Method guide · Kaizen and continuous improvement
PDSA cycle: plan, do, study, act, and how it differs from PDCA
Short answer
PDSA stands for plan, do, study, act. It is W. Edwards Deming's name for the improvement cycle: predict what a change will do, try it on a small scale, study the result against the prediction, then adopt, adjust or abandon the change.
PDSA is a four-step cycle for testing one change: plan the change and write down what you predict it will do, do it on a small scale, study the data against the prediction, and act by adopting the change, adjusting it and running the cycle again, or abandoning it. The steps match PDCA. The difference is in the third step: check asks whether the plan worked, study asks what you learned from the gap between the prediction and the result. Deming named it PDSA in 1993 and disliked the PDCA name.
Last reviewed:
What does PDSA stand for?
PDSA stands for plan, do, study, act. It is a cycle for learning whether a change is an improvement: you plan a test of the change and predict the result, carry out the test, study what happened against the prediction, and act on what you learned. Then you go round again with a better theory, a bigger test or a different change.
The cycle is also called the Deming cycle, the Shewhart cycle for learning and improvement (Deming's own name for it), and in most plants simply PDCA. In healthcare and education it is nearly always written PDSA, because the Institute for Healthcare Improvement built its Model for Improvement on it.
One cycle tests one change. A real improvement usually takes several cycles in a row: the first on one machine or one shift, the next under different conditions, the last to put the change into the standard. A single cycle that ends in a rollout has skipped the learning.
Plan, do, study, act: the four steps and what to write down
The cycle is only as good as what is written down before the test. The table gives each step, the questions it answers and what should be on the sheet.
Plan, do, study, act: the question each step answers
- 1PlanWhat are we trying to learn, what change will we test, and what do we predict will happen?
- 2DoDid the test run as planned?
- 3StudyWhat happened, compared with the baseline, the prediction and the target, and what does the gap teach us?
- 4ActAdopt, adjust or abandon?
| Step | What it answers | What to write down |
|---|---|---|
| Plan | What are we trying to learn, what change will we test, and what do we predict will happen? | The problem and its baseline figure, the suspected cause, the change, the prediction with a number, where and for how long the test runs, who does what, and how the data will be collected |
| Do | Did the test run as planned? | What was actually done, and anything that went differently from the plan: a shift missed, a part that did not arrive, an operator who did it another way |
| Study | What happened, compared with the baseline, the prediction and the target, and what does the gap teach us? | The result, the comparison with all three, the side effects, and what the team now believes about the cause that it did not believe before |
| Act | Adopt, adjust or abandon? | The decision and the reason, what changes in the standard if adopted, and the plan and prediction for the next cycle if not |
The three decisions in act are Deming's own, from The New Economics (1993): adopt the change, abandon it, or run through the cycle again.
The prediction is the part most teams skip, and it is the part that makes PDSA work. Without a number written down before the test, any result can be read as a success. With one, a result that meets the target but misses the prediction tells you the theory behind the change was incomplete, which is worth knowing before you roll it out to the whole plant.
Keep the first test small: one machine, one line, one shift, one or two weeks. Deming's 1986 version of the cycle says the change should be carried out preferably on a small scale. Small tests are cheap to undo, and the people running them do not need everyone's agreement first. Agreement matters for implementation, not for a test.
PDSA vs PDCA: what is the difference?
The steps line up one for one, and a plant that runs PDCA well is doing PDSA. The difference is the third step and what it is for.
| PDCA | PDSA | |
|---|---|---|
| Stands for | Plan, do, check, act | Plan, do, study, act |
| Third step asks | Did the plan succeed or fail against the target? | What happened against the prediction, and what did we learn? |
| Prediction written before the test | Not required, and often missing | Required: it is what study compares the data with |
| Where it came from | Japanese quality control in the 1950s, recast from Deming's 1950 wheel | Deming, 1986 and 1993, building on Shewhart's 1939 cycle |
| Mostly used for | Implementing and standardizing a known fix, and keeping to the standard | Testing and learning before a fix is standardized |
| Who says it | Manufacturing, lean and ISO management systems | Healthcare, education, the Deming community |
The history and the two uses follow Moen and Norman's account in Quality Progress (2010). They conclude that the two cycles are related only through the scientific method.
Deming did not accept PDCA as his. Writing to Ronald Moen in 1990, he asked for the cycle to be called PDSA and not PDCA, and in his seminars he warned that check means to hold back. The Deming Institute puts the practical difference this way: check looks at whether a plan succeeded or failed, study compares the result with a prediction so the theory behind the change can be revised.
