You meant to price the new menu this week. You meant to call the inspector back, order the hinges, sort the walk-in. The intention was real. What was missing was a moment: a point in the day when that job, and not the next thing in front of you, would happen.
An if-then plan supplies the moment. It has two halves: if (a situation you will meet: a time and place, a point in the shift, a thing that goes wrong) then (one action you will take). "If I've finished the lunch close and sat down with a coffee, then I open the supplier list and price the fish." Psychologists call these implementation intentions, a term Peter Gollwitzer introduced in the 1990s; they have been tested more than six hundred times since.
We touched on them in why you put things off. This piece reads the evidence properly: the first experiments, the two big meta-analyses, what happens to the effect once you correct for unpublished null results, and which parts of a plan seem to do the work. Then it turns that into plans for a kitchen, a site and a desk.
The first experiments: Christmas break, 1997
Gollwitzer and Veronika Brandstätter ran two small studies with university students in Munich around Christmas (Journal of Personality and Social Psychology, 1997).
Study 1. Students named a project they wanted to finish over the break and said whether they had made a plan for when and where to start it. Afterwards the researchers asked who had finished:
Difficult projects, no plan: 22% completed.
Difficult projects, with an if-then plan: 62% completed (68 projects in this comparison).
Easy projects, no plan: 78% completed.
Easy projects, with a plan: 84%, a difference too small to count (83 projects).
Study 2. Everyone was asked to write a report on how they spent Christmas Eve, and to send it within two days. Half were also asked to write down when and where they would write it:
With a when-and-where plan: 12 of 17 (71%) wrote it within the two days; on average 2.3 days after Christmas Eve.
Without: 6 of 19 (32%); on average 7.7 days after.
Postmarks agreed with the dates on the reports: 4.9 days against 12.6 days.
Two things are worth keeping from these early studies. The plans helped with hard jobs and barely mattered for easy ones, which people finish anyway. And the samples are tiny: Study 2 rests on 36 students, out of 86 who signed up. Results like these are where an idea starts, not where it is proven.
The two meta-analyses: d = 0.65, then d = 0.36
A meta-analysis pools many studies into one estimate. For implementation intentions there are two big ones, by the same research group, twenty years apart.
Gollwitzer and Sheeran (2006): 94 independent tests. Average effect on reaching the goal: d = 0.65, which psychologists call medium to large. This is the number most articles quote.
Sheeran, Listrom and Gollwitzer (2025, European Review of Social Psychology): 642 independent tests, most from the UK (157), Germany (148) and the USA (127). The sample-weighted average: d = 0.36 (95% interval 0.33 to 0.40).
The bigger review also checked for publication bias: small studies that found nothing tend to stay in a drawer, so the published ones look better than the truth. The authors found "substantial" bias. They corrected for it in two ways:
Trim and fill: adds the studies that seem to be missing; the estimate barely moves, to d = 0.35.
Robust Bayesian meta-analysis (RoBMA): a stricter correction; the estimate falls to d = 0.15 (0.08 to 0.22). It also found extreme evidence that the effect is real (Bayes factor 319) and that it varies a great deal from study to study.
Their own summary: the effect is robust, but small by conventional standards, and highly heterogeneous. That is the honest headline. If-then plans work; they do not work three times over, as the Christmas study suggested.
If-then plans work. The best current estimate says they work modestly, and much better for some jobs than others.
What those numbers mean in people
A d is hard to picture. One standard way to turn it into odds (Chinn, 2000) is to multiply it by 1.81 to get a log odds ratio. Here is a worked example: say that, on a normal week, 40 of 100 people start the job they had been putting off. With an if-then plan, at each estimate:
d = 0.65 (2006 review): odds ratio 3.25, so about 68 in 100 start. That is 28 extra.
d = 0.36 (2025 review, raw): odds ratio 1.92, about 56 in 100. 16 extra.
d = 0.27 (2025 review, behaviour only): odds ratio 1.63, about 52 in 100. 12 extra.
d = 0.26 (2025 review, workplace samples): odds ratio 1.60, about 52 in 100. 12 extra.
d = 0.15 (2025 review, bias-corrected): odds ratio 1.31, about 47 in 100. 7 extra.
