Analyse your running using AI

Strava Athlete Intelligence

About a year ago Strava announced Athlete Intelligence. Every tool out there had an AI feature, so Strava needed one too.

The launch promised 'personalised feedback' and 'performance insight'. 

https://stories.strava.com/articles/meet-athlete-intelligence-personalized-ai-insights-that-help-you-reach-your


I didn't love it and turned it off. It stated the obvious, it was overly complimentary and didn't recognise the difference between recovery easy runs and runs where you'd obviously put the effort in. And this was even though I tagged training sessions as 'workouts' and races as 'races'.

I noticed negative feedback on forums - and an unscientific search of forums as shown below seems to lean towards a 'not very good'/'how do you switch it off' type of vibe.

Do it yourself with Google

Meanwhile, the out of the box AI tools were getting better and better. Google Gemini side bar 'Ask Gemini' within the Google Chrome web browser is pretty amazing for example analysing data on any web page.

So, why not 'Do it yourself' I thought. And it works pretty well.


Things I like about the integrated browser AI include:
  1. You can ask any queries you like
  2. You can adjust the tone of the feedback - do you want praise or hard facts or tips for improvement
  3. Once you perfect a prompt, you can save it as a 'Skill' (see screenshot below)
  4. ...and it'll get better and better - Google are not going to stop making improvements to their AI

The screenshot shows how I have saved the AI Instructions as a skill. And within the Google sidebar I can invoke this at any time by typing /run-analysis-concise.

Strava MCP Connector with Claude

So, if Strava's own efforts were only 'OK' and my Ask Gemini efforts were really good BUT only focused on analysing one run with little context. Then how do I make the most of this huge wealth of running data I have built up inside of Strava?

The answer has been answered (I think) by Strava by launching their 'MCP Connector':

So, we're going pretty geeky here - but stick with me.

MCP = Model Context Protocol

In simple terms an MCP exposes data, in a secure read only form, to external AI engines. So you can get the best of both worlds. And in this case it is:
  • The best data - your unique data from all of your runs on Strava
  • The best AI - the AI platform you choose to use
Well for the first release Strava have chosen Claude.AI as the AI platform. And here's a few screenshots on how it works.

Firstly, sign up for a free Claude account - https://claude.ai/ - you'll get free credits for queries every day

Then click to enable the MCP in your Strava settings:

Then just chat away to Claude about your running - ask anything you want. Some examples below...

Example 1 - analyse a race - just say the race name... (similar example to Ask Gemini)




Example 2 - Make a query, in plain language, across many runs

Example 3 - Generating training plans, that are considerate of what your weekly mileage is


To summarise - I really think the MCP approach is the way to go. It's your data. Use it with the best AI technology.

...or if you don't like tech - feel free to just enjoy running and keep a notebook of your runs like the old days - as long as you're having fun...





Comments

Post a Comment

Popular posts from this blog

Laps / splits on Garmin / Strava

Image
When using a running watch (I use a Garmin) the Lap feature is pretty useful. It allows you to break your run into set distances (a Lap). Then when analysing on the Garmin or Strava website/app you can see your Splits, which are an analysis of your performance for each Lap. For example, ' For the 8th km of the race my average pace was 3:40/km, my heartbeat was 178bpm, and the elevation difference was +6m giving me a grade adjusted pace (gap) of 3:38/km '.  1. Auto lap Out of the box, most running watches automatically record a lap at 1.00 km (or 1.00 miles if you want to be old school ). The image below shows the default setting on my watch. Auto laps - the default option The screenshot below shows the automatic analysis of a recent race I did (Jarrow 10k) - I knew I needed 3:33/km average for a PB. I very nearly got it - but the laps/splits into the breeze where the pace was slipping to 3:40/km meant I just missed out. During the race by watch notified me every km so I knew wh...
Read more

Welcome to my running blog

Image
Main areas to explore: -  North East half-marathon routes - Over 20 routes from Holy Island in the north down to Hamsterley Forest in the south. -  Training workouts - A few different running workouts to keep training fun.  -  How to run faster - How to go from a 22 minute parkrun to breaking the 18 minute mark in five years. Hope you find it of interest.
Read more

2024 - Looking back at the year

Image
2024 was a year that always going to be a big one. I'd steadily improved for a number of years and knew with a bit of extra training and commitment I could enter the bottom of the M50 category and be competitive. I posted in January that I was going to give everything and see where things would end up.   Looking back on the year - I PB'd ever major distance apart from 10k - which is something I'm very proud of. In terms of rankings across the M50 category - top 10 in the North East region and there or thereabouts top 100 in the UK. Best achievements and rankings Best races parkrun was where my love for running started. So getting a sub-17 (16:57) at the 5k distance was something I was over the moon about. Averaging 3:29/km on the way out but conserving energy tucked into a group running into the wind I knew I had something in the bag for the way back. 3:21 and then 3:16 for the 4th and 5th km was something I didn't think I had in me - but the adrenalin and slight tail ...
Read more