderboards to create official race leaderboards on Strava. They themselves said that more than half of runners at major marathons like NYC, Boston, and London logged their races on Strava last year, so what if runners could see how they ranked among Strava users and compare their race times against previous years?
Salazar answered very carefully that they “can’t speak towards future roadmap, but…all I can say is that that use case is very important for us, and it’s very much present in our minds. So I’m thinking about how we can help users celebrate that moment; we know it’s a moment they want to celebrate.”
While this isn’t exactly a confirmation, it does make me optimistic that Strava might be working on race leaderboards that make it easier to connect with fellow racers you met. Labeling races and grouping times might make it easier for Athlete Intelligence to rate your effort compared to others in your age group and gender, which would be enlightening!
Where Strava goes next
Just in 2024, Strava added Athlete Intelligence, weekly and night heatmaps, a dark mode, and closer integrations with Oura and Apple Fitness Plus. So far this year, they’ve improved leaderboard integrity, made it easier to share your workout flyovers to brag on social media, and just this week added 3D FATMAP terrain data designed for winter sports athletes to see gradients, aspects, and even the nearby danger areas for avalanches.
Strava didn’t have any other future info to share yet, aside from telling me about FATMAP the day before its reveal. But I did get Strava’s execs to hint at “future stuff that we shouldn’t talk about,” in their words.
When I asked if Athlete Intelligence — which currently seems to focus on your last 30 days of activities for relative effort — would ever look at older data or take into account goals like “training for a marathon,” Terrell circled back to his initial point:
“We’re not just stuffing a prompt into an LLM; we are building custom models that will allow us to do different levels of interesting analysis. And what we can feed into the LLM will be much more specific to the users, sort of their lifelong fitness journey. So the LLM is going to be able to speak to a much broader array of aspects of your active life.”
Salazar jumped in to agree that they’re always asking “how can we develop more sophistication in what insights we can give the users,” and that Athlete Intelligence “is not done by any means.” They’ll continue to grow and smarten up the feature, making it more useful for a wider range of athletes.
It’ll be fascinating to see how Strava’s AI efforts compare against a hypothetical Gemini Coach in Fitbit or the Samsung Health Coach coming to Wear OS 6. At the moment, Strava has a head start on Android watches and doesn’t intend to slow down.
