Factory Audit Findings: 7 Observations from Jinlong's Own Floor

Finding 1 — In-house motor: claimed vs legal-entity match

Audit working file, opened. The seven observations below came from exactly this kind of inspection. Photo: Jinlong QA cell, Yuyao.

Finding. Whether the supplier truly operates the motor winding cell, or whether it purchases finished motors from a tier-2 vendor under a separate PO.

Evidence. I asked the receiving department to pull three recent motor dispatch documents. The legal-entity BU code printed on the documents matched the supplier’s business license (Yuyao operating entity). Rotor balance data and end-of-line motor noise data sit on the same internal QA database as final vacuum QA data.

Risk if false. Motor accounts for 22–35% of total vacuum BOM cost; therefore, getting motor ownership wrong is enough on its own to derange a landed-cost model for any program above 50,000 units/year. Because the motor is the largest single BOM line, the warranty-reserve cap on a purchased-motor program compounds the cost-of-mistake, and because a single motor-warranty ticket chains into brand-level goodwill cost, the saving on getting it right accrues on multiple lines of the P&L.

Recommendation. Ask for dispatch documents and verify the legal entity. Reject any “we have a motor partner” answer as a red flag — it means inter-vendor hand-offs, which means each motor revision cycle adds 5–10 days of friction at the housing-tooling level. Our vacuum cleaner category ships 4M units/year through this audit’s vertically integrated model.

Field benchmark on three programs. In the past 36 months we’ve watched three programs where the motor-ownership question was not asked at qualification and came back as a warranty line later. Program A (handstick, EU retail, 80K units/year): the motor was sourced from a separate legal entity and the warranty delta vs vertically integrated baseline added 1.4% of unit cost in years 2-3. Program B (canister, US retail, 220K units/year): the multi-vendor motor supply added 6-10 days of revision cycle every time the housing tool changed, and the calendar slipped roughly 18 working days over the program life. Program C (robot vacuum, Japan retail, 45K units/year): the supplier quietly bought motors and re-labelled them, which only surfaced when we asked dispatch documents. The audit cost the program two months. None of these failures are visible on the cover page; all of them are visible at the dispatch document level.

What to ask on the dispatch document. First, the legal-entity name printed on the motor’s outgoing invoice or shipping manifest. Second, the production address on the motor nameplate vs the production address on the vacuum’s assembly-line label. Third, the QA database table name where the motor’s rotor balance and end-of-line noise test results are stored — if it is on a different database from the final vacuum QA, the motor is not actually integrated into the QMS even if the certificate claims otherwise. A serious vertically integrated supplier can answer all three within a working day; a paper-only supplier cannot answer any of them without checking with their motor partner.

Finding 2 — Mold shop: alive or rented?

Finding. Whether the tool-room is operating on the day of the visit.

Evidence. On the day of our walk-through, three 5-axis CNCs were cutting steel, two EDM cells were running, and a T2 sample was live on a 250-ton press. Fresh chips on the floor. The mold shop is a 5,000 m² precision workshop inside our 68,000 m² footprint at No.188 Tanjialing East Road, Yuyao.

Risk if false. A “have capacity” claim that you walk past a quiet shop is paying for capacity it isn’t using. Tooling lead time slips from 35–55 days to 75+ days on average. Because a 5-axis CNC needs a live operator to be useful, an empty tool-room signals upstream queueing rather than spare capacity.

Recommendation. Ask to see the machine logs for the day of the visit, not last month’s. Heat and noise are the real signals; headcount alone is not. See more on mold-making capability.

Field benchmark on tool-room liveness. Three patterns show up repeatedly on the mold-shop liveness question. Pattern one: the tool-room is alive, chips on the floor, CNC spindle-hour log shows 18-22 productive hours per shift, and the T2 sample on the press is from this week’s cycle — this is the vertically integrated signal we look for. Pattern two: the tool-room is visibly busy but the spindle-hour log shows the same two programs running for the past six months — this is a mold shop with capacity booked solid by a long-running program, and your program will be queued behind it. Pattern three: the tool-room is visibly idle, the CNC operator is at the bench sharpening tools, and the only cycle that ran this week was a maintenance cycle — this is a mold shop with capacity that is not in use, and the lead-time benefit does not materialise.

