Zwift's New Ramp Test (FTP)

Coaches that are just scheduling workouts are already out of a job. If a coach isn’t adding something additional on top from a relationship aspect, they probably don’t have many clients at this point.

As for the AI/machine learning, there is a lot of hype and not a lot of results yet. Certainly not to the point that would replace structured assessments. There is not a single product on the market yet that is even what I’d consider 25% of the way to being a minimum viable product for “adaptive training”. It will happen eventually, but it’s further away than people think. But that’s just my opinion as someone who works with high level ML engineers regularly.

Also: if someone isn’t willing to empty the tank and give a max effort to get an FTP, chances are they have different motivations entirely that might be better served with a different type of experience.


I’m no expert but it would appear to me that for average folks who are not racing frequently or willing to do fairly frequent testing at a full range of duration, the data is not going to be good enough to use AI to come up with high quality program and adjust it on the fly. At least not whereas you’d be getting anything much better than a good off the shelf plan.

I love trainerroad, but my “winter” data vs my “summer” data look like two completely different riders. Its not because the TR plans are bad, its just that I have very little maximal data getting in there in the winter and what there is is solely FTP tests, not full range power profile tests. Personally, while I enjoy maximum efforts during races and group rides and do enough varied stuff to get a decent power profile, I have no interest in repeated full range power profile testing on the trainer just to feed data into an AI program. Be careful what you ask for :wink: If you are not willing to test a lot, the current version just might be good enough!

As a side note, in the long run Zwift may end up having an advantage in this space as they are collecting a full range of maximal data for folks who race or do the hard group rides.


TR is hinting at going in the opposite direction, no more testing and using your performance on your workouts to adjust your FTP.

No more of an advantage than any company that authenticates with Strava’s API and pulls in ride history (i.e. everyone). If you’ve sync’d your Strava account with TR and you ride on Zwift and send it to Strava, TR has the data. Zwift also has the data that you ride on TR, but they don’t have the advantage of knowing what the workout was supposed to be (similar if you do workouts on Zwift, TR won’t know what the workout was exactly).

TR has the biggest advantage of any company on the structured data front currently.

Exactly. You really don’t need AI to make a training plan for the average person, and their FTP doesn’t have to be perfectly set. They just need to do the workouts and they will be better off than 95% of the people training out there. You could probably just remove the FTP and the wattage entirely and they would get faster.

For an advanced rider, if they aren’t willing to do assessments but they want high level guidance, I’m not really sure what to tell them other than HTFU. In order to win, you have to be willing to empty the tank.


All true. Unfortunately it’s the average person (enthusiast level) that seems to be most concerned with FTP and wattage. If coaches and apps can’t deliver more watts (or whatever metric of improvement they offer) they may struggle to maintain users. Unfortunately there’s not much sex appeal in selling the idea of consistent training, adequate rest, and common sense nutrition.

The people who I know personally that worry the least about numbers day to day tend to be elite athletes.

1 Like

It’s like personal finance. Make more, spend less, invest the difference. Can’t sell a lot of books or charge high commissions that way unfortunately.

1 Like

The problem with that approach is that you need maximal efforts to build a full picture of a rider and workouts don’t necessarily give you that, especially TR’s workouts which don’t have much in the way of maximal level efforts much over 8 minutes (nothing wrong with that, a workout and a maximal test are two different things). TR workout alone are not going to build out the data for your full power curve. Anyone who gets stuck inside for most of the bsae, build specialty cycle can see their data does not paint a full picture. So, unless you race, or do some really freakin’ hard rides, you’ll have to test, at least if you want a sophisticated output.


I am going to do the Z ramp test today. I have a pretty decent idea where my FTP is based on what wattage I am hitting in my workouts. I will share the results here.

Anyway, being quite biased to the shortest part of the power curve, I am expecting a ramp test to overestimate my FTP since the anaerobic contribution for building my power in the short steps will be much greater than the average,

Coach Alan Couzens (a mad scientist - as he calls himself - that works mainly with triathletes) is always warning about the need for the ramp tests to be more gradual and with bigger time intervals for each level. This applies to all the measured parametes like LT, VO2Max, FatMax, Metabolic Ratio, etc…
He says that labs try to fit the ramp tests in a short period of time due mainly to economic reasons and most of the times the results suffer.

