Implementation Guide
Layered Process Audits.
Layered process audits (LPAs) are short, frequent checks that work at the process still matches the standard, performed by every layer of management: team leaders daily, area managers weekly, plant leadership monthly. Their product is findings, not scores. Audits measure the system's honesty, and findings are free consulting; an LPA program that celebrates 98 percent conformance has usually stopped asking real questions.
What layered process audits are
A layered process audit is a short, scheduled check that the work happening at a workstation still matches its standard: the settings, the sequence, the checks and the conditions the standard sheet specifies. Layered means every level of management runs the check against the same standards, each on its own cadence: the team leader daily, the area manager weekly, plant leadership monthly. Each audit is a handful of questions answered by looking at the process, not a review of documents in an office.
The method grew up in the automotive supply chain, where AIAG formalized it as the CQI-8 guideline, and it exists because standards decay silently. A tool gets substituted, a setting drifts, a check gets skipped on a busy Friday, and within a week the workaround is tribal knowledge. Annual system audits catch that drift months later, after it has cost something. An LPA catches it in days, while the fix is cheap and the trail is warm.
The output that matters is the finding. Each finding is a specific, dated statement of where reality has separated from the standard, discovered by your own people before it became scrap, a customer complaint or an accident. That is free consulting, and it is the entire return on the program. The conformance score is a byproduct, and a program that celebrates 98 percent conformance has usually stopped asking questions capable of failing.
Where LPAs sit in the transformation roadmap
LPAs are a sustain mechanism, which is why they come late. On the TeamGuru deployment roadmap they belong to the Standardize & Scale practice: they protect what the earlier stages built by verifying at the process that standardized work is still followed and followable, that 5S still makes abnormality visible, and that poka-yoke devices still work and have not been quietly bypassed. Without an audit layer, every one of those practices depends on nobody ever being busy, promoted or absent.
- Before Standard Work & 5S
- Before Obeya & Management Reviews
- You are here Standardize & Scale
- In parallel Skills Matrix
When audits backfire
An audit program can make a plant worse, and it usually does so in one of three ways. Do not launch until all three are ruled out:
- There is nothing to audit yet. An audit compares reality to a standard; if standards do not exist, or live in a binder nobody at the line can find, write and post the standard first. Auditing opinions produces opinions.
- Audits are used to grade people. The moment a finding hurts someone's bonus or shows up in a performance review, honesty dies: findings get negotiated down, auditors soften their questions, and the boards turn a comfortable green. The program keeps its cost and loses its information.
- Nobody has capacity to react. If findings pile up without dispositions, every audit teaches the organization that looking closely changes nothing. Five questions with a working reaction loop beat forty questions with none.
The common thread: LPA programs fail socially before they fail technically. Most of the design work below is protection for honesty.
How to build the program
Start from what decays, not from what is easy to check. Three sources fill the first question pool:
- Standards that changed recently. Improvement gains revert fastest in their first 90 days, so anything a kaizen or a workshop changed goes into the audit rotation immediately.
- Critical quality controls: the checks your FMEA and control plan assume are happening every shift, such as torque verification, first-piece checks and error-proofing devices.
- New standards: new products, new equipment, new people. Wherever a standard is young, drift is likely and cheap to correct.
The second of those is where the audit program connects to your FMEA: a control that exists only in the analysis and never in anyone's checklist is not a control. Then structure the layers. The layering is the method's real leverage: the same standards seen by fresh eyes, with growing authority and growing distance from the area's daily habits.
| Layer | Who audits | Cadence | Scope | Questions | Time per audit |
|---|---|---|---|---|---|
| Layer 1 | Team leader or shift supervisor | Daily | Their own area, every station over a week | 5 to 7 questions | 10 to 15 minutes |
| Layer 2 | Area or value stream manager | Weekly | Rotating across the areas they manage, including a sample check of layer 1's recent audits | 8 to 10 questions | 20 to 30 minutes |
| Layer 3 | Plant manager and leadership team | Monthly | Anywhere in the plant, plus the audit program itself: are the lower layers auditing, and are findings getting dispositions | A subset of the same questions, plus program-health questions | 30 to 45 minutes |
Put the layer 1 and layer 2 audits on the leaders' routine cards as part of leader standard work, so the audit has a protected slot instead of competing with the day's noise. Then keep the question pool alive: rotate roughly a third of the questions monthly, and retire any question that has not produced a finding in three months. A question that is always green is either a solved problem or a badly asked question. Both mean the minute it costs belongs to a different question.
Writing questions that find something
Question design decides whether the program finds anything at all. Four rules cover most of it:
- Answerable by direct observation in about one minute: the auditor looks, counts or watches one cycle. Anything that needs investigation is a project, not an audit question.
- Verified against a written standard: the correct answer exists on a standard sheet, a chart, a label or a photo of the standard state, never in the auditor's judgment.
- Specific: the question names the station, the parameter and the document, so two different auditors reach the same answer on the same day.
