The habit window is much shorter than 21 days

Where “21 days” came from

A plastic surgeon named Maxwell Maltz noticed in the 1960s that his patients took about 21 days to stop reaching for a limb they no longer had, or to get used to a new face. He wrote that it seemed to take “a minimum of about 21 days” for an old mental image to dissolve.

That is an observation about adjusting to something that already happened. It became “it takes 21 days to build a habit,” which is not what he said, not what he measured, and not a claim about starting anything. The number survived because it is short enough to sound achievable and round enough to remember.

The more careful work that followed put the range far wider and far messier — some behaviours settling in a few weeks, others taking the better part of a year, with enormous variation between people. So “how long does it take to build a habit?” does not really have an answer.

But there is a related question that does, and it turns out to matter much more.

The drop-off is not at day 21. It is in the first hour.

RankUp asks people to log real sessions — gym, study, deep work — and prove each one with a photo. That gives a timestamp for the exact moment someone first does the thing, and the moment they signed up intending to.

Across 154 accounts, 46 people have ever completed a session. Of those 46, here is when the first one happened, measured from the moment they signed up:

Nearly four in ten people who ever start, start within sixty minutes. Three quarters start on day one. By the end of the first week the door is essentially shut: only two of forty people ever began after that, and nobody at all began after a month.

There is no version of this chart where day 21 is interesting. The decision has been made long before then.

And then there is the second session

The other cliff is immediate. Of the 46 people who completed a session, 17 of them — 37% — completed exactly one and never came back.

The rest of the distribution is barbell-shaped. Once people get past two or three sessions they tend to go a long way: 15% of everyone who ever finished a session has now finished more than twenty-five.

• Exactly 1 session: 17 people (37%)
• 2 sessions: 5 (11%)
• 3–5: 7 (15%)
• 6–10: 5 (11%)
• 11–25: 5 (11%)
• 26 or more: 7 (15%)

So there are really two questions, and neither is about three weeks. Do you start immediately, and do you come back once. Clear both and the evidence here says you are more likely to end up in the group with dozens of sessions than the group with two.

The result I am throwing out

I expected to find that people who did their second session quickly went on to do more overall. The data says the opposite — the people who waited longer than 48 hours had a higher lifetime average.

I am not publishing that as a finding, because there are five people in that group. One unusually committed person moves an average of five by an enormous amount. It is noise that happens to look like a result, which is what most surprising findings from small samples turn out to be.

I am mentioning it because the numbers above come from the same dataset, and you should know that I looked at more than I reported.

What to actually do with this

If you are trying to start something, the useful reframe is that you are not embarking on a 21-day project. You are trying to clear two specific gates.

• Do it in the first hour. Not today, not tomorrow — in the hour where you still feel like the sort of person who is going to do it. The intention decays faster than almost anyone plans for.
• Make the first one small enough to be certain. The goal of session one is not progress. It is to exist, so that session two has something to follow.
• Schedule the second before you finish the first. 37% of people never get there, and that gap is the single largest loss in the whole funnel.
• Put a small number of people in the room. In a separate cut of the same data, users in a squad of fifteen or fewer completed a first session 57% of the time, against 17% for those whose only group was a large one.

The honest limitations

This is 154 accounts on one app, self-selected by having downloaded something specifically about discipline. Every number here has a small n and I have printed all of them. The time-to-first-session shape is the one I would bet on holding elsewhere, because it is the largest effect and the least dependent on anything specific to this product. The barbell distribution I would treat as suggestive.

I will republish these as the sample grows, including if the picture changes.