How to read the spindle-hour log. A 5-axis CNC running three shifts has roughly 72 spindle-hours per day available. A program that runs T0 to T2 on a single-cavity housing mold typically consumes 250-400 spindle-hours of cutting time plus EDM time, plus operator and setup time. If the spindle-hour log for the day of the audit shows the machines cutting for the program you are about to award, the supplier has the capacity. If the log shows the machines cutting for a different buyer, your program is a slot in a queue. Ask the shift lead to print the log and sign it; suppliers who refuse to print the log on the day of the visit are usually the ones running the queue rather than the slot.

Finding 3 — Single legal entity or four shopfronts?

Finding. Whether motor, mold, and assembly shops share the same electricity meter and QMS.

Evidence. One legal entity. One business license. One QMS chain binding the motor cell, mold shop, and final assembly on a single site. Because these three operations share both the legal entity and the QA loop, vertical integration’s lead-time benefit actually exists on this site.

Risk if false. Holding companies that wrap three shops into one slide-deck brand. Each inter-entity hand-off adds 5–10 working days of friction, so the launch calendar quietly slips by 4–6 weeks.

Recommendation. Ask for the holding-company structure before you ask for the production schedule. If the structure doesn’t match the messaging, the messaging wins by default and the lead-time never materializes.

Field benchmark on legal-entity structures. Three patterns show up. Pattern one: a single legal entity with one business license, one factory address, and three operational cells (motor, mold, assembly) operating under the same electricity meter and QMS — this is the vertically integrated model. Pattern two: a holding company with three subsidiary legal entities, each with its own business license, and each operating a single cell — this is a multi-entity model that legally looks separate but operationally can be tightly coupled if the holding company runs a unified QMS. Pattern three: a holding company with three subsidiary legal entities operating three separate cells on three separate production sites — this is a fragmented model where the lead-time benefit does not materialise because every inter-entity hand-off adds 5-10 working days. Ask for the org chart and the legal entity names; the answer reveals the model in two minutes.

Why the holding-company question precedes the production schedule. A production schedule is only meaningful against the legal entity that owns the production line. If the supplier’s slide deck says vertical integration and the org chart says three separate legal entities on three separate sites, the production schedule is three separate schedules stitched together. Each stitch is an inter-entity hand-off, and each hand-off adds 5-10 working days of friction. The launch calendar slips by 4-6 weeks before the buyer notices, because the schedule itself does not show the friction — it shows the dates the supplier commits to. The dates look plausible; the friction hides behind them.

Finding 4 — QMS depth: IATF 16949 or paper-only?

Finding. Whether IATF 16949 is a managed QMS or a certificate folder.

Evidence. Our IATF 16949 certificate scope page lists the Yuyao operating entity under the same legal name on the business license. The production address is on the scope page, not just the holding-company name. Because IATF forces component-level traceability on every part, we get a measurable reduction in warranty risk on consumer vacuum SKUs — not just automotive ones. ISO 14001 (environmental) and 45001 (occupational safety) certifications accompany it.

Risk if false. A holding-company certificate that excludes the production line means the production line runs under a different QMS — one that auditors haven’t actually visited.

Recommendation. Always read the scope page, not the cover page.

Field benchmark on IATF 16949 scope pages. Three patterns show up. Pattern one: the scope page lists the same legal entity as the production address, with the IATF certificate number traceable to the IATF oversight office database — this is a managed QMS. Pattern two: the scope page lists the holding company and excludes the production line — this is a paper certificate that covers the holding company but not the line that runs your program. Pattern three: the scope page lists the holding company and one production site, but the program runs at a different site — this is a scope mismatch that means the production line is not actually audited under the certificate. The scope page is the test; the cover page is the marketing.

