TL;DR. A habit rarely dies from one missed day: in the best-known field study, missing one opportunity did not materially affect habit formation. Yet in check-in records from 30,110 gym members, the longer the gap since the last visit, the less likely the next one, for 76% of them. The miss is cheap; what turns it into a gap is the failure, and the causes differ. The cue disappears: students who changed university kept old habits only where the context survived. The plan is too rigid: a paid daily two-hour window gained 0.39 in-window workouts a week and lost 0.56 outside it. The payoff ends: in the first month after a workplace gym incentive stopped, 25% of the gain was left. Or a reminder did the remembering: 94% adherence, lowest automaticity. This piece sorts them into nine graded failure modes. The CEO move: run a post-incident review and fix the failed component, not your willpower. The student move: treat the break as data and run one small repair experiment with a measurable check.
This is the diagnosis companion to how long a habit takes to form, habit stacking for interrupted days, which keystone habits pay most and what an AI assistant can do for accountability. Those cover installation. This one starts where a routine was running and now is not.
What counts as a broken habit?
Three different things get called a broken habit.
A routine that never became a habit. In Lally and colleagues’ 12-week study, the habit curve fitted well for 39 of the 82 people with usable data, 47.6%. A study of 111 new gym members put the minimum for an exercise habit at four sessions a week for six weeks. Many “broken habits” were routines still running on intention, a reminder or an incentive.
A lapse. One or a few misses inside a routine that otherwise runs. The Lally group later wrote that automaticity gains soon resumed after one missed performance.
A collapse. The routine is abandoned. The term pair comes from relapse-prevention research on addiction, which separates an initial lapse from a full-blown relapse and says the step between them is not inevitable.
So a missed day resets nothing, but it starts a clock. In Buyalskaya and colleagues’ gym data, the strongest predictor of attendance was time since the last visit, pointing the same way for 76% of members. That is prediction, not proof of cause: people who stop for other reasons also show long gaps.
A caution about self-diagnosis: in two studies by Mazar and Wood (809 adults online, 112 students tracking coffee drinking), habit strength predicted behaviour as well as or better than inner states such as mood, yet people explained their behaviour by inner states. The first story about a broken routine, usually a motivation story, deserves suspicion.
What do the key studies show?
Table 1. How routines stop: ten studies
| Study | Sample and setting | Measured | Result as printed |
|---|---|---|---|
| Lally et al. 2010 (abstract only) | 96 volunteers, one daily behaviour, 12 weeks | Self-reported automaticity | Missing one opportunity did not materially affect habit formation; good curve fit for 39 of 82 analysed |
| Wood, Tam and Witt 2005 | 115 students changing university | Exercise, newspaper and TV habits, 1 month before and after | Habits survived only when aspects of the context did not change |
| Buyalskaya et al. 2023, with correction | 30,110 gym members, over 12 million check-ins; 3,124 hospital workers | Predictability of attendance and handwashing | Attendance less likely the longer since the last visit for 76%; gym plateau at 68 to 78 days (first printed as 122 to 226); handwashing after 9 to 10 shifts |
| Beshears et al. 2021 | 2,508 Google employees, 4 paid weeks | Workplace gym visits | Pay for any visit: 0.19 more visits a week (regression estimate) than pay limited to a chosen two-hour window |
| Milkman et al. 2021 | 61,293 gym members, 54 four-week programmes | Weekly gym visits | 45% of programmes raised visits by 9% to 27%; 8% still significant in the 4 weeks after; best: a bonus for returning after a missed workout, 0.40 more visits a week |
| Royer, Stehr and Sydnor 2012 | 1,000 employees of a Fortune 500 company | On-site gym attendance | First month after the incentive: 25% of the increase persisted; commitment contract taken up by 12% |
| Acland and Levy 2015 | 120 recruited, 111 analysed, university gym | Weekly visits and own predictions | 0.188 more visits a week after the paid month; 0.047 (not significant) after a 4-week break |
| Silverman and Barasch 2023 | 980 employees, 30-day step challenge; 601 online participants | Continuing after an intact or broken streak | 93.14% continued after an intact streak, 68.66% after a broken one, 85.20% when the break could be repaired |
| Stawarz, Cox and Blandford 2015 | 133 enrolled, 96 completed, 4 weeks | Adherence; self-reported automaticity (0 to 28) | Reminders 94%, 15. Event cue 76%, 20. No cue 75%, 19 |
| Oscarsson et al. 2020 | 1,066 people with New Year’s resolutions | Self-rated success after one year | 58.9% with approach goals, 47.1% with avoidance goals |
Sources: the papers named in the first column.
