Lean teams use jidoka to build quality in at the source instead of inspecting it in later. Stopping feels costly in the moment but prevents batches of bad parts. Each stop also becomes a chance to learn and improve the process.
Where it comes from
Jidoka traces back to Sakichi Toyoda's automatic looms, which stopped themselves when a thread broke, so one weaver could tend many looms without making bad cloth. His Type G automatic loom was completed in 1924. Jidoka became one of the two pillars of the Toyota Production System, alongside just-in-time.
At a glance
Detect, stop, signal, fix
Swipe sideways to see the whole diagram.
How to apply Jidoka
- 1
Detect the abnormality
Use sensors, error-proofing devices or clear visual standards so abnormal conditions are noticed immediately. Detection should not depend on someone remembering to check.
- 2
Stop the process
Have the machine stop automatically, or give the operator authority to stop. The goal is that no defect passes to the next step.
- 3
Signal for help
Link the stop to an andon so a leader responds quickly. Make the location and type of problem visible.
- 4
Fix the immediate problem
Contain any suspect parts and restore normal conditions. Restart only once the process is producing good work again.
- 5
Find and fix the root cause
Investigate why the abnormality happened using tools like five whys. Add a countermeasure, often a poka-yoke, so it does not recur.
Worked example (illustrative)
A bottling station adds a sensor that stops the capper when a cap is missing or crooked. The stop lights an andon, and the team leader clears the jam and quarantines the last 20 bottles for a check. The capper stops about 9 times in the first week, nearly all from one point in the cap chute. The team adjusts the chute, and stops fall to 1 or 2 a week. Before the sensor, crooked caps were found only at the end of the line, often after a full case had been packed.
Common mistakes
- Letting the line keep running with known defects to protect output numbers.
- Restarting after the quick fix without investigating the root cause.
- Relying on end-of-line inspection instead of detection at each step.
When Jidoka is not the right tool
Some processes, such as continuous chemical flows, cannot simply stop. There, jidoka means detecting the problem early and containing it, then fixing the cause.
Stopping also needs a response. A stop without fast help only lowers output.
Jidoka meaning
Jidoka means stopping a process at the first sign of an abnormality, so defects do not move downstream. It is often translated as automation with a human touch.
Jidoka and just-in-time are the two pillars of the Toyota Production System.
Autonomation
Autonomation is the English word often used for jidoka. A machine detects a problem and stops itself, so one person can watch several machines.
The human part matters as much as the machine part. People also stop the line, often with an andon cord or button, when they see a problem.
The first minute after a stop: two examples
These examples are illustrative. A press stops because a sensor finds no part in the die. The light shows yellow, the team leader arrives within about a minute, removes the misfed blank and checks the last few parts. If the cause is not clear, the press stays stopped while the team asks why the feeder misfed.
A build pipeline stops a release because 2 of 1,400 automated tests fail. Nothing ships. The engineer who made the last change is alerted, reproduces the failure and either fixes it or reverts the change. The team then asks why the defect was not caught earlier, such as a missing test at the point of the change.
Machine jidoka and human jidoka
Toyota describes two forms, and neither needs advanced automation or AI:
Machine jidoka
The equipment checks its own work and stops when a condition is wrong, such as a loom stopping on a broken thread or a press stopping on a misfed part. One operator can then look after several machines instead of watching one.
Human jidoka
Any operator who sees something wrong can stop the line, usually by pulling an andon cord or pressing a button. The duty to spot problems sits with the people doing the work, not only with an inspection team.
Most real systems combine both. The useful question is which checks should stay with people, which can be built into the equipment, and how a missed check would become visible.
The five links in a jidoka response
A sensor alone is not jidoka. The response works only when every link is in place:
- 1
A clear normal
A written or visual standard for good work.
Without it, nobody can say what counts as abnormal.
- 2
Detection
The machine or the operator notices the change, written so it can be seen: the bracket is missing, not poor assembly quality.
Without it, defects surface at final inspection or at the customer.
