What is Enfusion and why should I care? Part 1

What is Enfusion™ and why should I care?

Enfusion is the first model of human performance based on machine learning and big-data analytics. That’s great, but what does that mean, and how can Enfusion make me faster? In the next series of blog posts we’ll explore the current standards for tracking fitness, along with their limitations. Then we’ll demonstrate how Enfusion can address these limitations and provide critical insight into your training where not previously possible.  

The Old Way: Tracking stress  

The most popular model of performance used by athletes today is based on tracking accumulated training stress over time, which was first put forward by Dr. Andy Coggan well over 10 years ago. Dr. Coggan adapted this approach from Dr. Eric Banister’s Impulse-Response model, first proposed in the mid-70’s. In cycling, a gold-standard measurement of overall performance is maximum 1-hour power (FTP). Dr. Coggan proposed that a potential way to benchmark training stress could be to measure today’s training relative to your current maximum ability during a maximum 1-hour effort. This produces what he calls the “Training Stress Score,” or TSS. The formula for TSS is as follows:

TSS = (seconds X Normalized Power X Intensity Factor) / (FTP X 3600) X 100

Dr. Coggan’s proposal of TSS was the first practical method for athletes to track and quantify their daily training load based on their current physical ability. TSS could then be averaged over time to monitor chronic and acute training loads in order to manage what Dr. Coggan calls, Training Stress Balance (TSB). By tracking the balance of acute and chronic training loads, coaches and athletes could make sure athletes had trained hard (chronic), yet were rested (acute) prior to key events. Together, Dr. Coggan’s work developing TSS and TSB helped increase the scientific rigor for the training of mainstream athletes and has served as the standard for the last several years. In fact, Dr. Coggan’s model or a derivative is the basis for nearly all online training analysis platforms used today.

However, despite all the good, there are many limitations to this approach. Here, we’ll take a look at some of the limitations in these stress-based models and provide evidence for how Enfusion can address these problems, as well as help athletes learn more than they ever thought possible about their training.

A single-value stress metric can mask the underlying differences in your training:

It is certainly acknowledged that any scientific model or approach has limitations. Despite the utility of TSS, there are some real limitations that must be considered. First, using a single metric to assess the physiological impact of training means that rides or workouts that are dissimilar can have equivalent TSS values. In some respects, this is the whole point of TSS — to put dissimilar workouts on a single scale to allow comparison. However, let’s take a look at what happens in this real-life example below.

For a rider with an FTP of 300 watts, a 1-hour max time trial is equal to a TSS of 100. Now let’s say the same rider does a steady 4-hour ride at 150 watts average.  Using the formula above we find this ride also has a TSS of roughly 100. Thus, according to TSS, these rides are equivalent. However, from practical experience we can say that these 2 efforts are clearly very different. Here are 3 examples why.

  • The 2 rides feel different. Although simplistic, this is an intuitive way to infer the load on the metabolic fuel systems (aerobic vs glycolytic) in the 2 different efforts.  In this example, 4 hours at 150 watts would be an easy/moderate pace, where one’s legs likely don’t “burn” very much, if at all. In fact, you could carry a conversation the entire time. That’s because all of the metabolic substrates are used to completion, and glycolysis never occurs at a high-enough rate to overwhelm the buffering capacity of the muscle (or blood). This is clearly not the case in a 1-hour time trial, which is an excruciatingly painful effort of mammoth proportions. Glycolysis is proceeding at such a high rate that lactic acid begins to accumulate, and pH in the muscle (and blood) drops, producing the characteristic burn we all know.  Thus, overall energetic flux during the TT is maximal for an all-out 1-hour effort.
  • Total work is different: In contrast to a maximal 1-hour effort, the energetic flux during a 4-hour ride at 150 watts is submaximal. What we mean by submaximal is, at any point in this ride the rider could go faster, releasing more energy. However, the overall energy expenditure at the end of the ride is 2-fold higher, again supporting the underlying difference. Let’s take a closer look.

1-hour ride at 300 Watts = (300W X 3600 seconds) / 1000 = 1080kJ

4-hour ride at 150 Watts = (150W X 14,400 seconds) / 1000 = 2160 kJ

  • The physical adaptations are not the same: Again, we can intuitively assume from practical experience that the physiological adaptations that occur following these 2 rides with equivalent TSS are not the same, which shows that the stress of training is not the same. Again, we can look to practical experience for an example. Let’s assume we have 2 riders with equivalent talent, and the same FTP. Imagine that one person trains for 4 hours per day at a low intensity. The other trains at an equivalent TSS, but with less volume and higher intensity. The first person doing high volume, low intensity training will rarely (if ever) be able to keep up in a time trial with a person who trains to be at their maximum for a 1-hour max power effort. We can look to other examples: people who target centuries often do poorly in criteriums, and vice versa.

Does TSS actually measure stress?

