Why your to-do app died

You reopen it on a Sunday evening, because something has to change.

Thirty-four items. You scroll. Three have been done for ages, you just never came back to tick them. Two are for a trip that is already over. One says "sort out garage", and has been saying it since February, which means you have read it at least twenty times and done nothing about it.

And then there is the date. The most recent thing you added is five weeks old. That is not the day you decided to stop. It is just the day you stopped.

You look at it for a minute, and then you do what everyone does. You start again. New list, better categories, this time properly.

Before you start again, one question

Why five weeks?

The usual answer, the one you give yourself as you close the app, is a lack of discipline. You did not keep it up. Other people keep it up.

Except nobody told you what the pass mark was. And it was much higher than you thought.

Where the number in your head comes from

You probably know it takes twenty-one days to form a habit. Everyone knows. It is in the self-improvement books, in the tracking apps, on the posters in the gym.

That number came from a cosmetic surgeon.

In the 1950s, Maxwell Maltz noticed that his patients often took about three weeks to adjust to their changed appearance, and he wrote it down in 1960 in a book that sold in the millions. It was an observation about recovering from surgery. It was not a study, he was not counting anything, and he was not talking about habits.

The number outlived its context. It arrived in your app.

What happened when somebody actually counted

In 2009 a team at University College London decided to measure it properly. Phillippa Lally and her colleagues recruited 96 volunteers. Each had to pick one thing to do every day, always in the same context. A piece of fruit at lunch, a glass of water after breakfast, a fifteen-minute walk after dinner. Nothing ambitious, deliberately.

Then, every day for twelve weeks, each person recorded two things: did I do it, and did it take effort. That second point is the real subject. A habit is doing the thing without having to decide to.

They then plotted, person by person, the point at which the effort stopped decreasing.

The result is not a number. It is a range. From 18 to 254 days. The median was 66 days.

One caveat before going further: Lally's paper is more careful than the use made of it. Her model fitted the data well for only about half the participants. Which means "66 days" is a median drawn from a very wide spread, not a rule that applies to you. The figure is now repeated with the same automatic confidence as the 21 days it replaced, which is fairly ironic.

Now look at the upper end of that range. In a study, on a behaviour the person had chosen themselves, with a researcher checking in daily, some people needed the best part of a year before a tiny action became automatic.

Your app, meanwhile, was waiting for you to maintain a list. Not a piece of fruit at lunch: a list, which has to be fed, sorted and cleared. And it was going to wait however long that took before it gave you anything back.

You gave it five weeks.

Lally's study also contains a result almost nobody quotes, although it is the most useful one: missing a day did not stop the habit forming. No break, no reset. So the unbroken run is not what forms the habit. Every app that has punished you with a snapped chain was working from a belief, not from data.

And meanwhile, the list grows

A second mechanism explains the thirty-four items.

In 1994 three Canadian researchers, Roger Buehler, Dale Griffin and Michael Ross, asked students writing their dissertation a simple question: when do you think you will finish?

Average answer: thirty-four days. Actual: fifty-five. Fewer than a third met their own estimate.

These were not strangers guessing at someone else's work. These were people estimating their own work, which they knew, and had done before in another form. And the work took about 60 percent longer than they said, consistently, in the same direction.

So you dated your tasks with the same optimism, because everyone does. Deadlines slipped. What was not done on Monday reappeared on Tuesday, then Wednesday.

A list that grows faster than you clear it stops being a memory aid. It becomes a daily statement of how far behind you are. From that point, opening it costs more than not opening it, and you stop opening it.

At that point, avoiding the app becomes entirely understandable.

For scale, you are far from alone. Industry data aggregators, such as Business of Apps compiling AppsFlyer figures, put average day-30 retention at around 5 per cent on iOS and 4 per cent on Android, across all categories. These are industry averages rather than a scientific study, so treat them as an order of magnitude. The order of magnitude is that almost everybody leaves.