For a plant the advice is simple. Keep calling it PDCA if that is the word your people know. Add the prediction to the plan and compare the result with it in check. The six cycles in our PDCA examples all do this, and one of them met its target while missing its prediction, which is exactly the case where check and study give different answers.
Deming's PDSA cycle: where it comes from
The cycle is older than the acronym. The dates and books below follow Moen and Norman's history, which is the most careful account of who added what.
| Year | Who | What changed | Where it is written |
|---|---|---|---|
| 1939 | Walter Shewhart | Turned specification, production and inspection from a straight line into a circle, and compared the three steps to forming a hypothesis, running an experiment and testing the hypothesis | Statistical Method from the Viewpoint of Quality Control (edited by Deming) |
| 1950 | W. Edwards Deming | Presented a four-step wheel to Japanese engineers: design the product, make it, sell it, test it in service and find out what users think, then go round again | Deming's lectures to the Union of Japanese Scientists and Engineers (JUSE) |
| 1951 | Japanese quality control | Recast the wheel as plan, do, check, act; act meant standardizing what worked or going back to plan. Kaoru Ishikawa later added goals and methods to plan and training to do | Masaaki Imai, Kaizen (1986), on the recasting; Ishikawa, What is Total Quality Control? (1985) |
| 1986 | Deming | Taught the Shewhart cycle in four steps: plan a change or test, carry it out preferably on a small scale, study the results, act on what was learned | Out of the Crisis |
| 1991 | Moen, Nolan and Provost | Made a written prediction part of plan, and had the study step compare the data with it | Improving Quality Through Planned Experimentation |
| 1993 | Deming | Named it the Shewhart cycle for learning and improvement, PDSA: plan, do, study, act. Act means adopt the change, abandon it, or run through the cycle again | The New Economics |
| 1994 and 1996 | Langley, Nolan, Nolan, Moen, Norman and Provost | Put three questions in front of the cycle: what are we trying to accomplish, how will we know a change is an improvement, what change can we make. This is the Model for Improvement | Quality Progress (1994) and The Improvement Guide (1996) |
Sources: Moen and Norman, Circling Back, Quality Progress, November 2010; Moen, Foundation and History of the PDSA Cycle (Associates in Process Improvement, 2009); the W. Edwards Deming Institute; the Lean Enterprise Institute Lean Lexicon.
Two things follow from the history. PDCA came first and came from Japan, so it is not a corruption of PDSA in time, only in Deming's view of its purpose. And the prediction step is not Deming's alone: it was added in 1991 by his colleagues and then built into his 1993 cycle, which is why older descriptions of PDSA do not mention it and newer ones treat it as essential.
A worked PDSA example from the shop floor
This example is illustrative, written from the method with realistic numbers. It shows why study is a different act from check. An assembly line stops several times a shift because a part is missing at the station. The team runs one PDSA cycle on two stations.
Plan
Problem: missing-part stops on assembly line 2, day shift. Baseline, five shifts from the stop log: 9, 11, 8, 12 and 10 stops, 50 in total, so 10.0 stops a shift. At the log's average of 4 minutes a stop that is 40 minutes of lost line time a shift.
| Reason from the stop log | Stops in 5 shifts | Stops a shift |
|---|---|---|
| Bin empty, one of the 6 fast-moving parts | 24 | 4.8 |
| Bin empty, any other part | 7 | 1.4 |
| Wrong part in the bin | 9 | 1.8 |
| Part damaged | 6 | 1.2 |
| Other | 4 | 0.8 |
| Total | 50 | 10.0 |
Suspected cause: the material handler refills bins on a fixed hourly round, and the six fast-moving parts run out between rounds. Change to test: a two-bin kanban for those six parts at stations 4 and 5, so an empty bin is the signal to refill and the handler no longer waits for the round. Target, set by the supervisor: fewer than 5 stops a shift. Prediction, written before the test: the 24 empty-bin stops on the six parts go to almost none, and nothing else changes, so stops fall to about (50 minus 24) divided by 5, which is 5.2 a shift, or about 21 minutes. Test: two weeks, ten day shifts, stations 4 and 5 only. Measure: the same stop log, with the reason recorded for every stop.
Do
The two-bin racks went in over a weekend and the handler was shown the new rule on Monday. Everything ran as planned, with one note: because every fast-moving part now had two bins, the number of bins on the two stations doubled, and two of the six parts are connectors that look alike.
