Layered Process Audits (LPA): the practical CQI-8 guide

What is a layered process audit?

A layered process audit is a short, standardized check of a manufacturing process against its own documented standards — the control plan, the standardized work, the error-proofing, the setup sheet — performed frequently, by several layers of management, on a defined cadence, with a defined reaction to every answer that comes back “no.”

Each of those italics is doing work, and each is where a program usually goes wrong:

• Standardized. The same short checklist, every time, for that process. It is not an inspection of the parts and not an open-ended walk-around; it asks whether the process is being run the way it was designed to be run.
• Frequent. Daily or per shift at the first layer. The point is to catch drift in hours, not to discover it at the annual audit.
• Layered. Supervisors do it most often, managers less often, the plant manager less often still — but everyone does it. Management presence on the floor is a designed feature, not a side effect.
• Reaction plan. A “no” is never just recorded. It triggers a defined action on the spot, and repeated “no”s escalate into corrective action.

Some plants — and some OEM customer-specific requirements — call the same thing a layered process confirmation (LPC), a layered audit, a layer audit, or simply LPA. A layered process confirmation is not a different method: same short standardized checklist, same layers, same reaction plan. The name changes; the mechanics don’t.

Where LPAs come from: CQI-8, TS 16949 and IATF 16949

A note on names first: ISO/TS 16949 (often written TS16949) was the automotive quality standard from 1999 until it was replaced by IATF 16949:2016. Searches for “layered process audits TS16949” and “IATF 16949 LPA” are asking the same question — LPAs were never a clause in either; they arrive through the customer-specific requirements that sit on top of the standard.

LPAs were popularized in the North American automotive supply chain in the early 2000s, driven largely by General Motors, and were codified in AIAG’s CQI-8 Layered Process Audit Guideline(first published 2005; second edition 2014). CQI-8 is the reference most customers mean when they say “do LPAs”: it lays out the purpose, the layer structure, how audit items are chosen, the reaction plan, and how results are tracked and reviewed.

Where the standard fits is a common point of confusion, so to be precise:

• ISO/TS 16949 was the automotive QMS technical specification until 2016. It required internal audits of the QMS, the manufacturing processes, and the product — but did not use the words “layered process audit.” People still search for “layered process audits TS 16949” because that is the standard many programs were built under.
• IATF 16949:2016 replaced TS 16949. It also does not name LPAs. Clause 9.2.2.3 requires manufacturing process audits, and the standard repeatedly defers to customer-specific requirements (CSRs).
• The CSRs are where the obligation lives. GM’s customer-specific requirements call for layered process audits and reference CQI-8 by name; other OEMs have comparable expectations. If a customer’s CSR requires LPAs, that requirement is part of your contract and an IATF auditor will assess against it.

Practical takeaway: read your customers’ CSRs before you design the program. If none of them require LPAs, you are free to build a lighter program shaped to your own risks — and many non-automotive shops do exactly that, because the method works regardless of who mandated it.

The layers and their cadences

“Layered” means layers of management. Operators are not the auditors — their process is what is being audited. A typical structure looks like this; the number of layers and the exact frequencies are yours to choose, but every layer must have a defined cadence:

Why the pyramid shape works:

• The high-frequency bottom layer catches drift quickly and keeps standard work honest day to day.
• The upper layers audit the same checklist. When a manager’s audit finds a “no” on an item the supervisor has been marking “yes” all week, that gap is itself the finding — the layers audit each other.
• Managers on the floor with a checklist in hand changes behaviour on both sides of the clipboard. That is the part CQI-8 is really after.

Two design choices worth making deliberately: rotation (assign each auditor a different area each period so every process gets covered and nobody cherry-picks the easy cell) and independence (a supervisor should not routinely audit only their own line — have Layer 1 auditors swap areas at least some of the time).

What an LPA checks — and how to write the questions

The checklist is short — 5 to 10 yes/no questions that take about 10–15 minutes at the process — and every question is aimed at something that would hurt if it silently stopped being true. Good sources for questions:

• Error-proofing devices. “Was the poka-yoke on OP30 verified with the red-rabbit part this shift, and is the verification recorded?”
• Key / special characteristics from the control plan. “Is the in-process check for the Ø12.00 bore being done at the frequency the control plan says, with the gauge the control plan names?”
• Standardized work. “Is the operator performing the sequence on the posted work instruction? Is the instruction the current revision?”
• Setup and change control. “Does the setup sheet match the machine parameters?” “Is the first-off approved and tagged?”
• Gauges and equipment. “Is every gauge at the station within its calibration date?”
• Recent problems. The last customer complaint, the last internal NCR, the last audit finding — put a question on it until it stays fixed, then retire it.
• Safety and 5S basics — one or two, not the whole checklist.

