From the cycle of Walter A. Shewhart (1939) as W. Edwards Deming taught it in Japan in 1950, where Japanese executives and JUSE gave it the Plan, Do, Check, Act wording.
A four-step loop that plans a change with a predicted result, tries it small, checks the outcome against the prediction, and then adopts, adjusts, or drops it.
Two turns of plan, do, check, act on a clinic's missed appointments
Play both turns, or select a stage. Each turn starts with a prediction, tries the change at one clinic, checks the measured result against the prediction, and acts on the gap; the second turn starts from what the first one learned. Illustrative.
- The next turn starts here (Act again to Plan)
- The next turn starts here (Act again to Plan)
- Way off: rethink the cause (Check to Plan)
- Plan. One clinic misses 84 of its 600 appointments a month. The team predicts that a text two days ahead will cut that by a third, and decides to count no-shows for new and returning patients separately.
- Do. The reminder texts run at one clinic for one month. Nothing else changes, so any difference can be traced to the texts.
- Check. No-shows fell from 84 to 69, an 18 percent drop against the 33 predicted. Returning patients missed 30 percent less; new patients barely changed, and they are now over half the no-shows.
- Act. The texts stay, since they work for returning patients. The gap points at new patients, who may not know the clinic or trust a text from it, so the next turn targets them.
- Plan again. The new prediction: a phone call from the front desk the day before will halve new patients' no-shows, from 37 a month to about 18.
- Do again. For a month, the front desk calls every new patient the day before the visit. The texts continue for everyone.
- Check again. New patients' no-shows fell from 37 to 20, a 46 percent drop, close to the prediction. The clinic's total is now 52 a month, down from 84.
- Act again. Texts for everyone and calls for new patients become the standard at all five clinics, written into the front-desk checklist. The next turn starts from that standard.
An example
A group of five outpatient clinics was losing about 14 percent of its appointments to no-shows. The operations manager wanted text reminders at all five clinics by the next month. The quality lead asked for one clinic and one month first, with a written prediction: texts two days ahead would cut no-shows by a third.
They cut them by 18 percent. Because the team had decided in advance to count new and returning patients separately, the check showed why: returning patients missed 30 percent fewer appointments, while new patients barely changed and now made up more than half the misses. The second turn kept the texts and added a phone call from the front desk to every new patient the day before. The prediction was a halving of new-patient no-shows; the result was a 46 percent drop. Only then did texts plus calls become the standard at all five clinics, cutting the first clinic's no-shows from 84 a month to 52.
Why it works
Most process changes are judged by whether things got better, which almost any change can claim. Writing a prediction before the trial gives the check something to fail against: the reminders helped, but by half what was expected, and the size of that gap is what sent the team looking for the reason.
Trying the change small keeps the cost of being wrong low and the result readable, since one clinic for one month shows the effect without other changes mixed in. Deciding in advance how to measure, here by type of patient, is what turned a disappointing number into a finding. And the loop's last step closes it: a change that works gets written into the standard, so the next turn starts from a better baseline instead of drifting back.
The thinking behind it
Walter A. Shewhart Walter A. Shewhart, a physicist and statistician at Bell Telephone Laboratories, is called the grandfather of statistical quality control. He sketched the control chart in a 1924 memo to separate ordinary variation from variation with an assignable cause, and in Statistical Method from the Viewpoint of Quality Control (1939, edited by W. Edwards Deming) he drew specification, production, and inspection as a cycle. That cycle, which Deming called the Shewhart cycle, is the root of plan, do, check, act.
W. Edwards Deming W. Edwards Deming, an American statistician and quality consultant, taught Japanese managers from 1950 to see production as one system linking suppliers, producers, and customers, and he carried Shewhart's cycle into management practice. The Plan, Do, Check, Act wording was fixed by Japanese executives and the Union of Japanese Scientists and Engineers after his lectures; Deming himself kept calling it the Shewhart cycle and preferred Study to Check, restating it as Plan, Do, Study, Act in 1993.
How to run it
- Plan the change and predict the result. Write what you will change, the number you expect, and how you will measure it, before anything runs.
- Decide how to break down the result. Choose in advance the groups you will count separately, so a miss can be traced.
- Do it small. Run the change in one place for a fixed period, with nothing else changing at the same time.
- Check the result against the prediction. Compare the measured number with the predicted one and ask what explains the gap.
- Act on what you learned. Make a change that worked the new standard, revise one that half worked, and drop one that did not.
- Go around again. Start the next turn from the new standard, with a new prediction aimed at what the last check found.
Where it goes wrong
The most common failure is skipping the prediction. Without a written number, any improvement looks like success, and the team learns nothing about why the change worked or by how much. Ask to see the prediction before the trial starts.
A second failure is a trial too big to read. A change rolled out to every site at once, alongside other changes, produces a result nobody can attribute. Shrink the trial until one cause can explain the outcome.
The third is never reaching Act. Teams check, discuss, and start another trial, and the improvement never becomes the standard, so practice drifts back within months. When a change works, write it into the checklist, the system, or the job description before moving on.
When to use something else
- DMAIC. When the defect is costly, the causes are unclear, and the project needs a formal measure and analysis phase before any change.
- Toyota Kata. When the aim is to build the habit of small experiments in a team, with a coach asking the same questions daily.
- Randomized Controlled Trial. When the change can be randomized across many users and the effect has to be separated from chance.
Where it sits in SOLVE eX
Go to the source
- Statistical Method from the Viewpoint of Quality Control, Graduate School of the US Department of Agriculture; edited by W. Edwards Deming, 1939
- Out of the Crisis, MIT Center for Advanced Engineering Study, 1986