We can offer you truly personalized training because our AI algorithm can predict your performance from your data. So we can provide you the best possible training results for your time constraints and tell you what performance you're going to achieve if you stick to the plan.

We deeply care for a scientific and data-driven approach. We know AI Endurance works because not only have we compared it to the most recent peer-reviewed research in exercise physiology, we have shown how our AI optimized training easily outperforms even the best one-size-fits all approach current research has to offer.

We can predict your future performance at about 5% if you have at least 100 runs with GPS data or rides with power data.

With AI Endurance every user gets their own 'digital twin' - a machine learning model that represents all the possible ways you would respond to different training routines. The model is updated every 24 hours for every user taking into account each new activity you have uploaded via Garmin, Suunto or Strava.

Running, cycling and triathlon.

Garmin Connect, Suunto and Strava.

Gradient Averaged Pace takes into account the elevation gains and losses during your run and normalizes your running pace to an equivalent pace on a flat course. For example, if you're running up a steep hill your raw pace might be slow but your GAP can be high as it takes the steepness of the hill into account. AI Endurance uses GAP to predict your running performance. Your running workout instructions are always in GAP.

Yes, at least for some activities. Power is the only metric in cycling that we found to be predictive. There are too many variations in speed and heart rate data that are not indicative of performance. You don't need all your activities to be recorded with a power meter. If your activity has heart rate and cadence data, we estimate power from those measurements.

You can manually set current performance levels under Account if you feel like the data does not represent your current fitness state well, for instance if you haven't done a test in a while. This should be used sparingly. Our AI automatically picks up most race and test performances. We define FTP as 95% of your best 20 minute power or in a ramp test as 75% of your best 1 min power. We define 5k pace as your best 15 minute GAP and 10k pace as your best 30 minute GAP. For more information, see training zones.

Yes. If you don't have enough data yet we will predict your performance based on the accumulated data of athletes in our database similar to you. We will automatically switch to your personal data as you accumulate more data on your connected apps.

First, make sure that if you manually set your current performance level it is reflective of your actual current fitness state and not an all-time best that might not be representative of your current level. The best way to gauge your current level is to do an FTP test for cycling or a 15 min time trial for running (if you're training for shorter distances) or a 30 min time trial (for longer distances). Our algorithm will automatically pick these up and adjust your current level (see Do you detect my race performances and FTP tests? ). There can also be situations where it's hard for our algorithm to detect your fitness level, for instance if you have rarely done a test or pushed yourself in the past. In this case, our AI might underestimate your level systematically - however the predicted improvements relative to the level we detect are still going to be realized if you follow the plan. Finally, if you have always trained 10 hours a week and you are requesting a plan with 3 hours of training there might be no way to hold your current level. Our predictions are based on the learnings from your data and not just wishful thinking. Rest assured that if this is the case our algorithm is finding the plan that lets you decrease the least for your time constraints.

Here are a few possible reasons:

1. Garbage in garbage out - is your data clean, i.e. from a well calibrated GPS device or power meter? If not you can exclude time intervals where bad data was taken under 'Data Exclusions' in the Dashboard. A common culprit are saved runs that are actually bike rides. You can usually identify those via spikes in pace in your 'My Past Data' plot in the Dashboard.

2. For cycling, have you been riding with multiple bikes and not all of them are equipped with a power meter? If you record an activity without a power meter but wear your heart rate monitor and preferably have a cadence sensor installed, we estimate your power from those measurements. However, if there are lots of rides without heart rate data the model would be blind for a significant fraction of the actual training. Also if you have/had multiple power meters/smart trainers and those read inconsistently relative to one another that can cause issues.

3. Similarly, if you are not recording all your runs with a GPS device the model is blind to those activities you are not recording.

Yes, under 'Account' you can choose between metric (min/km) and imperial (min/mile).

We support 5k, 10k, (Half-)Marathon for running; FTP for cycling; Sprint, Olympic, (Half-)Ironman for triathlon.

No - we do not recommend any structured training in this case. Focus on getting healthy first. The more you focus on getting healthy as opposed to a feared loss of performance the more you minimize your time away from structured training and racing. We always advise you to consult with your physician before following any exercise plan. Our AI does not detect injuries in any form and it is very important that you listen to your body and make game time decisions before and during the workouts that are most beneficial to your health.

We always advise you to consult with your physician before following any exercise plan. The predicted performance is based on the workouts we provide. However, it is very important that you listen to your body and use your own judgement before doing a workout. Pushing yourself too much can lead to injury. We always offer alternative instructions for an easier workout if you're not feeling great and sometimes you just need to skip a workout for extra rest and recovery and that's ok too!