The arithmetic for the bias-corrected line: 0.15 × 1.81 = 0.27; e^0.27 = 1.31. The starting odds are 40/60 = 0.67; 0.67 × 1.31 = 0.87; 0.87 / 1.87 = 47%.
A real field trial lands at the low end. Katherine Milkman and colleagues (PNAS, 2011) mailed 3,272 employees of a Midwestern utility the times and places of the company's flu-shot clinics. Some letters added one line, "Many people find it helpful to make a plan for getting their shot. You can write yours here", with a box for a date, or for a date and a time:
Letter only: 33.1% vaccinated (1,268 people).
Plus a box for the date: 35.6% (734 people), not a reliable difference.
Plus a box for the date and time: 37.1% (1,270 people), 4.2 points more after adjusting.
Where the clinic ran on a single day, the date-and-time box added an estimated 7.9 to 9.5 points; where it ran on several days, 1.7 to 4.3 points, not significant. A plan seems to matter most when the chance is short-lived.
Four points from one line of text in a letter costs nothing. But it is a long way from 22% to 62%.
Where the effect is big, and where it is small
The 2025 review breaks the 642 tests down by what was being changed. Some rows, with the number of tests (k):
Task completion: d = 0.72 (k = 14)
Remembering to do something later (prospective memory): 0.40 (k = 62)
Medication taken on time: 0.40 (k = 15)
Cancer screening attended: 0.43 (k = 13)
Eating more fruit and vegetables: 0.32 (k = 38)
Academic work, such as studying: 0.25 (k = 14)
Physical activity: 0.14 (k = 64)
Voting: 0.14 (k = 6), not reliably above zero
Vaccination: 0.11 (k = 5), not reliably above zero
And by setting:
Lab studies: d = 0.49 (k = 325). Field studies: 0.27 (k = 317).
Plan made in person: 0.53. On paper: 0.35. Online: 0.31.
Workplace samples: 0.26 (k = 23). University students: 0.39 (k = 363).
Outcome measured objectively: 0.43. Self-reported: 0.28.
Highly motivated vs. weakly motivated people: 0.79 (k = 8).
The pattern is clear enough to use. Plans do most for one-off actions with a clear moment (take the pill, go to the appointment, finish this task, remember to do X when Y happens). They do least for behaviour that has to repeat for months against habit and fatigue, like exercise. They work better in a lab than in a life, and better when you already want the goal. A plan does not create motivation; it turns motivation you have into a start.
Effects did not simply fade with time: tests measured at under a week averaged 0.48, at one to six months 0.19, and the twelve at six months or more 0.51. The long-run row is too small to lean on.
Which parts of the plan do the work
The same review coded what each plan said and how it was made. These are the findings a crew lead or a freelancer can act on (sample-weighted d; higher is better):
If-then form vs. a schedule. "If X, then I do Y": 0.43 (k = 373). A plain schedule ("I'll do Y on Tuesday at 3"): 0.29 (k = 269). The if matters.
Time only: 0.25. Time and place: 0.46. Time, place, and how or how long: 0.24. Adding the detail of the work almost halved the effect. Decide what the first action is before you write the plan, then keep the plan short.
A point in the task as the cue (starting a phase, finishing one): 0.50 (k = 32). "When I finish the walk-through, then..." is a strong trigger.
An obstacle as the cue (facing a setback): 0.31; realising you are distracted: 0.71 (only 2 tests).
One plan: 0.41 (k = 478). Three plans: 0.07. Five: 0.10. The authors are unsure why and say the number of plans may be mixed up with the type of goal; still, one plan per goal is the safe choice.
Said or rehearsed once: 0.50, against 0.33 for no rehearsal. Repeating it more often added nothing.
Recording the plan (keeping a copy to look at again): 0.31, against 0.44 without. The authors think going back over the plan turns it into something to deliberate about, which undoes the automatic link between cue and action.
People picking their own cue, with no guidance: 0.16 (k = 20).
That recording result is awkward for anyone who sells a task app, so to be plain about it: the studies did not test plans written on a task list. Our reading is that the plan should be written once, read back once, and then left alone, not reworded every morning. The next section is built that way.
If-then plans for a shift or a site
A good plan names a moment you will certainly meet, and one small first action. Some examples, written to the rules above:
In a kitchen
"If the lunch rush is down to the last two tickets, then I start the fish order for Friday."
