Lesson 14 of 23 · 6 min read
Every change is an experiment: the PDCA cycle
In Lean: PDCA (plan, do, check, act)
Builds on: Countermeasures, not solutions
In one minute
PDCA (plan, do, check, act) is a four-step cycle that treats a change to how work is done as an experiment: plan the change and write down what you expect, do it on a small scale, check what actually happened against what you expected, then act on the result. Until something checks it, every change is only a guess about cause and effect. The prediction is written before the change, and specific enough that a check could prove it wrong. Without one, almost every result can be made to look like a success.
The idea
Plan, try, check, then decide (in Lean: PDCA, for plan, do, check, act) is the basic cycle of improvement.
- Plan. State the problem and its cause, choose the countermeasure, write the prediction, and decide what you will look at and when.
- Do. Put the change in place as planned, on a small scale: one team, one form, one type of request. Note anything that did not go as planned.
- Check. Compare what happened with the prediction. Not "did people like it" but "did the thing we predicted happen".
- Act. If it held, make it the normal way of working (lesson 16). If it partly held, adjust and run another cycle. If it did not hold, go back to the cause with what you learned. Then the cycle starts again.
- Plan. State the problem and its cause, choose the change, and write down what you expect to see.
- Do. Make the change on a small scale: one team, one form, one kind of request.
- Check. Compare what happened with the prediction, in the work itself, at set dates.
- Act. Held: make it the standard. Partly: adjust and go again. Reverted: back to the cause.
The step people skip is writing down what you expect, before you try (in Lean: prediction, or hypothesis). It matters because hindsight is generous. After a change, nearly any result can be described as "it helped a bit". A prediction written beforehand fixes what success would look like while you are still neutral about it.
A useful prediction has four parts: what will change, how you will see it, by when, and which result would show you were wrong.
- Weak: "Onboarding will go more smoothly."
- Strong: "None of the next four new starters will arrive without system access. We will note each start on the board, and one starter without access means the plan missed something."
Use the measure you already have. If you noted how often the friction happens and how long it takes, predict in those terms. If you have no number, predict something you can observe: the step no longer happens, the question no longer arrives, the form no longer comes back. Do not invent a number to make the plan look complete, and do not mistake a prediction for a saving. "This will save an hour a week" is a forecast; only the check tells you what happened.
Change one thing at a time where you can. If you change three things and the problem goes away, you will not know which change did it, or which two you could have skipped.
Example
Illustrative example, not a real company.
In a recruitment team, job offers go out days after the candidate is chosen. The whys show that each offer waits for the hiring manager's director to approve the salary by email, case by case.
Plan: the director approves a salary range when the role is opened, so offers inside the range need no further approval. The team writes before starting: "For roles opened from next month, no offer inside the range will wait for a salary approval. We will look at every offer at the one-week and one-month checks. Any offer inside the range that waits means we missed something."
Do: the change runs for roles opened that month.
Check: no offer inside the range waited. A few offers above the range still waited, longer than before, because nobody was sure who approves exceptions.
Act: the change stays. The exceptions become the next, smaller cycle: who approves an offer above the range, and how quickly.
Common mistakes
- Writing the prediction afterwards. By then it describes what happened, not what you expected.
- Predictions that cannot fail. "Things will improve" is true of any result.
- Skipping Check. Moving to the next idea without looking means nobody learns whether the last one worked.
Words to know
- Plan, try, check, then decide (in Lean: PDCA): plan, do, check, act, repeated.
- What you expect, written before you try (in Lean: prediction or hypothesis): a statement a check could prove wrong.
- Check: comparing what happened with what you predicted.
Check your understanding
Try it on your own work
Describe one recurring task that wastes your time, and see these ideas applied to it.
Worksheet for this lesson: PDCA Template (free download, no sign-up).
Where this comes from
The plan-do-check-act cycle grew out of twentieth-century quality management and became the working rhythm of improvement in the Toyota Production System and the wider Lean tradition.