The Buyalskaya abstract says gym habits take months, and the text first gave 122 to 226 days. The authors’ correction four months later changed that to 68 to 78 days: the first figure was 1.8 to 2.9 times too high.
Table 2. What the same figures imply about failure
CEOtudent analysis of Table 1 sources: left column as printed, middle column our calculation.
| Figures as printed | CEOtudent calculation | Reading |
|---|---|---|
| In-window workouts a week 1.34 (window pay) against 0.95 (any-time pay); out-of-window 0.42 against 0.98 (Beshears) | The window gained 0.39 and lost 0.56: net -0.17 | A rigid slot cost more than it bought |
| Adherence 94% against 76%; automaticity 15 against 20 (Stawarz) | 18 points more adherence, 5 points (25%) less automaticity | Reminders bought repetition and cost independence |
| 93.14%, 68.66%, 85.20% (Silverman and Barasch) | A broken streak cost 24.48 points; repair recovered 16.54, or 67.6% of the gap | A repair rule removed about two thirds of the penalty |
Which nine ways do routines fail?
A systematic review of 100 behaviour theories by Kwasnicka and colleagues found five themes behind maintained change: maintenance motives, self-regulation, resources, habits and contextual influences. The taxonomy below turns them into nine failure modes that can be told apart.
Table 3. The habit failure-mode taxonomy
CEOtudent editorial framework. Grade rule: A = a meta-analysis, or two or more randomized field experiments with objectively measured behaviour. B = one randomized field experiment or large objective dataset, or a peer-reviewed review plus a prospective study. C = one small or short study, qualitative work, or a mechanism borrowed from another field. D = argument only. Grades rate how well a mechanism is documented, not how often it occurs; no opened study counts that, and CEOtudent collects no such data.
| # | Failure mode | Typical symptom | Diagnostic question | Research basis | Grade | First repair |
|---|---|---|---|---|---|---|
| 1 | Cue loss | Stopped soon after a move, new job, new timetable or home-office days | Did place, time or the preceding activity change before the first miss? | Wood, Tam and Witt; Wood and Rünger; Lally et al. 2011 | B | Re-anchor to a cue that exists in the new context |
| 2 | Borrowed cue | Ran only when an app or a person prompted it | Did it ever start without the notification? | Stawarz et al.; Renfree et al. | C | Tie the alert to an event, then test a week without it |
| 3 | Vague plan | Sincere intention, new decision every day | Can you say when and where in one if-then sentence? | Gollwitzer and Sheeran; Sheeran et al. | A | Write one if-then plan |
| 4 | Rigid plan | Miss the slot, lose the day | On missed days, was there a later moment that “did not count”? | Beshears et al. | B | Add a written fallback slot |
| 5 | Capacity mismatch | Grew longer or harder to start; or several new routines at once | Is it harder to start than when it worked, or one of many? | Kaushal and Rhodes; Acland and Levy; Renfree et al. | C | Shrink it to the version that worked; pause the rest |
| 6 | Payoff withdrawal | Faded after a challenge, incentive, course or partner ended | Was something external paying for it, and did that end? | Royer et al.; Acland and Levy; Milkman et al.; Kwasnicka et al. | A | Name a maintenance motive or renew the stake |
| 7 | Monitoring drop-off and broken streak | Logging stopped first, or a streak broke | Did the record stop before the behaviour did? | Harkin et al.; Silverman and Barasch | B | Track completion rate; add a repair rule |
| 8 | Lapse-to-collapse | One or two ordinary misses, a “ruined it” thought, then nothing | Was the first miss ordinary and the stop a decision? | Buyalskaya et al.; Milkman et al.; Larimer et al. | B | Write a lapse rule; reward the comeback |
| 9 | Framing and goal conflict | Defined as avoiding something, or always loses to another goal | Is it phrased as “stop” or “no”, or beaten by a higher priority? | Oscarsson et al.; Kwasnicka et al. | C | Rewrite it as an approach behaviour, or retire it |
That is 2 grade A, 4 grade B and 3 grade C. Modes can overlap. The order is deliberate: causes outside you come first, the story about yourself last.
How do you recognise and repair each failure mode?