"If the walk-in reads over 41°F, then I move the dairy to the reach-in and call the number on the door." (An obstacle plan; the same kind of line we suggested in shift handovers.)
"If I clock in for a close, then the first thing I do is check the closing list for anything marked from the morning."
On a site
"If the concrete truck is booked and the forms aren't signed off by 2 pm the day before, then I call the inspector before I leave."
"If I finish the morning walk-through, then I write the three things that must happen today on the board."
"If the client asks for anything extra, then I say I'll send the price in writing tonight, and I add the task before I get back in the van."
At a desk, for an owner or a freelancer
"If it's Monday and I've opened the laptop, then the first thing I open is the unpaid invoices list."
"If I notice I'm on my phone instead of the quote, then I write one line of the quote." (A distraction cue: few tests, but large effects.)
Each has one cue you can't miss, one action small enough to start in two minutes, and no detail about how long it will take.
Turning a plan into a task trigger
Five steps, taking about three minutes per job you keep putting off:
1. Pick the job you actually want done. Plans did most where motivation was high (0.79). If you don't really mean to do it, delete it or hand it on.
2. Decide the first physical action. Not "sort the rota" but "open last month's rota and copy it". Do this before you write the plan, not inside it.
3. Choose one cue you will certainly meet. A time and place ("Tuesday after the lunch close, at the pass"), or a point in the shift ("when I finish the walk-through"). Avoid cues that might not happen.
4. Write it once, as "If ..., then I ...". One plan per job. In tick-tock-done, put the then in the task title ("Copy last month's rota"), the day and time on the task, and the if as the first line of its note. On a daily checklist, put the item right after the step that is its cue, so the list itself is the trigger (see opening and closing checklists).
5. Say it back once, then leave it. Read it aloud, or say it to the person you work with; in a crew, the pre-shift huddle is the natural place. Then stop editing it. If the cue passes and nothing happened, write a new plan with a different cue rather than rewording the old one.
Our free two-minute start page does steps 2 to 4 for you: type the task, pick its kind, and copy three two-minute first steps and a "when this happens, I will" line. No sign-in.
What the research does not show
It doesn't show a large effect in everyday work. The headline d = 0.65 comes from an older, smaller and more biased set of studies. The best current estimate after correcting for bias is d = 0.15; field studies average 0.27 before correction, and the 23 workplace tests 0.26.
It doesn't show that a plan beats habit over months. Physical activity, the most studied long-term behaviour, shows d = 0.14.
It doesn't test task apps. None of the 642 tests, as far as the review reports, used a to-do list. The finding that kept copies weakened plans comes from studies that were not designed to test that, and is a correlation across studies, not an experiment.
It doesn't show what happens in crews. Nearly all tests are of individuals planning their own actions, often students in a lab. Plans made by a lead for someone else, or a crew's shared plan, are a different thing and have far less evidence.
Most effects are small and vary a lot. The bias-corrected spread between studies (tau = 0.27) is larger than the average effect. Some plans in some settings do much better than 0.15; some do nothing.
Write tomorrow's hardest job as a task with a time, its first step in the title and its if-then cue in the note.
Open tick-tock-doneSources
Gollwitzer, P. M. & Brandstätter, V. (1997). Implementation intentions and effective goal pursuit. Journal of Personality and Social Psychology 73(1), 186-199. PDF at Stanford SPARQ
Gollwitzer, P. M. & Sheeran, P. (2006). Implementation intentions and goal achievement: a meta-analysis of effects and processes. Advances in Experimental Social Psychology 38, 69-119. KOPS Konstanz
Sheeran, P., Listrom, O. & Gollwitzer, P. M. (2025). 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. doi:10.1080/10463283.2024.2334563; accepted manuscript at KOPS Konstanz. Tables 1 to 4 and the publication-bias section are the source of every d in this article after 2006.
Milkman, K. L., Beshears, J., Choi, J. J., Laibson, D. & Madrian, B. C. (2011). Using implementation intentions prompts to enhance influenza vaccination rates. PNAS 108(26), 10415-10420. PMC3127912
Chinn, S. (2000). A simple method for converting an odds ratio to effect size for use in meta-analysis. Statistics in Medicine 19(22), 3127-3131. PubMed 11113947 (the d × 1.81 conversion used for the worked example).
Cover: clipart from Openclipart, released under CC0 (public domain).