Maybe a 1 minute 20w ramp has the same problem,

1 Like

I think you’re bringing in a big assumption about how you can set training levels based on how people currently do set training levels.

TR has other useful information. They know what you were supposed to do in a workout, and they know what you actually did – whether you succeeded. Further, if you use one of their plans, they control what workouts you do. So, they have a feedback system where they could, say, manipulate your workouts until you start failing.

It seems to me, based on their recent workout and plan redesigns, that they’ve been working internally on compliance metrics – judging whether you succeeded in a workout and, if not, why not (or how not). It also seems from the workout text that many workouts are designed with the expectation that there will be some acceptable rate of failure. Take from that what you will.

1 Like

I did the Zwift test in an erg rollers and the results look far from ideal.

The main problem came from using Zwift’s powermatch.

At a certain moment in the test, when the power went up 20w the resistance of the roller decreased like 50w at first and then overcompensated.

This made the test quite erratic and difficult so I guess that I could have reached a higher power even with the added cognitive load that come from the rollers vs a trainer.

This is the first time I had problems with their powermatch.

I could have done the test using the power measured in the roller but the FTP estimate would be completely context dependant and probably useless in a outdoor situation.


So, for me it gets a C-

I will not set training zones based on that figure.


Zwift has powermatch?

Sort of…
When pairing the user can setup the
roller/trainer ANT FE-C device for erg and slope but the bike’s PM for power and cadence. This should mean the erg mode gets the power feedback from the bike’s PM and not the roller/trainer.

In addition to my previous comments, yesterday I did a short race in Zwift that took me 35 minutes and my 20m power during the race multiplied by 95% is spot on with their ramp test results. However, because I raced 15 minutes more, those 20m cannot be considered max effort so I woul say Zwift indeed underestimated my FTP by 10w or so.
I am doing a one month test of Zwift to see how it works and so far the conclusion is that I will be using only Traineroad as soon as that month ends.


I would not draw that conclusion because the 95% of 20min test protocol expects a 5min max effort beforehand.


Oh yeah, I see how that can be. Thanks for the tip!

1 Like

Good point there. Then maybe it was ball park.
I will repeat the ramp test soon to get a clearer idea.

I’ve used the TR ramp test for quite some time now (from beta) and have generally tested pretty consistently with nominal improvement. After spending some time doing weekend zwift races, I found my ftp increased nicely from 208 to 216 (using the same step test), probably due to the sustained nature. After another 4 weeks riding at 216 I tested again, this time using zwifts step test.

First thing I noticed was the steps are 20w vs 6% of ftp, which in my case was 20w vs around 12w. As a result, I was able to sustain a higher wattage (300+) for 1min, possibly because I got there quicker, rather than the slower ramp to get to my previous max of 288. Zwift then determined my ftp to be 230 (which I’d surely take than increase from 216).

So the question is… What’s going on? Is this zwift thing accurate? Have I actually improved from pushing harder? Or is it a result of being fresher at higher wattage resulting from a shorter test?

Granted, ftp is nothing but a benchmark to set yourself to for intervals and after trying 230, if I’m able to suffer my way though, I can’t imagine being any worse off, if anything it may result in more gains?

Any thoughts here?

Whay don’t you take part in the FTP challenge and find out. The FTP Challenge



You can’t compare FTP tests with two different protocols, that’s with TrainerRoad, Zwift or any other program. Consistency in testing is the only way to ensure improvement, aside from any discussion if one test is better than the other.

You would be worse off if you are using TrainerRoad with a higher FTP. Essentially you would be training in the wrong zones. When you think you’re doing a zone 2 workout, you would actually be in tempo. Then all your sweet spot workouts become threshold. It is possible that you could maintain this in the short term, but eventually you would be overreaching and start to dig yourself a hole.

Yes, that is the only risk there, in terms of overreaching. I’m a low volume guy so could definitely use a bit more to see real gains, but after doing TR for a few years and going from 201-203-208-216-230 seems a bit insane at the end of that sequence.

Just listened to the Zwift podcast with Greg Henderson, Kev Poulton and Matt Rowe and Poulton made the interesting point that he uses the ramp test to verify the results of an all out 8 minute test which is done after the ramp test but within the same session.
From memory he uses 72-78% of the ramp then confirms by taking 90% of the 8mins. That would be a brutal session. Definitely had never heard of this protocol before.

Here is the link for those interested