A control plan is a written document that lists every product and process characteristic you need to control, and for each one records its specification, the control method, the sample size and frequency, the measurement system, and the reaction plan for when it goes out of limits. It is the document that keeps an improvement in place after the project team goes home.
Most quality problems are not new. They are old problems that were fixed once and quietly came back because nothing was written down to hold the fix. The control plan is the antidote. In lean and Six Sigma work it is the deliverable of the final Control phase, the point where a proven improvement stops depending on the people who made it and starts depending on a system. This guide covers what belongs in a control plan, the three types the automotive standard defines, how to fill one out column by column, and how to keep it from turning into a binder nobody opens.
What is a control plan?
A control plan is a structured summary of how a process will be kept in control, characteristic by characteristic. Where a work instruction tells an operator how to do the job, the control plan tells the organization how the job will be checked: what gets measured, how often, by what method, and what happens when a measurement falls outside its limit.
The concept comes out of automotive quality. The AIAG (Automotive Industry Action Group) publishes the reference manual that most industries borrow from, where the control plan is a required output of Advanced Product Quality Planning, APQP (AIAG, APQP & Control Plan). You do not need to be an automotive supplier to use one, the same structure works for a bakery controlling dough temperature or a machine shop holding a bore diameter.
The point of writing it down is transfer. A control plan lets the person who understood the process on the day it was validated hand that understanding to every shift that runs it afterward, without being in the room. That is also why it sits at the heart of the fight against tribal knowledge it converts "Maria always checks the third fill head because it drifts" into a documented control anyone can run.
What goes in a control plan?
A control plan is a row per characteristic and a fixed set of columns. The columns are what make it powerful, each one forces a decision that vague "we check quality" talk skips over. The core columns, in the order they usually appear:
| Column | What it captures | Example |
|---|---|---|
| Process step | Where in the process this is controlled | Op 30, fill |
| Characteristic | The product or process feature being controlled | Fill weight |
| Specification / tolerance | The target and allowed range | 500 g ± 3 g |
| Measurement method | The gauge or test used | Checkweigher |
| Sample size | How many units per check | 5 |
| Frequency | How often the check runs | Every 30 min |
| Control method | How the data is used | X-bar & R chart |
| Reaction plan | What to do when it's out of limits | Hold lot, adjust filler, quarantine |
Two columns deserve special attention. The specification with its tolerance has to be a real, measurable limit, "smooth finish" is an argument, "Ra ≤ 1.6 µm" is a control. And the reaction plan is what separates a control plan from a checklist: it names exactly what happens when the check fails, so the decision is made in advance in a quiet room, not improvised in the middle of a bad shift.
What are the three types of control plan?
The AIAG framework defines three control plans that track a part's life from first prototype to full production. They are not alternatives, a part moves through all three (AIAG, APQP & Control Plan).
- Prototype. Used during the build of early prototype parts. It carries heavy dimensional measurement and material and performance testing, because at this stage you are still confirming the design is right. It typically checks more, more often, than production ever will.
- Pre-launch. Used after prototype and before full production is approved, during trial runs and launch. It often keeps enhanced checks in place: extra measurements, tighter frequencies, added validation, because the process is not yet proven at rate.
- Production. The steady-state plan for mass production. It documents the full set of product and process characteristics, the ongoing controls, and the measurement systems used once the process is validated and running. This is the one that lives on the floor for years.
The through-line is confidence. Early on you check a lot because you know little; as the process proves itself, checks shift from broad verification toward targeted control of the characteristics that actually drift.
How do you fill out a control plan?
You build a control plan by choosing what to control before you decide how to control it, and you choose using risk, not habit. The sequence that produces a plan people actually run:
- List the characteristics from your risk analysis. Pull them from the process FMEA and from the customer requirements, the critical-to-quality features. Do not try to control everything; control what carries risk. A special or "critical" characteristic that affects safety or function gets tighter control than a cosmetic one.
