Strava’s AI – in the real world

Take a look at the image above. It’s indicative of how Strava is not getting AI right. Remember the principle and read on…

Strava AI – A Very Quick Recap

Strava subscribers recently received free access to the new AI workout feedback.

From your regular workout summary, Strava’s AI gives you a quick overview plus details if you click through.

The whole point of this use of AI is to present you with information you might not have spotted by looking at the same old charts, or it should guide you towards further exploring the details and spending more time in the app. The more time you spend in the app, the more value you get and the more likely you are to remain a subscriber and boost Strava’s coffers.

Strava’s Quick Message – Examples

I received many AI messages on my Strava app, from the most trivial workout to a near-PR/PB effort. Here are some examples.

Whilst each message is perhaps unique and different from the next, they all exhibit a particular pattern, repeating broad phrase types and presented in the same format. Only a few would encourage me to click through to learn more. The blandness would probably stop me from reading any of those AI summaries after a week.

I said one workout was a PR/PB. Which one was it? Maybe that could even have been properly highlighted. Surely, I might learn the most from such an important workout.

From reading professional mainstream media outlets, you will know that the headline has to grab you to bring you in to spend more time reading its content. For a website, you might use the phrase click-bait, but almost everyone doesn’t mind ‘snappy’ headlines so long as they represent the content they draw you towards – that said, trust me, it is HARD to write headlines like that with tight word limits. Strava’s AI finds it hard, too.

Strava’s Detailed Message – Examples

There’s more detail if you want it. Here are examples of what I clicked through to.

#NotVeryExciting

Tell Me Something I Don’t Already Know

A few weeks ago, I read a tweet where someone criticised Strava AI: “Tell me something I don’t already know.” The context of the phrase stuck in my mind.

I’m unsure if that was fair criticism, though. Strava DOES tell me something I didn’t already know. Admittedly, I could quite quickly have learnt as much by looking at some charts in the app and then simply ‘knowing’ how that particular performance probably compared to recent performances. Mostly, Strava AI seems to save me a bit of time by not bothering to look through my stats.

Perhaps a better but less catchy criticism is implied by: “Tell me something I WANT/NEED to know.” This brings us back to the image at the top.

Strava is almost giving a boring speech like the person on the left of the image rather than the life-saving screech from the person on the right.

Data > Information> Knowledge> Insight> (Wisdom)

From an IT presentation I received while working in the real world (not the blogosphere), Strava AI must transform our DATA into INSIGHT. That imperative is unchanged from decades ago. Everything Strava gave me could have been done without AI thirty years ago. Admittedly, replicating it would have needed many template sentences and rules. But it could have been done straightforwardly.

Strava AI should also be used to make deep discoveries and present them in a more unique and engaging way.

Does Strava have the data to do that?

Strava has a lot of data. A lot. But does it have the correct data to deliver insights?

I looked at Whoop Coach, and it has access to a library of sports science (rules), your performance data (workout results) and data you have tagged with whatever you deem to be interesting, like alcohol consumption, sex, and food eaten late at night (personal engagement factors). It determines correlations between tags and performance and presents them in the light of science. Whatever you think about how well Whoop‘s AI Coach does that, I hope you agree it uses the correct data and broadly approaches the problem correctly. The main difference is that you have to ask Whoop Coach a question that immediately gives the coach the context of your interest.

AFAIK, Strava AI doesn’t have access to sports science. It doesn’t use any subjective tagging/highlighting or the notes for each workout, and it doesn’t have much access to how your body responds to changes in physiology, i.e. all the wellness info, although it does have some.

What Does Strava Have Access to?

Strava’s AI should potentially use richer data like weather data or the kinds of riders you rode with.

Q: Do you remember the whole point of Strava?

A: Segments, right?

Why hasn’t its AI even mentioned those to me?

Q: Do you remember why Strava has done so well as a business?

A: Sociability, right?

Why hasn’t its AI even mentioned my performance compared to my club mates or the people I follow-back?

Each plugin’s words added to the workout summary automatically produce rich tags which an AI could use. I am interested in them because I installed those specific plugins. You will have different tags installed based on your plugins and interests. #Personalisation. Here is an example of the notes of one of my recent Strava rides

Strava knows I use the WANDRER plugin (looks at what percentage of roads you’ve ridden in a given area). Strava knows I use KLIMAT to provide air density to support my interest in performance metrics. Why doesn’t it use the summaries I get from each plugin? For example, a near PB effort, adjusted for air density, could be known as a real segment PB/PR.

Take Out

Strava AI does add something. I’m just not sure it’s of any value to me other than a passing interest.

Strava’s AI tries to be a clever reporting and analysis tool primarily based on raw performance data like HR, speed, and power, contextualised over time. I’m not sure I want Strava for that purpose at all.  I use it for segments (a mess), routes (excellent) and social aspects (good).

Changes

I don’t expect Strava to rein in their efforts and start from scratch. The most productive thing they could do is stop the AI from appearing on irrelevant (short) workouts. Just show it for the more important or exceptional ones where it might have something interesting to say.

In any case, the insights I envisage would be highly resource-intensive and probably grind the entire Strava ecosystem to an expensive halt.

That said, I might be interested in periodically hearing what Strava’s AI says. So, perhaps even a new area in the app where I can look at the REAL insights from the last 7, 14 or however many days.

Or perhaps there could be an AI icon next to the Achievements icon, and I often click on the Achievements icon to marvel at my PRs on subtly different segments I PRd last week. I might click on an AI icon to see what insight Strava has for me.

Thirdly, the scope of AI needs to be radically extended to cover what each user is uniquely interested in, whether engagement, performance, or whatever. A simple Ra-Ra message after each workout is not good enough.