Study
Stops a shift over the ten shifts: 6, 5, 7, 5, 6, 4, 6, 5, 5 and 6, which is 55 stops, or 5.5 a shift, about 22 minutes at 4 minutes a stop. Against the baseline: down from 10.0 to 5.5. Against the prediction of 5.2: close. Against the target of fewer than 5: missed. A check that stops here says the test failed. The study looks at the reasons.
| Reason | Baseline, stops a shift | Trial, stops in 10 shifts | Trial, stops a shift |
|---|---|---|---|
| Bin empty, one of the 6 fast-moving parts | 4.8 | 2 | 0.2 |
| Bin empty, any other part | 1.4 | 14 | 1.4 |
| Wrong part in the bin | 1.8 | 26 | 2.6 |
| Part damaged | 1.2 | 10 | 1.0 |
| Other | 0.8 | 3 | 0.3 |
| Total | 10.0 | 55 | 5.5 |
Every figure in both tables can be recomputed from the raw counts: the test file for this page does so.
Missing-part stops a shift, baseline against trial
Example numbers- Empty: fast mover: 4.8 a shift at baseline, 0.2 in the trial
- Empty: any other part: 1.4 a shift at baseline, 1.4 in the trial
- Wrong part in the bin: 1.8 a shift at baseline, 2.6 in the trial
- Part damaged: 1.2 a shift at baseline, 1 in the trial
- Other: 0.8 a shift at baseline, 0.3 in the trial
Stops a shift by reason, baseline then trial: fast-moving bins empty 4.8 to 0.2, other bins empty 1.4 to 1.4, wrong part in the bin 1.8 to 2.6, part damaged 1.2 to 1.0, other 0.8 to 0.3.
What the study found. The prediction about the cause held: empty-bin stops on the six parts fell from 4.8 to 0.2 a shift, so the hourly round was the cause. But wrong-part stops rose from 1.8 to 2.6 a shift. The change had a side effect: twice as many bins at the station, two of them holding connectors that look the same, and the handler put the wrong connector in the wrong bin several times. The target was missed because of a problem the change created, not because the theory was wrong.
Act
Adjust and run again. Keep the two-bin system. Add a photo and a colour band to every bin label so the two connectors cannot be confused, and brief the handler on the second shift as well. Prediction for the next cycle: wrong-part stops back to about 1.8 a shift or lower, everything else as in the trial, so about 0.2 plus 1.4 plus 1.8 plus 1.0 plus 0.3, which is 4.7 a shift and under the target. Only when that holds does the change go into the handler's standard work and spread to the other stations.
Compare this with a pass or fail check. A check would have recorded 5.5 against a target of 5 and either dropped the two-bin idea or lowered the target. The study kept the part of the change that worked, found the side effect, and set up a second cycle that is likely to pass. Our PDCA template has a line for the prediction and a check against the goal; six more full cycles, with their run charts, are in the PDCA examples.
PDSA in healthcare and the Model for Improvement
Most of what you will read about PDSA online comes from healthcare, because the Institute for Healthcare Improvement (IHI) made the cycle the engine of its Model for Improvement: three questions (what are we trying to accomplish, how will we know a change is an improvement, what change can we make that will result in improvement) followed by PDSA cycles to test the changes. The Agency for Healthcare Research and Quality (AHRQ) and many health systems publish PDSA worksheets built on the same model.
IHI's advice on running the cycles reads well on a shop floor too: test on a small scale first, with willing volunteers, and widen the test under different conditions before implementing; plan the next cycle or two while designing this one; ask what can be tested by next Tuesday; and reflect on every result, including the failures, by asking what was expected, what happened and what to do next.
Healthcare has also measured how well the cycle is used in practice. A systematic review of 73 published PDSA studies by Taylor and colleagues (BMJ Quality and Safety, 2014) found that fewer than 20 percent fully documented a sequence of iterative cycles, and only 15 percent of those with enough detail reported data at monthly or more frequent intervals to guide the cycles. The lesson is the same in a plant: one cycle, no prediction and an annual figure is not PDSA, whatever it is called.
FOCUS-PDSA (or FOCUS-PDCA) is a hospital variant from the 1980s that adds five steps in front of the cycle: find a process to improve, organize a team, clarify the current process, understand the variation, select the change. It does the same job as the plan step of an A3 or a kaizen event charter: make sure the team is working on the right process with data before it tests anything.
A PDSA template: what belongs on the sheet
A PDSA template is a one-page form with a box for each step. Any PDCA form works if it has a line for the prediction. The free PDCA template is a one-page PDF and an Excel cycle sheet that records the baseline, the goal and the result of each cycle, and a log for numbering cycles as you go round again.
- Header: the problem in one line, the owner, the team, the start date and the cycle number, starting at 1.
- Plan: the baseline figure, the suspected cause, the change, the prediction with a number, the scope (which machine, line or shift), the dates and the data to collect.
