Why I Beat Strava’s AI Predicted Marathon Time By 40 Min

(And What That Means for You)

Let me start by saying, I love data. I track my workouts, review heart rate zones, and keep a close eye on progress. I’ve coached hundreds of athletes to personal bests using structured training and metrics. But when it comes to race day?

Data doesn’t always tell the whole story.

This became clear at my Eugene Marathon this last weekend. When I checked Strava’s new AI prediction feature the night before race day what I saw was discouraging. It told me it predicted I could run a 3:44:43 marathon. Now, that might seem respectable, but my goal was to qualify for Boston with a time of 3:10:00. I ended up running a 3:05:56.

So how did I beat the algorithm by nearly 40 minutes?

Let’s dig into why I believe race predictions from even smart tools like Strava’s Performance Prediction can miss the mark and what that means for you as an athlete.

1. Strava Didn’t See the Full Picture of My Training

The biggest reason? I’m training for an Ironman, not just a marathon. My weekly run mileage isn’t super high compared to a pure marathon plan, but my total training volume (long rides, swim sets, brick workouts, strength sessions) has been much greater .

AI saw a lower run volume and assumed less run fitness. What it didn’t recognize was that my aerobic engine has been in overdrive thanks to multi-sport training, something endurance athletes know well, but most algorithms haven’t caught up to yet.

Research backs this up: cross-training (especially cycling and swimming) supports cardiovascular fitness and improves endurance without the extra wear and tear from running only programs. It's one reason triathletes can hold strong run performances even on lower run mileage.

2. I Train at Altitude Then Raced at close to Sea Level

I live and train in the mountains. That means cold air, snow, and thinner oxygen. It also means slower training paces, especially on the hilly terrain of my daily runs.

Strava’s race prediction assumes you’re training and racing on similar terrain. But the Eugene Marathon is a flat, close to sea-level course, and when you go from training at 3,500–6,000 feet down to sea level, it’s like flipping on an extra oxygen tank.

Altitude training has been shown to increase red blood cell count and improve oxygen delivery, huge advantages when you race at lower elevations. That edge doesn’t show up in your training paces, but it absolutely shows up on race day and I felt it.

3. My Speed Workouts Weren’t Reflected

Strava from what I see pulls from dozens of data points, but it doesn't always pick up the intent behind a workout. I had run multiple sessions of Yasso 800s, a classic marathon predictor workout that gave me confidence I could hit my Boston qualifying time.

Yasso 800s aren’t perfect, but many athletes (including me) find them useful. When I can string together 8-10×800 meters at a pace faster than a 3:05, it’s a pretty strong signal that I’m ready to race around 3:05. The AI didn’t weigh that as heavily as I did.

Lesson here? Context matters. Your workout purpose, structure, and progression count, but only you or a coach can fully interpret that. Have you ever completed a peak training workout, looked at your garmin and had it tell you it was an “unproductive workout”? 

4. It Didn’t Factor In Strength Training

Every week, I lift. Heavy. Smart. Focused. This keeps me injury-resistant, improves my running economy, and helps me hold form late in the race. It also helps me physically and mentally to finish strong. 

But Strava doesn’t (yet) factor in your strength sessions. Even though studies show that strength training improves running efficiency and reduces injury risk, these benefits stay invisible to algorithms focused solely on run data.

5. It’s Still New, It Is Not Human and I’m Not A Robot. 

Strava’s Performance Prediction tool is new, launched in early 2025, and like all AI, it’s learning. It’s grounded in millions of runs and uses machine learning to predict likely outcomes.

It doesn’t understand:

• Your mental toughness
  

• Your taper strategy
  

• Your terrain or weather
  

• How much you left in the tank during training
  

• Your years of race experience

And it definitely doesn’t understand your “why.”

I’ve been running and racing for 25+ years. I train smart, but I race harder—on emotion, on grit, and with a level of experience that can’t be predicted by pace averages alone.

So What’s the Takeaway?

Use data. Reference predictions. Check your trends.  But don’t blindly trust the algorithm. In this case I quickly questioned my capability then I shook my head and told myself this prediction it was giving me was flat out wrong! 

Use your brain. Use your coach. And most importantly, know your body.

Your watch might tell you your last workout was “unproductive,” even though it felt great and built confidence. The algorithm might say you’re not race-ready, even though your workouts, fueling, and recovery all tell a different story.

That’s why coaching and self-awareness still matter and always will.

For My Athletes and Endurance Community:

If you’re building toward a race, don’t let a prediction hold you back. Your training is more than the sum of your Strava stats. Your performance is built from:

• Smart structured programming
  

• Structured variety
  

• Real recovery
  

• A belief in your capabilities
  

• Your ability to go out race with grit and stay mentally strong. 

I qualified for Boston not because of what the data said…but because I knew I could.

And that’s something no AI can calculate.

Need help building a training plan that goes beyond the numbers? A focus on strength, endurance and nutrition as well as health and mental toughness?  I got you.

Reach out and let’s build something powerful together.

Make today your best, 

 Coach Joe Beckerley
Fuel The Finish Coaching
"Train for life. Fuel for performance. Live fully."