- 3
Stop or hold
Suspect work stops moving.
Without it, the alarm sounds while bad parts keep flowing.
- 4
A call for help
The problem reaches someone who can act.
Without it, work sits stopped and output falls.
- 5
Fix and learn
The immediate condition is restored and the cause is studied.
Without it, the same stop comes back next week.
Setting the stop boundary
A stop does not always mean every machine stops. Decide how far the hold reaches for each kind of problem:
| Stop boundary | Fits | Watch out for |
|---|---|---|
| Hold one item at its station | A part not seated in a fixture | Work piling up behind it if the fix takes long |
| Block the next machine cycle | A broken or missing tool | Parts made since the last good check |
| Stop the line at a fixed position | Manual assembly lines with an andon cord, where the line stops at the end of the cycle if the problem isn't fixed in time | Leaders who can't reach the call before the cycle ends |
| Stop linked processes together | A fault that affects every unit downstream | The cost of a wide stop, which needs engineering review |
Hold one item at its station
A part not seated in a fixture
Watch out for: Work piling up behind it if the fix takes long
Block the next machine cycle
A broken or missing tool
Watch out for: Parts made since the last good check
Stop the line at a fixed position
Manual assembly lines with an andon cord, where the line stops at the end of the cycle if the problem isn't fixed in time
Watch out for: Leaders who can't reach the call before the cycle ends
Stop linked processes together
A fault that affects every unit downstream
Watch out for: The cost of a wide stop, which needs engineering review
Also decide what happens to work made since the last confirmed good check. The response should cover all of it, not only the item that set off the alarm.
Restart rules
Write these before the first stop, while nobody is under pressure to run:
- What to check: The specific conditions that show the problem is fixed, such as the fixture clear and a test part seated correctly.
- Who approves restart: A named role, not whoever is nearest.
- First-piece check: How the first parts after restart are verified.
- Suspect output: Who decides what happens to held work, and where that decision is recorded.
- Temporary fixes: Each one gets an owner, a scope and a review date, so a workaround doesn't quietly become the new standard.
Restarting fixes today's condition. The study of why it happened stays open until a countermeasure is tested, often with five whys.
Jidoka examples by process
Four illustrative cases. Each shows detection, containment, help and learning:
Fastening
A nutrunner checks torque against the approved range. Out of range, the station holds the assembly and calls for help. The responder checks the joint, holds the unit and studies whether the cause is the tool, the part or the method.
Machining
A probe finds a broken tool before the next cycle starts. The machine stops, and the team checks every part made since the last good check. The study asks why the tool broke and whether the check is in the right place.
Packaging
A scanner finds that the product and its label don't match. The case is held instead of shipped. The team fixes the mismatch and reviews how the wrong labels reached the line.
Manual assembly
An operator finds a part that doesn't fit and calls for help instead of forcing it. The team compares the part and the method with the standard, and fixes that gap rather than treating resistance as normal.
Planning a jidoka pilot
Start with one quality problem, not a list of sensors. Pick a process where the team can describe the work, see the problem and control the response, then write a one-page brief:
Process boundary
The step, its inputs and the next handoff.
Normal condition
The approved standard for good work.
Target problem
The observable change to detect.
Affected output
What gets held or checked after a detection.
Response roles
Who reviews, who fixes and who studies the cause.
Success evidence
How the team will judge whether the pilot works.
Review the brief with the operators. Ask about hard cases, informal workarounds and times when the written method is unclear, and change the brief before changing any equipment.
Questions for the design review
Test the plan against these before it goes live:
- Can a known bad condition pass unnoticed? Check that the detection confirms the real condition, such as the part installed, not only that someone reached into the bin.
- Can normal variation set off a false stop? Nuisance stops teach people to bypass the system.
- Can the responder see the signal from where they work? If not, who covers?
- Can suspect work leave by another route? A rework bench or a side door is a common gap.
- Is there a controlled way to handle exceptions? Overrides need a named approver and a record.