What to ask of the next tool

A last caution before the practical part: none of this proves any particular tool works better than another. It shows only how much you are asking of someone when a tool depends on weeks of uninterrupted daily use. That is an argument about design, not proof that anyone has solved it.

Judge a tool on a bad week. For the first three days everything is new and you are enjoying it. The real test is the week you are ill, travelling, or simply flat.

A gap has to be recoverable. If four days away produces a red wall of overdue items, the tool has turned an ordinary human pause into a reason to quit. Lally's data suggest the pause itself does little harm; it is the way the screen presents it that puts you off coming back.

Clear the backlog instead of letting it grow. What you have deferred three times will not be done on the fourth. Delete it, or give it a real date and a real slot. Keep rolling things forward silently and the list ends up holding only what you have not done.

Look at how long it actually took last time. You do not estimate badly through carelessness. Buehler's students were estimating their own work, and it took about 60 percent longer than they said. The only correction that works is to look at the record.

Prefer tools that do not depend on you maintaining them every day. That is the firmest conclusion in all of this. Rather than asking how to last 66 days, ask what carries on working on the days you do not.

A tool with no list to maintain

That is the thinking behind Nitka (heynitka.com): there is no list to maintain. You say the thing the moment it occurs to you, by voice, text or photo, and it does the structuring and brings it back at the right time.

It also does something about the task that keeps sliding. Once it has reminded you of something three times and the task is still open, it stops reminding you and asks a single question, headed "Still on your list": pick a day that really works, or mark it Not needed. It asks once per task, and it never closes a task by itself, because only you know whether it still matters.

It is not the only possible answer. A weekly review you genuinely stick to is another option, and if yours has survived a bad month, it works.

What it cost you to open the app

Do not start again with better categories. This is the fourth time, and the categories were not what went wrong.

Ask the smaller question instead. That Sunday, five weeks ago, what did it actually cost you to open this app? Thirty seconds of sorting? The sight of the garage from February? The arithmetic of what you had not done?

Whatever the answer is, that is the thing to fix. Everything else is just another redesign of the same system.

Disclosure: this article is published by Nitka, which appears above. The research cited exists independently of us, and the links below go to the original sources.

References

Phillippa Lally, Cornelia van Jaarsveld, Henry Potts and Jane Wardle, "How are habits formed: Modelling habit formation in the real world", European Journal of Social Psychology, 2010, 40(6), 998 to 1009. 96 volunteers, twelve weeks, 18 to 254 days to reach 95 per cent of maximum automaticity, median 66. Also contains the finding about the missed day. https://doi.org/10.1002/ejsp.674 (paywalled, the publisher gives access to the abstract only)
Maxwell Maltz, Psycho-Cybernetics, 1960. The origin of the twenty-one day figure, as an observation about cosmetic surgery patients rather than a study of habits. Available to borrow: https://archive.org/details/psychocybernetic0000malt
Roger Buehler, Dale Griffin and Michael Ross, "Exploring the 'planning fallacy': Why people underestimate their task completion times", Journal of Personality and Social Psychology, 1994, 67(3), 366 to 381. The students, their dissertations, and the gap between 34 and 55 days. https://doi.org/10.1037/0022-3514.67.3.366 Free PDF: https://web.mit.edu/curhan/www/docs/Articles/biases/67_J_Personality_and_Social_Psychology_366,_1994.pdf
Business of Apps, "App Retention Rates", from AppsFlyer data. An industry aggregator rather than a peer-reviewed study, and across all categories rather than specific to productivity apps. https://www.businessofapps.com/data/app-retention-rates/

Nitka, the app mentioned above, turns what you say, type, photograph or forward into dated reminders, and tells you when two things you told it do not fit together. It keeps only what you give it, and your data is never sold, never used for advertising and never used to train models.

Try it free for three months.* On the App Store now; on Android, ask for early access.

* Then $0.99 a month, cancel any time. More at heynitka.com.