- Do: what was actually done and what went differently from the plan.
- Study: the result against the baseline, the prediction and the target; side effects; what was learned.
- Act: adopt, adjust or abandon, with the reason; what changes in the standard; the plan and prediction for the next cycle.
For a bigger problem, the same cycle sits inside other formats. An A3 report is PDSA laid out on one page with the background and current condition in front of it. A kaizen event runs several fast cycles in a week. If you are not sure which frame fits the problem, the lean method navigator asks six questions and suggests one, and problem solving methods compares them side by side.
Common mistakes with PDSA cycles
- No prediction. Without a number written before the test, the study step has nothing to compare with and every result looks like a success. Write the prediction and the reasoning behind it.
- Testing on the whole plant. If the first test covers every line, it is an implementation, it is expensive to undo, and the side effects reach everyone. One machine, one line or one shift is enough to learn from.
- Lowering the target to match the result. A missed target is information. Find out why it was missed, as the worked example does, and run again.
- Stopping after one cycle. Most changes need two or three cycles: one to test the idea, one under different conditions (the other shift, the other product), one to put it into the standard.
- Studying only the headline number. The total hides the mechanism. Break the result down by reason, station or shift, and look for side effects as well as the intended effect.
- Treating a prediction as a target. The target is what the business needs. The prediction is what this change should achieve and why. They can differ, and the gap between them is where the learning is.
- No next cycle named. Act should end with a decision and the plan for the next cycle, or a note that the change is adopted and the standard updated. A sheet with an empty act box was a check, not a study.
- Not spreading what worked. Once a change holds under varied conditions, write it into standard work and share it with the other lines, which is yokoten. Otherwise the next line runs the same cycle from scratch.
Continuous Improvement Software
Run PDSA cycles in LeanSuite
Use A3, DMAIC, PDCA, PDSA or just-do-it templates, or define your own method and phases.
Go deeper
This page is the short version. These guides cover each part in full.
- PDCA Cycle: Plan, Do, Check, Act Steps ExplainedThe PDCA steps in full, where the cycle fits in lean, and how plants use it day to day.
- What is DMAIC and PDCA in Lean Manufacturing?When a problem needs the full DMAIC project instead of a short cycle, and how the two relate.
- A3 Problem Solving: The A3 Report and the 10-Step ProcessThe one-page A3 format that wraps a PDSA cycle in the background, current condition and follow-up.
- Yokoten: What Horizontal Deployment Means and How to Do ItHow to spread a change that passed its cycles to the other lines and plants.
- Coaching Kata: Toyota's Leadership RoutineA daily routine built on the same predict, test and study loop, with a coach asking the questions.
All our articles on this subject are in the kaizen and continuous improvement topic.
Free templates and tools
- PDCA templateOne improvement cycle on one page: plan, do, check against the goal, then standardize or adjust. PDF and Excel.
- A3 problem solving templateThe standard seven-box A3 from background to follow-up, with a filled-in worked example. PDF and Excel.
- Quick kaizen templateOne page for a small improvement: before and after, the cause, the countermeasure, the results and whether the standard changed.
- 5 Whys templateA problem statement, a why chain that can branch, evidence for every answer, root causes and countermeasures that are checked.
- Coaching kata template (improvement kata storyboard)The learner's storyboard (target condition, current condition, obstacles, PDCA cycles) and a fold-over card with the coach's five questions.
- Lean method navigatorSix questions about one problem, then the lean method to start with, the two that follow, why each fits, five steps for Monday and what it will not solve.
- Kaizen savings calculatorYearly savings, payback in months and first-year ROI of an improvement, from time saved, scrap avoided and its costs.
Doing this in LeanSuite
Kaizen & Projects
Manage every improvement project in one visual hub, from a small kaizen event to a full DMAIC project, and track each one through to results.
See Kaizen & ProjectsLean Studio
LeanSuite's whiteboard for lean teams, with 30 ready-made boards including the fishbone, 5 Whys, A3 and 8D.
See Lean StudioYokoten software
Captures local improvement wins and turns them into standards across every plant in your network.
See Yokoten software
See it on your own lines
Bring the problem you are working on to a 45-minute demo. Or look around with sample data first, or pilot LeanSuite on your own lines for 90 days, with the setup done by our team.
FAQ
Questions about pDSA cycle
PDSA stands for plan, do, study, act. It is a four-step cycle for testing a change: plan the test and predict the result, do the test on a small scale, study the data against the prediction, and act by adopting the change, adjusting it and running again, or abandoning it.