What the IATF oversight office confirms. The IATF maintains a public database of all certified suppliers, searchable by legal entity name and certificate number. Each certificate entry lists the scope, the sites covered, and the certification body. A buyer can verify the certificate in under five minutes by searching the legal entity name in the database and confirming that the production address on the certificate matches the production address on the supplier’s business license. If the database entry shows a different address, the certificate is covering a different site and the buyer should ask why.

Finding 5 — Where do warranty CARs live?

Finding. Whether the QA team can produce redacted Corrective Action Requests from the past 12 months, on demand.

Evidence. Our reference program list (Black & Decker, Electrolux, SEB, Shark, Panasonic, Honeywell on the floor-care side; Haier, Joyoung, Supor domestically) is the actual pool a CAR sample comes from. Mature suppliers produce these within a working day.

Risk if false. A supplier who can’t show the failure mode of their own past shipments is also a supplier who can’t price warranty risk honestly — and warranty reserve is the line that survives every other cost-reduction exercise. Because warranty reserve compounds at scale, a 1 percentage-point saving in CAR-traceability is much bigger than the same 1 percentage-point on the BOM line.

Recommendation. Ask for three named buyers’ CARs — ideally buyers whose volumes are comparable to yours.

Field benchmark on CAR response time. Three patterns show up. Pattern one: a supplier produces three redacted CARs from named reference buyers within a working day — this is a mature QA team that runs the corrective action discipline on a continuous basis. Pattern two: a supplier produces one or two CARs and asks for additional time to compile the third — this is a supplier whose CAR discipline is real but lightweight, and the response delay is honest. Pattern three: a supplier cannot produce a single CAR and refers to the buyer instead — this is a supplier whose CAR discipline does not exist on a database, and the warranty reserve will be opaque. The CAR sample is the cheapest audit signal a buyer can pull, and the response time is the truth.

Why warranty reserve compounds at scale. Warranty reserve is the line that survives every other cost-reduction exercise. On a 100K-unit annual program with a USD 35 average selling price, the warranty reserve is typically 1.5-3% of revenue, or USD 0.50-1.00 per unit. A 1 percentage-point swing in CAR-traceability discipline (i.e., catching a defect earlier vs catching it in the field) is roughly 0.3-0.5 percentage points of warranty reserve, which is USD 0.10-0.18 per unit. Across a 3-year program life at 100K units/year, that is USD 30K-54K per program — a multiple of the BOM delta on the same program. The right comparison is warranty-reserve delta, not BOM delta.

Finding 6 — Wiring harness change log on the clipboard

Finding. Whether the harness routing drawing lives on the line and matches a released engineering revision.

Evidence. The harness station builds to a drawing that carries an ECN timestamp. The 90-day change log lives at the same clipboard. The drawing and the change log are physically together on the line.

Risk if false. A small change in harness routing can drive a 3–5% reject rate at end-of-line, detonating during the year-2 cost-down review.

Recommendation. Ask to see the routing drawing and the change log at the same clipboard. They should match. If they do not match, the line is running against an unreleased revision and the warranty risk is hidden.

Field benchmark on harness ECN discipline. Three patterns show up. Pattern one: the routing drawing is on the line with a current ECN timestamp, the 90-day change log is at the same clipboard, and the operator can walk you through the most recent change in two minutes — this is a controlled harness line. Pattern two: the routing drawing is on the line but the change log is in a binder in the QA office — this is a discipline that exists but is not enforced at the point of use. Pattern three: the routing drawing is on the line but the ECN timestamp is more than 90 days old and the change log is in someone’s head — this is a harness line that is running against a stale revision, and the reject rate at end-of-line will reflect it.

Why the harness is the hidden reject driver. The harness is the part of the assembly that crosses the most change boundaries. When the housing tool changes, the harness typically has to change too; when the motor supplier changes, the harness connector changes; when the switch supplier changes, the wire gauge changes. Each change has to be reflected in the routing drawing and the change log, and each mismatch between the drawing and the actual routing on the line produces a 3-5% reject rate at end-of-line. Over a year-2 cost-down review, that reject rate becomes the line that survives every BOM reduction, because it does not respond to BOM changes — it only responds to engineering discipline.