Run the system with normal and deliberately bad conditions under an approved test plan before routine use. These questions don't replace engineering review; they make the assumptions visible to the people who do it.
Measuring whether jidoka works
Read a few measures together:
| Measure | What it shows | How to read it |
|---|---|---|
| Defects found downstream | Whether the target problem still escapes | A rise with fewer stops means detection got weaker |
| Repeat stops for the same cause | Whether countermeasures work | The same cause three times means the study stopped at the symptom |
| Time to acknowledge a signal | Whether help arrives | Long waits point to staffing or signal placement |
| Time to restore good conditions | Where diagnosis or restart checks are slow | Split it into waiting, diagnosing and verifying |
| False or nuisance stops | Whether the check needs tuning | Too many and people start bypassing it |
| Unplanned overrides | Where the designed response doesn't fit the real work | Each one is a design question, not a discipline problem |
Defects found downstream
Whether the target problem still escapes
How to read it: A rise with fewer stops means detection got weaker
Repeat stops for the same cause
Whether countermeasures work
How to read it: The same cause three times means the study stopped at the symptom
Time to acknowledge a signal
Whether help arrives
How to read it: Long waits point to staffing or signal placement
Time to restore good conditions
Where diagnosis or restart checks are slow
How to read it: Split it into waiting, diagnosing and verifying
False or nuisance stops
Whether the check needs tuning
How to read it: Too many and people start bypassing it
Unplanned overrides
Where the designed response doesn't fit the real work
How to read it: Each one is a design question, not a discipline problem
Fewer stops is not automatically good news. Early in a pilot, more stops can mean the system is finally showing problems that were always there.
Jidoka myths
Check a proposed system against these:
- 1
The alarm is the fix. Ask what the signal makes someone do and how suspect work is held.
- 2
Every stop needs the same response. Define roles and escalation for each kind of problem.
- 3
Operators should fix it quietly. Make asking for help easy, and keep approved fixes apart from improvised ones.
- 4
Restarting closes the issue. Track the cause study separately from getting the line running again.
- 5
A bypass is fine if output looks good. Exceptions need review and control, not casual overrides.
- 6
Jidoka needs expensive automation. An operator who can stop the line is jidoka too.
| Concept | Main role | Link to jidoka |
|---|---|---|
| Automation | Does work with less manual effort | Becomes jidoka only when it can detect a fault and stop |
| Andon | Makes status and problems visible | Calls help to the stop |
| Poka-yoke | Prevents or flags a specific error | Often provides the detection |
| Just-in-time | Makes what is needed, when it is needed | The other pillar of TPS; depends on good parts moving between steps |
| Kaizen | Improves the process step by step | Acts on the problems jidoka exposes |
Guides on this topic
- One-piece flow: how single-unit flow cuts waiting and WIPOne-piece flow explained: a worked comparison with batches, takt and cycle time, the benefits to check, production and transfer batches, flow and pull, where it fits, the conditions it needs, a pilot plan and how to measure it.Read the guide
- Muri: how to spot and reduce overburden in leanMuri explained: what overburden means, how it differs from muda and mura, the common causes, an observation checklist, six ways to reduce it, examples from factories to software, and how to measure progress without hiding overload.Read the guide
- Lean healthcare: principles, tools and how to startLean healthcare applies lean methodology to patient care. See the principles, hospital wastes, tools, patient flow and how to start in a clinic or ward.Read the guide
- 8D problem solving: the eight disciplines, step by step8D problem solving explained: D0 to D8, is/is not, containment, occurrence and escape causes, verified corrective actions, an example and the report.Read the guide
Sources
Books and papers show their publication year. Web pages show the month we last checked them.
- 1.Taiichi Ohno, Toyota Production System: Beyond Large-Scale Production (Productivity Press, 1988)
- 2.Toyota Motor Corporation, Toyota Production System (global.toyota, checked October 2026)
- 3.Jeffrey K. Liker, The Toyota Way: 14 Management Principles from the World's Greatest Manufacturer (McGraw-Hill, 2004)