Operations & process
PDCA Cycle
An iterative four-stage improvement cycle, plan, do, check, act, for testing a change on a small scale, studying the results against a prediction, and standardising what works before running the loop again. The engine of kaizen and of most modern quality systems.
Also known as Plan-Do-Check-Act, Deming cycle, Shewhart cycle, Deming wheel, PDSA cycle. First set out by Walter A. Shewhart and W. Edwards Deming; named PDCA by Japanese managers in 1950; the primary source is cited in full below.
Where this is contested
Widely called the Deming cycle, a name Deming himself rejected. He credited Shewhart's 1939 cycle, presented his own version to Japanese managers at JUSE's invitation in 1950, and it was Japanese practitioners who reworked it into Plan-Do-Check-Act in the early 1950s. From 1986 Deming insisted on PDSA, holding that 'study' demands learning where 'check' suggests mere inspection.
- Format
- Process / loop
- Level
- Team · Business unit
- Best for
- Plan execution · Evaluate options
- Decision stage
- Plan · Execute · Review
- Difficulty
- Introductory
- Time to apply
- A first cycle can run inside two to six weeks; the method only pays off across several consecutive cycles.
Plate · The model
The components
Plan
Define the problem and the current condition with data, set a measurable target, form a hypothesis about cause, and design a change together with a prediction of its effect and how it will be measured.
Signals of strength
No baseline measurement exists before the change · The plan describes a solution with no stated problem · No prediction is written down, so any result can be claimed as intended · The scope needs months rather than weeks to test
Do
Carry out the change at small scale in the live process, collecting data as it runs and recording deviations from the plan as facts to be studied rather than failures to be hidden.
Signals of strength
The trial quietly becomes a full rollout · Data collection is left until after the trial ends · Deviations from the plan go unrecorded · The people running the process were not told what is being tested or why
Check
Study the results against the baseline and the prediction, and explain the gaps. This is the learning stage, the one Deming argued should be called study, and the one most often reduced to a glance at a dashboard.
Signals of strength
Results are eyeballed with no comparison to the baseline · Success is declared from anecdote rather than measurement · Gaps between prediction and outcome are noted but never explained · The stage is skipped entirely when results look good
Act
Close the loop: adopt the change and standardise it into how the work is done, adapt it and run another cycle, or abandon it and keep the learning. Then set up the next cycle from what this one taught.
Signals of strength
Improvements decay within months because nothing was standardised · The report is filed but standard work, training and measures stay unchanged · Failed trials are buried rather than mined for learning · No next cycle is ever framed, so improvement stops at one pass
When it earns its keep
- A process problem is recurring and you want improvement grounded in evidence rather than in the loudest opinion in the room.
- You are about to change a live operation and need a way to trial the change at small scale before it can do damage at full scale.
- An improvement has been made but nobody can say whether it worked, because nothing was measured before or after. PDCA imposes exactly that discipline.
- You are building a continuous improvement habit in a team and need a shared, teachable structure for how changes get proposed, tested and adopted.
And when it doesn't
- The situation is an emergency. A safety incident or a service outage needs containment first; PDCA is for improvement once the process is stable enough to study.
- The question is strategic direction rather than process performance. Choosing which market to enter is not a cycle you can run weekly; use strategy tools and reserve PDCA for executing the choice.
- The change cannot be trialled small or reversed, such as a one-off acquisition or an irreversible system migration. The cycle's power comes from cheap, repeatable experiments.
- You have no access to data. A cycle without measurement is plan-do-plan-do, and it institutionalises guessing.
How to run it
Before starting, gather the inputs the analysis depends on:
- A defined process with an owner, and a problem statement specific enough to be measured.
- Baseline data on current performance, since without it the check stage has nothing to check against.
- A target and an explicit prediction of what the change will do, which is what separates an experiment from an initiative.
- Authority to run a small-scale trial in the live process, and time booked to study the results before scaling.
- 1
Choose a problem small enough to cycle quickly
Scope the first cycle so it can complete in days or weeks. The method compounds through repetition, and a cycle scoped to take six months will usually be abandoned or faked. Narrow beats ambitious.
- 2
Plan with a baseline and a prediction
Describe the current condition with data, state the target, form a hypothesis about cause, and design the change with a written prediction of its effect. The prediction is the most skipped and most valuable element; without it, any outcome can be declared a success afterwards.
- 3
Do the change at small scale
Run the trial on a limited slice of the operation, one line, one depot, one shift, and record what actually happened, including deviations from the plan. The deviations are data, and hiding them poisons the check stage.
- 4
Check results against the prediction
Compare measured results with the baseline and with what was predicted, and ask why any gap exists. Deming preferred the word study for this stage precisely because the job is learning why, and a pass or fail inspection wastes the experiment.
- 5
Act: adopt, adapt or abandon
If the change worked, standardise it, write it into standard work, train it, and spread it. If it partly worked, adjust and recycle. If it failed, keep the learning and drop the change. Standardisation is what stops improvement evaporating when attention moves on.