- The evidence is part of the question. Not "is torque controlled" but "is the driver set to the value on the standard sheet", so the answer is checked, not felt.
| Bad question | Why it fails | Better question |
|---|---|---|
| Is 5S maintained in the area? | A yes/no opinion with no evidence named. It is always yes by the afternoon. | Pick one shadow on the station 2 tool board: is the tool present or signed out? |
| Are operators following standard work? | Unobservable as asked, and it audits the person instead of the process. | Watch one cycle at station 3: does the sequence match the 7 steps on the standardized work chart? |
| Is the torque equipment properly maintained? | Nobody can verify "properly" in a minute, so nobody tries. | Is the torque driver at the fitting station set to the value on the standard sheet? |
| Is quality documentation up to date? | Too broad. It invites a glance at a binder and a checkmark. | Open this shift's first-piece check record: complete, signed, and within limits? |
A worked six-question audit
Here is what a layer 1 audit looks like for the assembly cell of the 450-person components manufacturer used across these guides: 456 units per day, takt 118 seconds. The questions are illustrative, and deliberately short. Watching one full cycle costs two minutes at this takt, and the whole audit fits inside 15 minutes.
| # | Question | What it protects |
|---|---|---|
| 1 | Ask the operator at station 3 to show the standardized work chart for the current product. Is it the posted revision, and is the standard WIP of 2 pieces it specifies actually in place? | Standard work integrity: the chart is current and the cell runs to it |
| 2 | Watch one full cycle at station 3. Does the work follow the 7 steps on the chart, in order? | The sequence as written, observed rather than asserted |
| 3 | Is the torque driver at the fitting station set to the value on the standard sheet, and does the fastening counter show the last cycle complete? | A critical quality control and the poka-yoke device behind it |
| 4 | Open this shift's first-piece check record. Is it complete, signed and within limits? | The control plan check the FMEA assumes is happening |
| 5 | Pick two shadows on the station 2 tool board. Is each tool present or signed out on the board? | 5S as an abnormality detector, not as housekeeping |
| 6 | Ask the operator: what changed at this station in the last month, and where is the current standard for it? | Whether recent improvements made it into standards and training |
Running an audit
Go to the process and say plainly what you are doing. LPAs do not need to be surprise inspections: you are auditing the process, not ambushing the person, and a standard that only holds when nobody is watching is a finding either way. Observe before asking. Watch the cycle, look at the settings, count the pieces, and only then start the conversation.
Ask people to show, not tell. "Show me how you confirm the torque" produces evidence; "do you confirm the torque?" produces reassurance. And when something deviates, write the finding without blame, because a violated standard is data. Either the standard is wrong or unfollowable, or the training did not hold, or the condition drifted. All three are system problems with different fixes, and none of them is a verdict on the operator. The question an auditor carries is what allowed this, never who did this.
Two behaviors decide the program's reputation on the floor. First, never fix silently and walk away: restoring a condition without recording the finding hides the signal the program exists to collect. Second, thank people for what they show you. The operator who says "this gauge sticks, we all tap it twice" just handed you a finding no checklist would have surfaced. Punish that honesty once and the program goes blind for a year.
From findings to actions
Every finding gets a disposition within a week, and there are only three:
- Fix. The standard is right and the condition drifted: a missing tool, a wrong setting, a worn label. Restore the condition, verify it, and note whether the same fix keeps recurring.
- Train. The standard is right and followable, but execution drifted. Retrain at the process, not in a room, and ask the harder question: why was the drift invisible until an audit found it?
- Change the standard. The standard is wrong, outdated or unfollowable, and the operator was compensating for it. Fix the document, not the person. In a healthy program this is a large share of dispositions.
Findings and dispositions are trended monthly, in the management review: findings raised and closed within a week, repeat findings on the same question, and audit completion by layer. Read the conformance percentage as an honesty signal, not a target. A healthy program sits comfortably below 100 percent and produces a steady stream of small, closed findings; findings per audit falling toward zero while the questions stay unchanged is a warning, not a victory.
Common failure modes
What bad looks like
- Checklists filled in from the office, from memory, at the end of the week
- A 40-question marathon that takes an hour, so it slips, then quietly stops
- Findings logged in a spreadsheet where nothing has an owner or a disposition
- 98 percent conformance celebrated in the monthly deck while scrap stays flat
- The same questions for a year, so everyone knows them, including the answers
- The quality department audits alone while line management signs the form
What good looks like
- 5 to 7 questions, 10 to 15 minutes, at the process, in a protected slot
- All three layers staffed by line management, up to the plant manager
- Every finding dispositioned within a week: fix, train, or change the standard
- Questions rotate monthly; always-green questions get retired
- Scores read as trends and honesty signals; findings treated as the product
- Results and open findings visible in the area, next to the team boards
What happens next
A running program produces two streams of value. The first is the finding-by-finding sustain loop above. The second is the pattern: a question that keeps producing findings marks a standard under systematic stress, a changeover standard that never survives the night shift, a check that every layer finds skipped. Those patterns are improvement targets with evidence attached, and they feed the kaizen pipeline with better input than any suggestion box.
What kills paper LPA programs is rarely the auditing; it is the bookkeeping. Three layers, rotating question sets, and dozens of findings per month, each needing an owner, a disposition and a due date. This is where TeamGuru holds the practice: the checklists and audits use case schedules every layer, runs the questions on a phone or tablet at the line with photo evidence on findings, and hands each finding to action management as an owned, dated item, so the monthly trend review reads live data instead of a collated spreadsheet.
On the roadmap, LPAs do not end; they run for as long as the standards they protect. They are the audit muscle of Standardize & Scale, and they work alongside the skills matrix, which develops the very capability the audits keep verifying: people who can run the process to standard, and leaders who can tell when it is drifting.
