How long does it take to raise your FTP by 20 watts?

Twenty watts is close to the smallest gain one re-test can prove

On a 250-watt FTP, the smallest change a single re-test separates from noise sits between 9 and 20 watts, depending on how tightly the test is controlled [MacInnis et al. 2019, Borszcz et al. 2020]. A 20-watt target is not a modest goal. It is roughly the resolution of the instrument.

The parent guide on how to increase your FTP works through the reliability studies in full; the two numbers that matter here are the ends of the range. Tested a week apart, 25 trained cyclists returned a coefficient of variation of 2.9% on the 20-minute protocol, with the prescribed 45-minute warm-up varying nearly twice as much [Borszcz et al. 2020]. Under laboratory control, eight highly trained riders held the typical error to 4.6 watts on trials averaging 343 watts, about 1.3% [MacInnis et al. 2019]. Your re-test sits somewhere between those two conditions, and almost certainly nearer the first.

Multiply the typical error by roughly 2.8 and you have the smallest change one individual can detect at 95% confidence. At the 2.9% end, a 250-watt rider needs about 20 watts before one re-test can tell a gain from the day. At the 1.3% end it is about 9 watts. The rider who asks how long 20 watts takes has, without meaning to, picked the first gain a single test can defend.

Two things follow. The date you own the watts and the date you can prove them are different dates, and the second one is later. And the looser your protocol, the further apart they drift — a re-test with a different warm-up, a different trainer, or a different week of the block is measuring something the first test did not measure. Holding the protocol still is cheaper than gaining the extra watts it takes to see through a sloppy one.

The same 20 watts is a different ask at 200 and at 350

Twenty watts is 10% of a 200-watt FTP, 8% of 250, and 5.7% of 350. In the trial that measured it most directly, three weeks of short intervals moved elite cyclists' threshold-region power between 2.0% and 4.7%, depending on which marker was read [Rønnestad et al. 2020]. Only one of those three riders is asking for something inside that band.

The 350-watt rider is asking for 5.7%, which sits above the top of that band, and those were riders who were already well trained. That does not make it impossible; it makes it a season-shaped target rather than a block-shaped one. The 200-watt rider is asking for 10%, which sounds worse and is usually the easier of the two, because a rider at 200 watts typically has training history still to accumulate rather than a ceiling to push. Percentage is the wrong instinct here. Training age is the variable that moves.

The clearest read on the less-trained end comes from 14 recreational male cyclists, mean age 41, who completed a 16-week pyramidal programme. Power at a 4 mmol/L blood lactate concentration improved with a large effect size, d = 1.21, and the relative improvement correlated most strongly with time spent in the moderate zone between 2 and 4 mmol/L rather than with time above it [Magalhães et al. 2024]. The study had no comparator group, so it shows that these riders improved, not that the distribution is what improved them. It is still the population where the larger relative gains live.

What the trials observed, and over how long

Three weeks is the shortest window over which the studies cited here detected a change in 20-minute power, and that was nine high-intensity sessions in elite riders [Rønnestad et al. 2020]. The others ran 10 and 16 weeks. Inside every one of them, the spread between riders is wider than the difference between protocols.

Nine elite cyclists at 73 ± 4 mL/kg/min did three weekly sessions of short intervals for three weeks — three series of 13 efforts of 30 seconds, with 15 seconds between them. Mean power in a 20-minute test rose 4.7 ± 4.4%. An effort-matched long-interval group of nine fell 1.4 ± 2.2% over the same three weeks [Rønnestad et al. 2020]. On a 350-watt rider, 4.7% is about 16 watts. The spread matters more than the mean here: ± 4.4% means riders running the identical protocol, under identical supervision, ranged from roughly no change to about 9%, which on that same rider is 0 to 32 watts.

The longer windows tell the same story with different measures. Over 10 weeks of twice-weekly high-intensity sessions, short intervals produced a VO2max improvement of 8.7 ± 5.0% against 2.6 ± 5.2% for effort-matched long intervals, with moderate-to-large effect sizes across 30-second, 5-minute and 40-minute all-out performance tests [Rønnestad et al. 2015]. Note what was measured there: that study reported effect sizes on those three performance tests, not a percentage change in 20-minute power, so it is evidence about the format rather than a number you can put on your own re-test.

Which leaves the planning answer as a range with a shape. Three weeks is the floor, and it produced 4.7% on average — about 16 watts for a rider at 350 — with a real chance of nothing. Ten to sixteen weeks covers the rest of the evidence here, and a rider who has been training consistently for years should plan across two blocks rather than one. Every one of those trials also controlled total volume and supervised the sessions. Your block will have a work trip in it.

What to watch while you wait for the number

Re-testing every four weeks mostly manufactures noise you then react to. Three signals move earlier and cost nothing: the same power at a lower heart rate, more time held at the power before you crack, and the power curve rising at durations you never set out to test.

Each of those is a within-ride observation rather than a test day. Heart rate at a fixed wattage falling across a block is a straightforward read that the same work costs less. Holding a threshold interval for 12 minutes in week one and 18 in week seven is a gain in usable time at that power, which is often what a rider actually wanted from the extra watts. And a power curve that lifts at 5 minutes or at 8 while the 20-minute point sits still is evidence the block is working on something, just not yet on the coordinate you keep checking. To use any of them you need the bands fixed first — a power zones calculator turns one FTP number into the seven ranges you ride to, and leaving those alone for the length of the block is what makes week one and week seven comparable.

Narrowing the wider question to a single target produces one uncomfortable conclusion worth stating plainly: for a rider already deep into their training history, 20 watts is not a one-block target, and a plan that promises it in eight weeks is selling the group mean as a forecast. Nothing in the studies here supports that promise for a named individual. What they support is a range, a floor of about three weeks before anything is detectable at all, and the ordinary advice to measure the same way twice.

Quick answers

Sources cited in this guide

1. 01
   Borszcz et al. 2020. Reliability of the Functional Threshold Power in Competitive Cyclists. International Journal of Sports Medicine.
2. 02
   MacInnis et al. 2019. The Reliability of 4-Minute and 20-Minute Time Trials and Their Relationships to Functional Threshold Power in Trained Cyclists. International Journal of Sports Physiology and Performance.
3. 03
   Rønnestad et al. 2020. Superior performance improvements in elite cyclists following short-interval vs effort-matched long-interval training. Scandinavian Journal of Medicine & Science in Sports.
4. 04
   Rønnestad et al. 2015. Short intervals induce superior training adaptations compared with long intervals in cyclists - an effort-matched approach. Scandinavian Journal of Medicine & Science in Sports.
5. 05
   Magalhães et al. 2024. Effects of a 16-Week Training Program with a Pyramidal Intensity Distribution on Recreational Male Cyclists. Sports (MDPI).

More inside How to increase your FTP

Other articles in this series

1. 01

   Watts or weight: which actually moves your W/kg?

   One kilogram is worth about your current W/kg in watts. Why that makes the denominator look cheaper than it is, and where W/kg is the wrong metric.

   →