Rules for writing them:

• Every question is observable at the process in minutes. If answering it means opening a database in the office, it belongs in a process audit, not an LPA.
• Every question states the expected condition so two auditors reach the same answer.
• Every question has a reaction plan for “no” (next section).
• The checklist changes. Retire questions that have passed for months; add one for every new escape. A checklist that has not changed in a year is a checklist nobody is reading.

The reaction plan: what happens on a “no”

This is the part that separates an LPA from a survey. For each question, the checklist says what the auditor does when the answer is “no” — before leaving the process:

1. Correct on the spot if it is correctable (wrong revision posted → get the current one; gauge out of date → swap it; parameter drifted → reset and note it).
2. Contain if product may be affected: quarantine what was made since the last “yes,” and raise a nonconformance so the disposition is recorded.
3. Escalate if the auditor can’t fix it — to the process owner, with a named person and a date.
4. Record the “no” and the action taken, in the audit record — not on a sticky note.
5. Watch for repeats. The same “no” on the same question across audits is a systemic issue: open a corrective action, and keep the question live until the fix is verified effective.

A “no” with no recorded action is worse than no audit at all: it documents that management saw the problem and walked past it.

Building the LPA schedule

The schedule is a matrix: layers × areas × periods. Build it once, publish it, and measure adherence to it — schedule adherence is the first metric a customer auditor will ask for.

1. List the processes / areas to be audited (cells, lines, work centers). Not every process needs the same frequency — weight toward high-risk and customer-facing ones.
2. Assign a cadence per layer (the table above is a starting point).
3. Assign auditors to areas per period — round-robin rotation is the simplest fair scheme; random assignment works too, as long as every area is covered every cycle.
4. Publish the plan so each auditor knows what they owe this week, and remind them before the period closes.
5. Track planned vs completed, per layer and per area, every period. Below 100% is normal; unmeasured is not.
6. Review at management review — adherence, top “no” questions, repeat findings, and open actions.

Charting and reporting LPAs

Four charts carry almost all of the value; anything beyond them is decoration.

• Schedule adherence by layer — completed ÷ planned, per period, one line per layer. If Layer 3 sits at 40%, that is the finding: management is not on the floor.
• Conformance rate — “yes” ÷ total answers, per area and per period. Watch the trend, not the level; a jump to 100% overnight usually means pencil-whipping, not perfection.
• Top failed questions (Pareto) — which items keep coming back “no,” plant-wide. These feed the corrective-action queue and tell you which questions to keep.
• Coverage — a heat map of areas × periods showing which processes were actually audited. Gaps hide here.

Report the layer-adherence and top-failed-question charts at management review, with the open action list. That package — schedule, records, adherence, actions, review evidence — is what a customer or IATF auditor checks when they assess an LPA program.

LPA vs process audit vs internal audit vs inspection

They are complements, not substitutes. LPAs keep the daily discipline that makes the deep audits find fewer surprises. (If you need the annual layer, our free ISO 9001 internal audit checklist for Excel covers the schedule, checklist and findings tracker.)

Eight ways LPA programs quietly die

1. Pencil-whipping. All “yes,” every day, from the office. Fix: rotation, upper-layer audits of the same checklist, and the occasional question that should come back “no.”
2. The checklist never changes. Fix: retire proven questions, add one per escape, review quarterly.
3. No reaction plan. “No” gets a tick mark and nothing else. Fix: an action is required to close the audit.
4. Only Layer 1 does it. Managers “don’t have time.” Fix: chart adherence by layer at management review; visibility fixes this faster than policy.
5. Auditing your own area, always. Fix: rotation and independence rules.
6. Treating it as inspection. Measuring parts instead of checking the process. Fix: rewrite questions to be about the process.
7. No evidence trail. Paper sheets in a binder nobody can summarize. Fix: whatever system you use, it must produce adherence and trend charts without a week of typing.
8. Findings never reach CAPA. The same “no” for six months and no corrective action. Fix: a repeat rule — same question, same area, N times → CAPA.

A 30-day rollout

1. Week 1 — scope and layers. Read the CSRs. Pick the areas, define 3 layers and cadences, name the auditors. Get the plant manager to commit to Layer 3 in writing.
2. Week 2 — checklists. Draft 5–10 questions per area from the PFMEA/control plan, error-proofing list, and last 12 months of complaints and NCRs. Write the expected condition and reaction plan for each. Pilot on one line.
3. Week 3 — schedule and training. Build the layers × areas matrix, publish it, and train auditors on how to audit: observe, ask, don’t coach during the audit, record what you saw.
4. Week 4 — go live and chart. Run every layer once. Chart adherence and the top “no”s. Take both to the first management review and fix the schedule where it is unrealistic.

Then keep it alive: monthly checklist review, quarterly rotation refresh, and adherence on the management-review agenda permanently.

Questions people ask

Written from 16 years in manufacturing quality — machining, fabrication and assembly — before building NominalQC. Corrections welcome via the contact page.