- Write a measurable spec and tolerance for each. Every characteristic needs a number and a range, tied to the drawing or the customer spec. If you cannot state a limit, you cannot control it, you can only argue about it.
- Pick a measurement method and confirm the gauge is capable. A control is only as trustworthy as the gauge behind it. Run a measurement systems analysis (gauge R&R) so you know the measurement error is small relative to the tolerance before you rely on it.
- Set sample size and frequency by risk and stability. Higher-risk, less-stable characteristics get larger samples and more frequent checks. A proven, capable, stable characteristic can be checked less often. Base this on process capability not on what feels safe.
- Choose the control method. Decide how the data gets used: 100% inspection, a control chart for a variable characteristic, a poka-yoke that prevents the error outright, or an attribute check. Prevention beats detection wherever it is possible.
- Write the reaction plan. Spell out exactly what happens on an out-of-limit result: contain the suspect product, who is notified, what adjustment is made, and how you handle everything produced since the last good check. This is the column that fails audits when it is vague.
- Assign an owner and a review trigger. Name who owns the plan and when it gets revisited, after any change to the product, process, or equipment, and whenever a corrective action touches one of its characteristics. A control plan with no owner is a control plan that is already out of date.
How does a control plan connect to DMAIC?
The control plan is the deliverable of the Control phase, the "C" in DMAIC (Define, Measure, Analyze, Improve, Control). ASQ describes DMAIC as the structured cycle that runs a problem from definition through a sustained fix, and the final phase is where controls are put in place to hold the gain (ASQ, DMAIC Process).
That last word, sustained, is the whole reason the document exists. A team can Define a scrap problem, Measure the baseline, Analyze the root cause, and Improve the process to cut defects in half, and it all evaporates in a quarter if nothing captures the new settings, the new checks, and the new reaction plan. The control plan is where the Improve gains get banked. Skip it and you have not finished a Six Sigma project; you have run an expensive experiment whose results you chose not to keep. That lost gain is not abstract: an improvement that reverts quietly reloads the cost of quality the project was meant to cut, and it undoes the work of the kaizen event or project that produced it.
It pairs naturally with statistical tools. Where a characteristic is charted, statistical process control is the engine and the control plan is the instruction sheet that says which characteristic, which chart, which sample, and which reaction. One tells you the process is drifting; the other tells you what to do about it.
Why do control plans go stale, and how do you stop it?
Control plans die for the same reason lean programs stall: the document and the floor drift apart until the plan describes a process that no longer exists. The failure modes are predictable.
The plan and the process change independently. Someone tweaks a fixture, moves a check, or swaps a gauge, and the control plan never gets updated. Six months later an auditor finds the floor doing one thing and the plan saying another. Every process change needs to trigger a plan review, that is the whole job of the owner column.
Reaction plans that only exist on paper. The plan says "hold the lot and notify quality," but at 2 a.m. the operator adjusts the machine and keeps running because that is what actually happens. A reaction plan nobody follows is worse than none, because it creates a false record of control.
Checks recorded on paper nobody reviews. If the check data lands on a clipboard that gets filed unread, the control plan is theater. The value of a control is the fast feedback when something drifts, and paper logs totaled at month-end are not feedback, they are history. This is where connecting the plan to live data changes what the document can do.
When checks are captured digitally at the station and flow into one operational layer, an out-of-limit result can trigger the reaction plan in the moment, flag the characteristic, prompt the operator, hold the lot, instead of surfacing days later in a review. Harmony digitizes the checks operators already run and connects them to the machines and systems around the line in one live layer with no rip-and-replace of your existing gauges or software. When CLS moved its production and quality logging off paper the checks stopped being a record filed after the fact and started being a signal the team could act on during the shift, which is exactly what a control plan is supposed to be.
Start with the handful of characteristics that carry the most risk, write real specs and real reaction plans for them, and make sure the check data comes back fast enough to act on. A control plan is not a compliance artifact. It is the memory that keeps a fix fixed.