The steps are the same. In PDCA the third step, check, asks whether the plan succeeded or failed against the target. In PDSA the third step, study, compares the result with a prediction written before the test, to learn why the change did or did not work. Deming named the cycle PDSA in 1993 and disliked the PDCA name. PDCA is mostly used to implement and standardize, PDSA to test and learn.
W. Edwards Deming, building on Walter Shewhart's 1939 cycle of specification, production and inspection. Deming taught a four-step version in Out of the Crisis (1986) and named it plan, do, study, act in The New Economics (1993). Moen, Nolan and Provost added the written prediction in 1991, and the Model for Improvement put three questions in front of the cycle in 1994.
Plan: define the problem, the change, the prediction and how the test will run. Do: carry out the test on a small scale and note anything that went differently. Study: compare the result with the baseline, the prediction and the target, and look for side effects. Act: adopt the change into the standard, adjust it and run another cycle, or abandon it.
The change is tried in the do stage, but only as a small test. It is implemented in the act stage, after the study stage has shown that it works and under what conditions. Implementation usually follows two or three cycles, not one, and ends with the change written into standard work.
An assembly line has 10 missing-part stops a shift, most because fast-moving parts run out between the material handler's hourly rounds. The team puts six parts on two-bin kanban at two stations and predicts 5.2 stops a shift. The two-week trial gives 5.5: empty-bin stops on those parts almost vanish, but wrong-part stops rise because the doubled bins hold two look-alike connectors. The team keeps the two bins, adds photo labels and runs a second cycle. This example is illustrative.
A one-page form with a box for each step: the problem and baseline, the change and the prediction, what was done, the result against the baseline, the prediction and the target, and the decision to adopt, adjust or abandon with the plan for the next cycle. Any PDCA template works if it has a line for the prediction.
A hospital variant from the 1980s that puts five steps in front of the cycle: find a process to improve, organize a team, clarify the current process, understand the variation and select the change. It makes sure the team has the right process and the data before it tests anything, much as the background and current condition sections of an A3 do.
Usually two or three for one change: the first to test the idea on a small scale, the next to test it under different conditions such as the other shift or another product, and the last to put it into the standard. Each cycle should have its own prediction and its own data. A single cycle followed by a rollout skips the learning.
As the testing engine of the Institute for Healthcare Improvement's Model for Improvement: a team answers three questions about its aim, its measures and its change ideas, then runs PDSA cycles to test each idea on a small scale before implementing it. A 2014 review in BMJ Quality and Safety by Taylor and colleagues found that most published healthcare PDSA studies did not document iterative cycles or frequent data, so the method is often named more than it is followed.
Related guides and tools
- Six worked PDCA cycles from the shop floor, each with a prediction, a small-scale test and the real check
- PDCA template: one-page PDF and Excel cycle sheet with a line for the prediction and the goal
- PDCA cycle explained: the plan, do, check, act steps in full
- Problem solving methods compared: when to use PDCA, A3, 8D or DMAIC
- Printable PDCA cycle poster for the line-side board
- Toyota Kata: the daily improvement routine built on the same predict and study loop
More method guides
- The 6 Ms of production
- Poka-yoke
- Model area 90-day plans
- Moving from Excel to one system
- Reducing machine downtime: a worked month
- Andon
- Cycle time vs lead time, worked through
- Root cause analysis
- How many operators a line needs
- Reducing scrap and rework: a worked month
- Unsafe acts and unsafe conditions
- Just culture
- Stop-work authority
- Material review board (MRB)
- Nonconformance and NCRs
- Problem solving methods
- Quad chart
- OEE calculation
- Quality assurance vs quality control
- SOP meaning and format
- Job hazard analysis
- Near miss reporting
- Fishbone diagram
- Pareto chart
- Histogram
- Control chart
- What is a gemba walk
- What is lead time
- What is operational excellence
- How to improve product quality
- Preventive vs predictive maintenance
- Kanban
- FMEA
- Safety moments
- Just-in-time manufacturing
- Kaizen
- 5S
- Total quality management (TQM)
- CAPA (corrective and preventive action)
- Layered process audit (LPA)
- Lean tools
- Lockout tagout (LOTO)
- A3 problem solving
- Preventive maintenance
- APQP
- High reliability organization (HRO)
- Manufacturing KPIs
- Quality metrics
- Practical problem solving
- CMMS
- Risk assessment
- Production efficiency
- Total productive maintenance (TPM)
- Reliability-centered maintenance (RCM)
Or browse the topics, the full library and the lean glossary.
Rather see it on your own problem?
Bring the problem you are stuck on, and we will show you on a call how LeanSuite handles it.