1. Cue loss: the trigger stopped showing up
Recognise it. It stopped soon after a change of place or schedule. Motivation feels the same; the moment no longer arrives.
Evidence (B). In the transfer study, exercise, newspaper and TV habits survived only where the performance context did. In a qualitative study of 10 people on a habit-based weight programme, weekends and vacations temporarily disrupted performance because cues were absent; habits were reinstated on return to work.
Repair. Re-anchor, do not re-motivate. Choose a cue that occurs in the new context on every relevant day; for hybrid weeks, write one anchor per location.
Retire when the habit only made sense in the old context.
2. Borrowed cue: the reminder did the remembering
Recognise it. It stopped when a notification was muted or an app was dropped.
Evidence (C). In a 4-week study of a daily text-message task, mostly with students, SMS reminders gave the highest adherence and the lowest automaticity. Interviews with 16 users of one habit app found that reminders and streaks supported repetition but created dependency: every participant stopped using the app at least temporarily, and tracked behaviours lapsed with it. Two small studies: suggestive, not settled.
Repair. Move the alert to just before an existing daily event, then switch it off for a week and count what still runs.
Retire when it only ever ran on prompts and you would not choose it unprompted: that is a task, so schedule it.
3. Vague plan: no when, no where
Recognise it. The aim is sincere and unspecific, so each day needs remembering and deciding.
Evidence (A). If-then plans improved goal attainment by d = .65 across 94 tests with 8,461 participants; the same chapter cites a review in which a median 47% of people who intended to act did not. A 2024 synthesis of 642 tests by the same group (abstract only) reports effects from .27 to .66 depending on the outcome. One limit: Wood and Rünger note that such plans do not control many strong existing habits.
Repair. One sentence: “If [existing daily event], then I will [the first two minutes of the routine].”
Retire when no real daily event can host it.
4. Rigid plan: miss the slot, lose the day
Recognise it. Misses cluster on days when the usual slot was blocked, and nothing happened later.
Evidence (B). The Beshears experiment paid Google employees for gym visits at any time or only inside a self-chosen daily two-hour window. The window group did more in-window workouts, far fewer outside it and fewer overall. One company, paid attendance; visits were measured, not automaticity.
Repair. Keep the default slot and add the fallback in the same sentence: “If I miss 7:00, then 12:30.”
Retire when no fallback exists and you will not change the schedule.
5. Capacity mismatch: friction creep and overload
Recognise it. The routine grew, several started together, or it was planned for an ideal week.
Evidence (C). Among the 111 new gym members, consistency, low behavioural complexity, environment and affective judgments predicted habit formation. People also plan for a more capable self: Acland and Levy’s participants over-predicted their gym attendance by a factor of 2.5 to 5.5. The 16 app users were tracking 3 to 15 behaviours at once. No opened study tested friction added to an established routine.
Repair. Cut back to the last version that ran, restore the set-up that made starting easy, and pause all but one new routine. An energy audit shows where the day has room.
Retire when the smallest useful version does not fit.
6. Payoff withdrawal: the incentive, challenge or novelty ended
Recognise it. It ran during a challenge, a bet or a course and faded within weeks of the end.
Evidence (A, for gym attendance). In Royer and colleagues’ experiment ($10 a visit for one month) most of the gain was gone by the second month afterwards, while the group offered a self-funded commitment contract still differed a year later. In Acland and Levy’s study the gain did not survive a winter break, and in the megastudy far fewer programmes mattered after they ended than while they ran. Kwasnicka’s review concludes that people need at least one sustained motivator.
Repair. Before the external payoff ends, name what will sustain the routine (enjoyment, visible results, identity) and build one immediate reward in, or renew the stake on purpose.
Retire when nothing in the routine itself is rewarding and the outcome can be reached another way.
7. Monitoring drop-off and the broken streak
Recognise it. Logging stopped before the behaviour did, or a long streak broke and the tracker went unopened.
Evidence (B). Prompting people to monitor progress improved goal attainment (d+ = 0.40 across 138 randomized studies with 19,951 participants), but those studies test starting to monitor, not stopping. Step-challenge participants were more likely to hit the target after an intact streak than after a broken one, and the experiment in Table 1 reproduced the drop, amplified when people blamed themselves for the break. The experiments used short online tasks; the field result is correlational.
Repair. Replace streak length with completion rate over the last 14 opportunities and add a repair rule, such as a make-up within 48 hours.