- 6
Connect the cycles
End each cycle by framing the next one from what was learned. A single cycle produces a tweak; chained cycles produce capability, and that chaining is what Deming meant by continual improvement.
Reading the result
A tested change with evidence of its effect: baseline, prediction, measured result and the decision taken. Over successive cycles, a rising standard captured in standard work and a team that treats changes as experiments rather than edicts.
- Judge a cycle by its check stage. If the write-up contains a baseline, a prediction and an explained gap, the cycle was real; if it jumps from action to conclusion, it was an initiative wearing PDCA's clothes.
- Failed cycles that produce learning are successes of the method. A team whose cycles never fail is either not measuring or not experimenting.
- Look at the chain, not the link. One completed cycle proves little; the trajectory across several cycles is where the improvement, and the evidence of an improvement culture, actually shows.
A worked example
A Midlands parcel carrier attacks failed first-time deliveries
A regional parcel carrier running 30 depots across the Midlands has a failed first-time delivery rate of 9 per cent, each failure costing a redelivery attempt and a customer service contact. The operations director suspects vague delivery windows and missing safe-place instructions, and commissions a PDCA cycle on two depots rather than a network-wide programme.
- Plan
- Failure-code analysis across three months shows 60 per cent of failures are 'customer not in, no safe place held'. Hypothesis: customers who know the window and have lodged a safe place fail far less often. Change: a day-before SMS with a two-hour window plus a safe-place prompt at booking. Prediction, written down: failed first-time rate on trial routes falls from 9 to 6 per cent within six weeks.
- Do
- The trial runs on 40 routes across the two depots for six weeks. One deviation is recorded honestly: the safe-place prompt shipped two weeks late and sat three screens deep in the booking flow, so uptake was low from the start.
- Check
- Failed first-time rate on trial routes falls to 6.8 per cent against 9.1 per cent on control routes, short of the 6 per cent prediction. Decomposition shows the SMS window did most of the work; safe-place capture reached only 11 per cent of customers, explaining the gap. Flats and apartment blocks barely improved at all.
- Act
- Adopt the day-before SMS network-wide and write it into the standard despatch process. Adapt the safe-place capture by moving it to the first booking screen, to be retested in the next cycle. Frame cycle two around flats and apartments, where the failure mode is access rather than absence.
The read. The cycle delivered a real improvement and an honest miss. Because a prediction existed, the 6.8 per cent result forced an explanation rather than a celebration, and the explanation, buried safe-place capture, became the next cycle's plan. Scaled naively without the trial, the carrier would have credited the wrong mechanism and stalled at the first plateau.
Pitfalls
- Launching at full scale on the first cycle. This converts a cheap experiment into an expensive commitment and makes honest checking politically impossible.
- Running plan-do-plan-do. Skipping check is the commonest corruption of the cycle, and it reduces the method to serial guessing with paperwork.
- Planning without a prediction. If nobody wrote down the expected effect, the check stage becomes an exercise in retrospective justification.
- Treating act as filing the report. Until the improvement is written into standard work and trained, it will decay as soon as attention moves elsewhere.
- Stopping after one cycle. The method's compounding value is in chained cycles, and a single loop, however tidy, is a pilot rather than continuous improvement.
What the critics say
In practice the method is mostly applied badly. A systematic review of PDSA use in healthcare quality improvement found that few published applications met the method's core principles, with iterative linked cycles, small-scale testing and documented predictions frequently absent, which undermines claims made on the method's behalf.
Taylor, M. J., McNicholas, C., Nicolay, C., Darzi, A., Bell, D. and Reed, J. E. (2014) 'Systematic review of the application of the plan-do-study-act method to improve quality in healthcare', BMJ Quality and Safety, 23(4), pp. 290-298.
The cycle's simplicity is deceptive and partly illusory. Reed and Card argue that effective use demands supporting infrastructure, measurement capability and leadership behaviours the four letters conceal, and that treating PDSA as a standalone tick-box method sets teams up to fail.
Reed, J. E. and Card, A. J. (2016) 'The problem with Plan-Do-Study-Act cycles', BMJ Quality and Safety, 25(3), pp. 147-152.
The PDCA formulation itself was disowned by the man it is usually named after. Deming held that 'check' invites inspection and holding back where the stage demands study and learning, and from 1986 he taught the cycle as PDSA, calling PDCA a corruption of what he had presented in Japan.
Moen, R. D. and Norman, C. L. (2009) 'The Foundation and History of the PDSA Cycle', Associates in Process Improvement; see also Deming, W. E. (1993) The New Economics. Cambridge, MA: MIT Press.
Sources and further reading
- Shewhart, W. A. (1939) Statistical Method from the Viewpoint of Quality Control. Washington, DC: Graduate School of the Department of Agriculture.
- Deming, W. E. (1986) Out of the Crisis. Cambridge, MA: MIT Center for Advanced Engineering Study.
- Deming, W. E. (1993) The New Economics for Industry, Government, Education. Cambridge, MA: MIT Press.
- Moen, R. D. and Norman, C. L. (2009) 'The Foundation and History of the PDSA Cycle'. Paper presented at the Asian Network for Quality Conference, Tokyo. ↗