One Runner Followed His Watch’s AI Training Plans for a Month. Here’s What Happened.

Published July 19, 2023 11:47AM

A few months ago, I noticed an odd thing about my Garmin Enduro 2 watch. At the start of my Saturday run, instead of immediately searching for satellites, it displayed a new screen, an AI-generated suggested workout customized just for me. I was planning to do a long run, but my watch said I should do intervals instead.

Initially, I scoffed. How could this wrist device know what my body needs, better than me? The longer I sat with it, the more curious I got.

In the last few years, running watches have made substantial leaps forward. They are now able to collect and analyze a wealth of information, stuff like heart rate variability, blood oxygen levels, respiration rates, sleep cycles, stress, acclimatization, stride mechanics, and other metrics. Even midpack models offer a robust feature set that would’ve been hard to imagine a decade ago. Most major watch brands like Suunto, Coros, and Polar now offer a suite of recovery recommendations, load tracking, and basic workout suggestions.

Could my Garmin watch actually know something I don’t about my training progress? Could it, heaven forbid, replace a coach? I have had a handful of coaches over my running career, usually whenever I had a clear race or goal in mind, so needless to say, I was skeptical. Currently I have no coach, and no specific race I’m preparing for, so I decided to be a guinea pig, making a resolution to wear my Garmin for a full month, 24 hours a day, and do everything it recommended.

Who Is This AI Coaching Watch Function Made For?

In 2020, Garmin bought Firstbeat Analytics, a Finnish company, and started to release suggested workouts a few months later, crunching numbers on your recovery, sleep, and body strain to recommend what you should do on any given day. The goal? To use machine learning to help improve users’ VO2 max and lactate thresholds, or in other words, your short- and long-term fitness capacity.

It’s easy to be skeptical—even this publication is dubious of Garmin’s recovery algorithm—but the more I asked around, the more I realized how few people had actually given these suggested workouts, or even the Garmin Coach feature, a fair shot.

So who, exactly, are these suggested workouts for? Joe Heikes, Forerunner Product Manager, says that Garmin developed these recommendations for “the middle of running’s society.” The top quarter of the sport—highly competitive, elite, pro runners—likely have a coach already, while most new runners aren’t ready for a structured training plan as they ease into the sport.

“This feature is for committed runners who need a little help,” says Heikes. “They don’t need a ton of hand-holding, but do want a plan and to see progress.”

Herman Bonner, who works for Firstbeat Analytics, says the biggest challenge is trust. “Anytime you’re giving advice, you first have to prove you are trustworthy. This takes a lot of time and effort.” But Booner is confident in the algorithm. “As an analytics company, we sifted through all kinds of data, applied accepted training philosophies, and tested for years, but our customers don’t see that,” he says. “They have to use the feature to see the benefits, but it’s hard to commit before you trust it. So there is a loop.”

Heikes says they first identified the need almost ten years ago. Garmin was getting feedback from users who wanted guidance on workouts, something to push them in the right direction. “I was skeptical when I started testing it; so was Herman,” Heikes says. “The watch will never be perfect. It’s built on a specific coaching philosophy that’s not right for every runner, but it’s far better than nothing.”

Automated Training Plans: First-Hand Experience

Because suggested workouts are informed by the aggregate of all the data your watch is able to collect, I committed to wearing it nonstop. This isn’t normal for me—I typically wear watches only for runs—but I knew it would be more accurate (and presumably beneficial) if I was fully committed.

From there, Garmin took over. I didn’t have to log my workouts or figure out my baseline fitness, like you would with a new coach. All of this hums along in the background and only gets more accurate the more you use the watch—the strange beauty of AI tracking everything you do.

As I learned, there are seven different types of recommended workouts: a mix of recovery runs, base endurance, high-intensity aerobic workouts, and anaerobic training efforts. I found this to be diverse enough to stay interested for a month, but I wonder if it would get repetitive in the long run. I would later learn it’s designed as a step in the process of establishing a running routine, not necessarily something you would rely on for years.

The workouts scale as you get more fit, in terms of duration and intensity. My ramp-up started slow, but as the weeks flew by, I did notice an acceleration, especially in the duration of my harder workouts. My biggest pain points were in the first week when the watch served up some questionable recommendations. For example, on my second day, Garmin suggested a tempo workout, but I was feeling terrible after sleeping poorly, so I did an easy run instead.

Garmin acknowledges its own imperfection on their website: “There will be days when the best option is for you to rest for your next challenge,” implying that you can skip workouts as needed. Blips like this are part of the process—the watch is trying to learn your long-term patterns, and errors are much more likely early on.

On the plus side, right away I noticed how well the algorithm incorporates data from other devices, like my Garmin Edge bike computer. This is critical, as someone who keeps a relatively even balance between running and riding. It also incorporates big efforts in the gym, telling me to back off my next day of running after a hard kettlebell workout. I liked this multi-sport integration, which applies to many other weekend dabblers like me.

The Final Verdict

Overall, I found the AI behind the watch to be mostly accurate, suggesting base efforts in line with my expectations, threshold workouts consistent with past workouts, and anaerobic sprint workouts to be hard, as they should be. While it took time to build trust, I feel confident that the logic is sound, although not for everyone. If you already have specific workouts you like that focus on key running metrics like VO2 max or lactate threshold, you’re probably not the intended user of the feature anyway.

While I enjoyed my month using the feature, I’ve already returned to my old habits of training by feel, an approach that offers more flexibility. But maybe this experiment says more about my personal preferences than it does about the watch and its algorithms.

Heikes says the suggested workouts feature was not developed to replace coaching, though he thinks they could certainly complement each other. “There is a lot of data for a coach to look at for high-level advice, while letting the watch operate as an AI training plan for day-to-day workouts,” he says.

Corrine Malcolm, who has been coaching since 2016 and founder of Foothills Endurance, sees an opportunity for coaches to work in tandem with AI coaching platforms.

“The biggest challenge of coaching is how to scale. If you’re doing it right, you can only support a few dozen people at a time,” says Malcolm, who believes a hybrid model could provide value to a runner who doesn’t want to pay a large monthly fee.

“Think of it like a low-cost, low-touch model. A weekly office hours to ask questions to a human, while mostly relying on the AI to give workouts.” Leveraging the algorithms ability to collect and analyze data, Malcolm says a hybrid model could help her coaching business scale while giving runners a new style of coaching to consider, and from which to benefit. “I could probably coach a few hundred people this way, which is a win-win for everybody.”

After my personal experiment, I can see the value of the data and suggested workouts to inform and simplify coaching. And for someone just a couple years into the sport without a coach, I’m confident that the suggested workouts will help improve fitness and variation, provided you stay consistent and (gasp!) trust the algorithm.