You can export your training plan to TrainingPeaks where you can execute your workouts in Apple Watch, Wahoo, Polar, Suunto and many more. You can also use our Zwift workouts or directly execute the workouts from your Garmin device if you connect your Garmin Connect account with AI Endurance. Simply go to the 'Apps' page and connect with 'TrainingPeaks - Export Workouts' or 'Garmin - Export Workouts'. If you make changes to your AI Endurance training plan, these will automatically be forwarded to TrainingPeaks and/or Garmin Connect. We continuously export your workouts to Garmin Connect and TrainingPeaks up to one week into the future. You may require a TrainingPeaks Premium account to access your AI Endurance workouts in TrainingPeaks.

If you haven't been following your current plan, there might be a better option available for your current fitness state. If your prescribed training significantly deviates from your actual training, we propose you to request a new plan that includes the updated information on your fitness state.

It's harder to provide our AI with clean data to predict swimming performance as it is inherently more noisy and less abundant. We are working on providing you a non-AI swimming plan as a first step soon.

Yes we automatically avoid duplicate activities during import. In the odd case that activities get imported at exactly the same time, we delete duplicates every 24 hours. Activities from Garmin Connect and Suunto always override Strava activities as Strava does not pass on HRV data.

If you have 'Garmin - Import Data', Suunto or Strava connected, only a few seconds to minutes after you upload the activity. If you have multiple connections, we import your Garmin/Suunto activity first and sync with Strava a few minutes later to check if there is any additional information, such as if your activity was a 'Race' or if the activity name tells us you did an FTP test prior to re-training our machine learning model for each user. We make sure there are no duplicates between Garmin, Suunto and Strava even if you have multiple connections.

If your import data includes HRV, it can take a few minutes for your activity to show up in AI Endurance as the DFA alpha 1 analysis is computationally expensive.

Garmin does not pass on third party activities to AI Endurance. To import your Zwift or other third party activities please additionally connect with Strava.

We use your heart rate data to gauge your performance. For instance, if you were able to run 5 min/km or at 200 W for an hour at a significantly lower heart rate than last week, the AI notices this as an improvement in your performance. However, generally heart rate is a much worse predictor for performance than power and GAP.

We calculate your aerobic and anaerobic heart rate thresholds from your DFA alpha 1 data, see this blog post.

For your cycling activities without power, we use heart rate and cadence to estimate your power through a personalized machine learning model.

We track your activity HRV data if you have 'Garmin - Import Data' or 'Suunto - Import Data' enabled and HRV logging enabled on your device. You can easily track your aerobic and anaerobic thresholds from your HRV data via DFA alpha 1. For more information, see also this blog post.

We also use DFA alpha 1 to track your readiness to train, durability and overall fitness improvements. Learn more here.

We automatically attempt to detect your DFA alpha 1 threshold for every activity with HRV data that you import into AI Endurance. For a threshold test, simply export the DFA alpha 1 ramp test to Zwift or your Garmin device from the Apps page.

In order to detect DFA alpha 1 thresholds automatically, AI Endurance has to detect an intensity ramp during your activity as this is the method that is backed by current research on DFA alpha 1 threshold detection. Our algorithm has to be able to identify a time frame that fulfills the following criteria for automatic threshold detection:

1. Steady and smooth increase of intensity, i.e. monotonically decreasing alpha 1 values and monotonically rising pace/power values.

2. Sufficient data points on both sides of the threshold, e.g. alpha 1 < 0.5 and alpha 1 > 0.5 for the anaerobic threshold detection.

3. A not too high ramp speed for increasing the intensity. If you ramp up the intensity too quickly there is a delay in your cardiovascular response relative to your power/pace that would lead to overestimated power/pace thresholds.

You also have the option to download the activity data from the activity page as a csv file and manually perform the ramp detection.

5-10% artifact corruption. Artifact corruption indicates the quality of your HRV data. For minimal artifact corruption, we recommend the Polar H10 heart rate monitor. In your activity page, we show your DFA alpha 1 threshold for any artifact corruption but for analysing your readiness to train we only take into account activities with less than 10% corruption. To be taken into account for AI predictions, we require less than 5% corruption.

The Polar H10 reliably records HRV data with artifact corruption of less than 5%. Other heart rate monitors can also in some cases deliver this quality but you will have to check with your particular heart rate monitor on a case to case basis.