TSS = (seconds X Normalized Power X Intensity Factor) / (FTP X 3600) X 100

Returning to the TSS formula, we see that the numerator contains ride time in seconds multiplied by Normalized Power and an Intensity Factor. The denominator contains maximum sustainable power for 1-hour (FTP) multiplied by 3600 second (number of seconds in 1-hour). Now let’s look at the formula for energy, or Joules:

Joules = Watts X seconds

Therefore, another way to look at TSS becomes:

(Joules for today’s workout*  X Intensity Factor) / (Joules at FTP) X 100

In other words, TSS puts the energy required to complete today’s workout on a scale relative to the energy required to complete a 1-hour max effort. This means that tracking TSS over time is more directly related to the relative amount of energy expended over time than with understanding accumulated stress associated with training.

Taken together, we can see describing the physiological “stress” of a training episode based off of a single descriptor is overly simplistic, and can actually mask important details about an athlete’s training. In our next blog, we’ll describe how these problems are compounded when trying to evaluate fitness and fatigue via Coggan’s Training Stress Balance approach.

Ok, so how does Enfusion pull out the critical details of endurance training data to help athletes know exactly what they’re doing and how it compares to their athletic goals?

*Normalized Power is measured in Watts. When multiplied by time, Normalized Power can be converted to Joules. However, Normalized Power requires raw average wattage to undergo a transformation that will be discussed in a later blog post.

The New Way: We don’t care about stress

The use of “stress-based” performance tracking systems have provided insight at a time when there was none, and moved the sports science field forward. However, training stress is extremely difficult to measure accurately without using invasive techniques (muscle biopsy, blood tests, etc.). In addition, current stress-based systems cannot utilize the flood of data collected by today’s training devices, and require a one-size-fits-all approach for all users. We also know that cycling, as a whole, has an extremely diverse population of athletes. If you’re a BMX racer or ultra-endurance racer, your events could last from seconds to days, and each of these disciplines has specialized training approaches.

There are also significant differences in the individual characteristics of athletes. We can look within the field of sports science, as well as other scientific fields (e.g. genetics), to support the idea that these differences among individuals are meaningful and need to be accounted for. Thus, our goal when creating Enfusion was to create a system flexible enough for every discipline and every athlete. We try to make as few assumptions about how you respond to training as possible, and let your data tell the story.

With this in mind we created a new model of performance from the ground up. This includes the development of a novel suite of performance metrics able to more completely describe an athlete’s unique qualities. Our goal is to, wherever possible, use simple percentage based metrics and physical units like miles, watts, beats-per-minute, etc.  Not only does it make the metrics easier to understand, but it means that our metrics are based in reality. Finally, these metrics must fit together to describe an athlete’s performance, both quantitatively and qualitatively, and that is where Enfusion really shines.

Tracking Training Load versus Training Stress:

Since stress cannot effectively be measured non-invasively, we chose to focus on what we could measure: energy expenditure. More specifically, we chose to focus on how much energy riders can release across every time point, from 1 second to 6+ hours. As you see in Figure 1, we can build a Power Decay Curve for each athlete by mining training and race data to reconstruct an athlete’s maximum ability for any time point, as well as track changes in max power over time. As we will discuss below, training and races can be compared against an athlete’s maximum ability to give a more precise estimation of training intensity.

Figure 1. Enfusion can mine rider data to reconstruct a rider’s max power across time. This curve can also be used to estimate a rider’s maximum workload by multiplying ride time by wattage. Workouts can be measured against this curve to understand Percent Effort.

In addition, Enfusion measures another critical aspect of energy expenditure: Variance, or how steady or chaotic an effort was. We make direct measurements of this aspect of training and put it in the context of an athlete’s current physical ability. Together with Enfusion’s other key metrics, we can accurately describe an athlete’s training, and are used in our proprietary model of performance (Enfusion Score™) to measure changes in fitness following each workout.

Finally, we are able to dispense with the limitations of stress-based performance models and create the World’s first method capable of tracking form quantitatively and qualitatively on a daily basis. Below, we outline and define some key metrics Enfusion uses to track/describe training load, and help quantify performance.      

Watt*Hours: We believe that the most direct method for tracking workload is through energy expenditure. Energy is most often tracked in joules (or kilojoules), which is equal to watts*seconds. However, this often gives you a large number and requires figuring out how many seconds are in your workout. This can be annoying if you are not looking at your training device or are not near a computer where you can see the actual file. We asked why not just use hours instead of converting to seconds and save yourself a step? Instead of Joules (or kilojoules), you wind up with Watt*hours, which are exactly the same units used for rating light bulbs. Now, all you need to know is average wattage and time, and you have an accurate measure of your workload. However, if you prefer to use kilojoules, you can certainly continue to do so. All the metrics that follow will still apply.