Retire the tracker, not the habit, once the routine runs without it.
8. Lapse-to-collapse: one miss becomes a verdict
Recognise it. An ordinary miss was followed by a judgment (“I have ruined it”), then a longer gap.
Evidence (B, mechanism borrowed). The abstinence violation effect in addiction research describes people who attribute a lapse to personal failure, feel guilt and become more likely to abandon the attempt. For everyday routines the evidence is indirect: the gym gap finding, and the megastudy’s best performer, a bonus of about nine US cents for coming back after a missed workout. The authors warn that top-performer estimates are probably inflated.
Repair. Write the lapse rule in advance: log the cause in five words; the next opportunity is the restart. Make the comeback the rewarded act.
Retire when restarts keep failing for the same reason. That reason is the real mode.
9. Framing and goal conflict: the routine is against something
Recognise it. It is defined as stopping or avoiding, or it loses every contest with something you care about more.
Evidence (C). The 11.8-point gap between approach and avoidance goals in Table 1 comes from goals people chose themselves and success they rated themselves (730 participants). Kwasnicka’s review adds that motivation to avoid negative consequences is hypothesised to be insufficient for behaviour that needs sustained effort.
Repair. Rewrite it as an approach behaviour on the same cue: “At 9 p.m. the phone goes on the hallway charger and the book comes out.”
Retire when it conflicts with a higher priority. Decide that with kill criteria set in advance, not in a low moment.
Do AI assistants and trackers change how habits fail?
Probably, but this section is argument, grade D: no opened study tests AI assistants against these modes. Three modes are tool-shaped: a notification is a borrowed cue, a streak is a tracker’s default display, and a new app’s novelty wears off like a withdrawn payoff. Hybrid work adds cue churn: for a habit anchored to arriving at the office, the vacation finding suggests every home day acts like a weekend. The safer split: the assistant reads the log and names the mode; the cue stays with an event in your day.
How do you run a 10-minute habit post-mortem?
CEOtudent editorial framework. The CEO half is the post-incident review: an operations team does not answer an outage by resolving to try harder; it finds the failed component. The student half is the experiment: one change, one measure, one review date.
- Minutes 0 to 3: timeline. List the last 14 opportunities as done or missed. Mark the first miss and what changed in the fortnight before it: place, schedule, tools, people, workload.
- Minutes 3 to 8: diagnosis. Ask the nine diagnostic questions in Table 3 in order and stop at the first clear yes. If the routine never ran for a few weeks without a prompt, a payer or a daily decision, it never formed: start at questions 2, 3, 5 and 6.
- Minutes 8 to 10: one sentence. “This routine stopped because of mode X, shown by Y.” Then fill in Table 4.
Table 4. One-page repair experiment
CEOtudent editorial framework; the defaults are not tested thresholds.
| Field | What to write | Default |
|---|---|---|
| Hypothesis | “It stopped because of . If I change , it will run on at least ___ of the next 14 opportunities.” | 10 of 14 |
| One change | A single repair from Table 3 | Never two at once |
| Measure | Completion rate; did it start without a prompt? | Tally marks on paper |
| Lapse rule | What a miss triggers | Log the cause; restart at the next opportunity |
| Review date and decision | Keep, try the next mode, or retire | After 14 opportunities; two failed experiments means retire or redesign |
Put the review date in your weekly review so the experiment ends with a decision.
What this does not mean
- The evidence is narrow. Most field data cover gym attendance, diet, handwashing and step counts, in students or single companies. No opened study tested knowledge-work routines.
- Automaticity is mostly self-reported. The large datasets measure attendance and predictability instead.
- The grades are not prevalence. Any claim that a given share of habits fail for a given reason would be invented.
- The repairs are inferences. Kwasnicka and colleagues write that it is currently unclear what conditions are required to re-establish a behaviour after relapse.
- This is about everyday routines, not addiction, eating disorders or any clinical condition. This article is general information, not medical or psychological advice.
Frequently asked questions
Does missing one day ruin a habit?
No. In the Lally study it did not materially affect habit formation. The risk is the gap that follows, so restart at the next opportunity.
Are streaks good or bad?
Both. An intact streak raised continued engagement and a broken one cut it, less so when the break could be repaired. Track a completion rate and define a repair rule.
Should I rely on reminders?
For starting, yes. In one small study they gave the highest adherence and the lowest automaticity, so tie the reminder to an event and test a week without it.