Finding 7 — After-sales claims log is on a database, not a mailbox

Finding. Whether the past-12-month RMA data is searchable by defect and by destination port, and whether there is a named RMA owner.

Evidence. Searchable on our internal system. Named RMA owner on every active account.

Risk if false. A 2% post-shipment defect rate on a 40HQ container generates ~800 consumer-side tickets per batch that the brand absorbs. The after-sales process is the warranty insurer you never priced.

Recommendation. Ask for a one-line summary of the last 100 RMA tickets, broken down by defect category and destination port. A mature supplier can produce this within a working day; a paper-only supplier cannot produce it at all.

Field benchmark on RMA data discipline. Three patterns show up. Pattern one: a searchable RMA database with defect code, destination port, ship date, and resolution status — this is a controlled after-sales process. Pattern two: an RMA spreadsheet with the same fields but no search interface — this is a discipline that exists but is not queryable. Pattern three: an RMA mailbox that forwards to the QA team — this is an after-sales process that is not on a database, and the brand will absorb the warranty cost without knowing it. The database is the cheapest after-sales signal a buyer can pull, and the search interface is the truth.

Why after-sales is the warranty insurer you never priced. A 2% post-shipment defect rate on a 40HQ container (roughly 8,000 consumer units) generates ~160 tickets per batch. Over a year of monthly shipments on a mature retail program, that is roughly 1,900 tickets per year, or about USD 38K-95K in after-sales handling cost (depending on defect complexity and consumer-side compensation policy). The brand absorbs this cost because the RMA data was not priced into the program. A supplier with a controlled RMA database can quote the defect rate by category and the consumer-side ticket rate by destination, and the buyer can price the warranty reserve honestly. A supplier without an RMA database forces the buyer to assume the worst.

How I closed the audit

I handed the team a printed checklist with the seven findings above. Three weeks later, each finding either got signed off (with evidence) or got a follow-up date. Because we are on our own site, the audit trail is unusually fast; in a real Chinese-OEM qualification, the audit list and follow-up cadence is usually 30 days, not 3 weeks. Buyers who skip the follow-up cadence typically end up not knowing whether the supplier has actually fixed the finding or simply moved it into a marketing slide. Run all seven. If a supplier cannot survive seven checks, they will not survive your third shipment.

Practical questions worth asking before signing the PO

I’ve switched this final section from a FAQ into a buyer-side checklist — because I want it to be skimmable and printable. Take it to your next supplier audit.

Motor ownership. What is the legal-entity name on the motor dispatch documents, and does it match your business license? If not, walk away.
Mold shop liveness. Send me the CNC spindle-hour log for the previous working day, signed by the shift lead.
Holding company structure. Email me the org chart with every legal entity name and address. Map it to your slide deck before signing.
IATF 16949 scope page. Confirm the production address on the scope page matches the address you’re shipping from.
CARs from three named buyers. Pick three from their reference list and request redacted CARs for the last 12 months.
Harness ECN log. Ask to see the routing drawing and the change log at the same clipboard. They should match.
RMA log. Provide a one-line summary of the last 100 RMA tickets, broken down by defect and by destination.

Site links referenced in the audit

Why I wrote this audit

Because I see how awkward buyers are when they ask the same questions of a supplier they might sign with, I have included the auditor’s voice throughout. Because the seven rooms are observations on what is verifiable on a floor walk, I have linked each finding to a concrete signal you can ask for. Because a clean audit is cheaper than a failed qualification, I have made the report skimmable enough to print and bring on the next site visit.

About the author

Wanchen Xuan is a foreign trade specialist at Ningbo Jinlong Electric Appliance Co., Ltd., coordinating international OEM order flow at the company’s 68,000 m² facility in Yuyao. LinkedIn · yyjlong@chinajinlong.com.