Percent Effort™: Ok, so now we have a quick and easy way to track workload, but what does it mean? As you see in Figure 1, we can model any athlete’s maximum power output over time. For any workout, we can now compare the length of time for that workout on the x-axis, and see what the corresponding max wattage value is on the y-axis (this is automated, of course). Now multiply time by wattage to see the maximum workload an athlete can handle. Let’s look at an example.

An athlete rides for 1.5 hours at 200W, or 300 Watt*Hours. As you can see in Figure 1, the athlete’s max wattage at 1.5 hours is 220W, meaning their maximum expected workload at 1.5 hours equals 330 Watt*Hours. If we divide today’s workload for 1.5 hours by their predicted max workload and multiply by 100, we see the following:

(300 Watt*Hours / 330 Watt*Hours) * 100 = 90.91%

This means that a workload of 300 Watt*Hours for today’s ride was 90.91% of what this athlete is currently capable of doing.

Percent Variance™: If workload is the “cost” of a workout, then Percent Variance describes how you “spent” the energy to get the job done. For example, a rider doing a flat time trial would have a steady and consistent release of energy. In contrast, a rider doing a 6-corner criterium or mountain bike race could be constantly surging and coasting during the course of the race. In Figure 2, you can see the difference between a recovery ride on the trainer (top) versus a mountain bike race (bottom). Why is Percent Variance so important to consider in your overall training plan? Because the energy systems required (along with the degree to which each system is used), can be very different. This can have serious implications on recovery, as well as the type of physical adaptation gained from training/racing. Therefore, this aspect must be accounted for to accurately train to the demand of a given event or goal. In a similar way to how we model max power over time, Enfusion models an athlete’s maximum tolerated variance across time. Each workout can then be compared against this maximum to calculate Percent Variance for any workout, which helps determine how difficult this aspect of the workout was for you. Enfusion is the only system available that directly measures variance, and can track changes over time.

Figure 2. Enfusion is the only system in the world that allows riders to measure how much variation of pace they can handle for any length of time. With our Percent Variance metric, riders can, for the first time, track this aspect of training. Top: Wattage chart for a recovery ride on indoor trainer equal to 0.53% Variance. Bottom: Wattage chart from a mountain bike race representative of 100% Variance. Images are real data. Percentage values stated represent an athlete’s actual max ability.

Pulling it all together:

Ok, so Enfusion offers a new way to look at training data. But how do you use it to take your training to the next level?

Enfusion allows you to define the physical demands of any event, and easily create concrete training goals in the context of your own ability using 4 key metrics: time, Watt*hours, Percent Effort, and Percent Variance.  Let’s look at an example: Your major objective every year is the state mountain bike championship. Last year, you finished 3rd. You look at your data from the race and you see that the race lasted 2:15:00, you produced 450 Watt*hours, at 96% Effort, and 88% Variance.  Together, these 4 values help decide not only what type of training you need to do, but they constitute the benchmark that you must meet or exceed in order to be prepared for this year’s state championships. For example, given the high degree of variance in last year’s race, you should likely focus on short, high-intensity intervals versus prioritizing 6-hour easy rides (not that you need to totally neglect endurance). You would also need to increase your max Watt*hours around 2 hours.

You can more precisely track the training load (pushing but staying within your limits), avoiding problems of single-metric “stress based” systems. Let’s continue with the same example. As you train in preparation for the state championship, you can precisely monitor how hard you’re training in relation to your key event with Percent Effort and Percent Variance. These two metrics help ensure recovery days are actually easy and that difficult days are appropriately difficult. You can design workouts or find test events (like a Tuesday night crit series) that mimic key aspects your goal event, such as Percent Variance.

You now can measure the progress in your training against that concrete objective.  Not only can you track training load, but you can easily see your progress. Let’s say you started training in January and the race is June 1. By March, you were able to do match your Watt*hours from last year’s race: 450 watt*hours at 2:15:00 which is equal to 96% Effort. In mid-May you do a tune-up race, and find out that 450 Watt*hours for the same time is now equal to around 94% Effort. Based on this, you can have concrete data to suggest you are well prepared for your race in 2 weeks.

Looking ahead:

Accurately tracking your training load is critical to achieving your athletic goals. But it is only half the story. You also have to account for the impact of the training on the athlete, with respect to fatigue, from day-to-day. It does no good to train hard if you don’t allow yourself to recovery appropriately. In addition to tracking your training with increased precision, Enfusion is also the only system currently available that can estimate changes in the quality of your performances on a daily basis. This means Enfusion can more accurately detect over-reaching, help avoid overtraining, measure recovery, as well as detect breakout performances.  In our next blog post, we will describe how we accomplish this through our machine learning approach, and how this moves beyond the current standard of evaluating acute versus chronic stress balance.

Thank you for reading and we hope you are as excited about the potential of Enfusion as we are!


-The Enfusion Team

TSS, TSB, Normalized Power, and FTP are all registered trademarks of Training Peaks, and PeaksWare, LLC.

Leave a Reply

Your email address will not be published. Required fields are marked *