Is a broken habit a willpower problem?
Rarely the first place to look. Wood and Rünger write that difficulty changing established patterns does not reflect a failure of willpower. Check cue, tools, plan and load first.
When should I give up on a habit?
When two repair experiments on different modes fail, or the smallest useful version does not fit your week. Retiring a routine by rule is a decision, not a collapse.
Sources
- Lally, P., et al. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998-1009. Abstract only.
- Gardner, B., et al. (2012). Making health habitual: the psychology of habit-formation and general practice. British Journal of General Practice, 62(605), 664-666.
- Lally, P., et al. (2011). Experiences of habit formation: a qualitative study. Psychology, Health and Medicine, 16(4), 484-489. Abstract only.
- Wood, W., et al. (2005). Changing circumstances, disrupting habits. Journal of Personality and Social Psychology, 88(6), 918-933.
- Wood, W., and Rünger, D. (2016). Psychology of habit. Annual Review of Psychology, 67, 289-314.
- Mazar, A., and Wood, W. (2022). Illusory feelings, elusive habits: People overlook habits in explanations of behavior. Psychological Science, 33(4), 563-578. Abstract only.
- Gollwitzer, P. M., and Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69-119. Scanned copy.
- Sheeran, P., et al. (2024). The when and how of planning: Meta-analysis of the scope and components of implementation intentions in 642 tests. European Review of Social Psychology, 36(1), 162-194. Abstract only.
- Beshears, J., et al. (2021). Creating exercise habits using incentives: The trade-off between flexibility and routinization. Management Science, 67(7), 4139-4171. Author manuscript.
- Buyalskaya, A., et al. (2023). What can machine learning teach us about habit formation? Evidence from exercise and hygiene. Proceedings of the National Academy of Sciences, 120(17), e2216115120; with the authors’ Correction, 120(34), e2312763120.
- Milkman, K. L., et al. (2021). Megastudies improve the impact of applied behavioural science. Nature, 600, 478-483. Author manuscript.
- Royer, H., et al. (2012). Incentives, commitments and habit formation in exercise: Evidence from a field experiment with workers at a Fortune-500 company. NBER Working Paper 18580; journal version in American Economic Journal: Applied Economics, 2015.
- Acland, D., and Levy, M. R. (2015). Naiveté, projection bias, and habit formation in gym attendance. Management Science, 61(1), 146-160. Accepted manuscript.
- Harkin, B., et al. (2016). Does monitoring goal progress promote goal attainment? A meta-analysis of the experimental evidence. Psychological Bulletin, 142(2), 198-229.
- Silverman, J., and Barasch, A. (2023). On or off track: How (broken) streaks affect consumer decisions. Journal of Consumer Research, 49(6), 1095-1117. Accepted manuscript.
- Stawarz, K., et al. (2015). Beyond self-tracking and reminders: Designing smartphone apps that support habit formation. Proceedings of CHI 2015, 2653-2662.
- Renfree, I., et al. (2016). Don’t kick the habit: The role of dependency in habit formation apps. CHI 2016 Extended Abstracts, 2932-2939.
- Kaushal, N., and Rhodes, R. E. (2015). Exercise habit formation in new gym members: a longitudinal study. Journal of Behavioral Medicine, 38(4), 652-663. Abstract only.
- Oscarsson, M., et al. (2020). A large-scale experiment on New Year’s resolutions: Approach-oriented goals are more successful than avoidance-oriented goals. PLOS ONE, 15(12), e0234097.
- Kwasnicka, D., et al. (2016). Theoretical explanations for maintenance of behaviour change: a systematic review of behaviour theories. Health Psychology Review, 10(3), 277-296.
- Larimer, M. E., et al. (1999). Relapse prevention: An overview of Marlatt’s cognitive-behavioral model. Alcohol Research and Health, 23(2), 151-160.
Table 1 reports figures as printed; Table 2 is a CEOtudent analysis of them. Tables 3 and 4, the grades, the post-mortem and every default (14 opportunities, 10 of 14, 48 hours) are CEOtudent editorial frameworks, not tested protocols. Of the 21 sources, 5 were read as abstracts and are marked; two were read through text recognition and checked against the page images. Deliberately not used: the “21 days” claim; any “X% of resolutions fail” figure (no primary source opened); any share of habits failing for a given reason; and per-day automaticity changes from the unopened full text of Lally et al. (2010).
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
