What a Battery QMS Must Prove Before You Award a Pack Contract

A pack supplier's quality management system only earns a place in the award decision when it proves three things: that the certificate scope covers custom lithium-ion pack assembly at the exact site that will run your build, that the control plan and inspection criteria were written against your configuration rather than a generic assembly, and that the records trail — cell lot, BMS firmware revision, end-of-line test data, nonconformance history — comes out of the same system that wrote the procedures. Miss any one of those and the certificate logo on the quotation cover page is decoration.

The room is usually the same. Conference table, projector, a binder with tabs, and a quality engineer who has been told to walk the visiting buyer through the system. I don't open with the binder. I ask which building on this campus actually assembles the pack, and whether we can pull the certificate and read the scope statement together, out loud.

That question sorts suppliers faster than anything else in the process. Some reach for the document without hesitating. Others explain that the pack line is at a different address, or that the certificate is held by the parent company, or that a "group" certificate covers it. All three answers are useful. None of them is the answer I want.

The certificate is a claim. The scope statement is the evidence.

A certificate number on a quote tells you somebody paid for an audit. The scope statement tells you what was audited. It's a sentence or two on the certificate itself or on an accompanying schedule, and it names activities. "Manufacture of lithium-ion battery packs" is not the same scope as "assembly of electronic assemblies" or "distribution of battery cells." If the scope language doesn't reach custom pack assembly, the audit didn't reach your product.

Then three more checks that take ninety seconds:

• The site address on the certificate. Not the sales office. Not the headquarters city. The plant that will build your pack.
• The accreditation mark. Certification bodies are themselves accredited, and the mark identifies which accreditation body oversees them. That's a check on the checker.
• Current status. Certificates get suspended, withdrawn, and expired. A PDF from three years ago and a current listing are different documents.

I've watched a buyer discover mid-program that the site on the certificate and the site on the purchase order were two different buildings, one of which had never been audited for anything. That isn't a paperwork problem. That's an unverified process making your product.

What the system has to actually cover on a pack line

Custom pack assembly is not one process. It's six or seven, and the QMS has to hold each of them under control.

Incoming cell inspection. Cells arrive with lot codes and date codes, and the inspection on receipt — capacity and internal resistance sorting data, open-circuit voltage, visual, packaging condition, storage state of charge, storage temperature records — is what makes a pack traceable later. If cells from several lots get mixed on one line and nobody records which went where, the whole downstream traceability argument collapses.

Workmanship control. Busbar torque with a calibrated driver and a logged value, weld pull or peel testing on the tab welds, ultrasonic weld parameters recorded per shift, harness routing to a defined standard, ESD controls at the bench. Whether the supplier works to IPC-A-610, IPC/WHMA-A-620, or an internal standard matters less than whether the acceptance criteria are written down, revision-controlled, and applied by an operator who can tell you what "reject" looks like.

Documented information. Which revision of the work instruction was in force on the day your pilot ran. Retention periods stated in years. Drawings and BOMs under change control, with the change record retrievable.

Nonconformance handling. Where suspect material goes, who authorizes rework, whether rework is documented, whether scrap is actually scrapped. I ask to see the quarantine area. Not the procedure for the quarantine area. The area.

Equipment calibration. Spot welders, torque drivers, power supplies, the end-of-line tester, thermal chambers, any measurement device used to accept or reject product, plus the gauge R&R behind the measurement system itself. Calibration stickers expiring next month on a line running your program are worth a conversation.

Corrective action. A CAPA log with real entries, root cause, containment, and a follow-up that confirms the fix held. An 8D that stops at D3 is a complaint log, not a corrective action system.

The document set to ask for before anyone quotes

Ask in the RFQ, not after. Ask before pricing, so the supplier understands the evidence is part of the deliverable.

The package I want in the first response:

• The certificate and its scope statement, plus the accreditation mark
• The quality manual, or if that's 60 pages of boilerplate, the process map and the level-2 procedures covering pack assembly
• The control plan, step by step, with characteristics, specification, measurement method, sample size and frequency, and the reaction plan for out-of-spec
• Work instructions with revision numbers
• Inspection criteria and sampling plan, including the AQL or equivalent for cosmetic and workmanship defects
• The calibration list for the line
• The traceability scheme, described in one page: how a finished pack serial number resolves back to cell lot, BMS firmware revision, operator, and test data
• End-of-line test specification with pass/fail limits
• One redacted nonconformance report and one redacted CAPA from the last twelve months

That last item does more work than the other eight combined. A live NCR includes handwriting, dates, initials, and a disposition. Boilerplate can be produced in an afternoon. A closed NCR with a real signature cannot.

Mapping the evidence to the sourcing stage

RFQ response. Certificate, scope, control plan, process map. I also ask which of these documents will be revised for our program and who signs the revision. A supplier who says "the control plan is generic, we'd write a program-specific one" has answered honestly and told me the generic one isn't worth reading.

Sample evaluation. First article inspection report against the drawing and the control plan, dimensional and functional. EOL test data for the actual samples in my hands, not a datasheet. Cell lot and date codes for the cells inside those samples. If samples were built on a bench rather than on the production line, say so — that's a real limitation, and it changes what the sample proves.

Pilot build. Now the control plan should be running in production mode, operators trained with records to show it, the PFMEA updated with what the pilot taught, calibration current. Ask for pilot yield and the NCRs from the run. The pilot is where I want to see a nonconformance, actually. A pilot with zero issues usually means nobody was looking.

Mass production. Evidence turns into monitoring here. Periodic EOL data, lot traceability records pulled on request, change notification for cells, BMS, connectors, and any component that touches form, fit, function, or safety. Surveillance audit reports on a schedule. A committed response time for a supplier corrective action request.

Ongoing. Certificate status re-checked annually, KPI review — defect rate, on-time delivery, NCR aging — and a plant visit on a cadence that matches the risk.

Where the gaps actually show up

Two I see more than any others.

The first is BMS firmware. The firmware that shipped isn't the firmware that was validated, and nobody can produce a revision log because the BMS vendor controls it and the pack assembler treats it as a black box. I want a firmware revision identifier per pack serial number, a record of what was tested with that revision, and a written commitment that a revision change triggers notification and re-validation. Then I ask how the assembler verifies the firmware loaded on the line — a checksum read-back, a version query over the communication interface, something. If the answer is "the operator selects the right file," that's an assumption sitting on the most critical control in the product.

The second is cell lot records. Packs built from mixed lots happen constantly in low-volume custom work, and that's fine if it's recorded. It stops being fine when a field issue surfaces two years later and the pack can't be traced to a cell population. The traceability page I asked for in the RFQ is the thing that makes an eventual failure investigation possible instead of theoretical.

Smaller ones: a control plan that lists process steps but no reaction plan, calibration that lapsed on a driver nobody uses anymore but that's still assigned to the line, and rework happening in a corner without a traveler.

Turning this into contract language instead of good intentions

Everything above is a request until it's in an agreement. In the RFQ, phrase it as a question that has to be answered in the quote: "State whether you will provide pack serial number to cell lot traceability for all units, and describe the record retention period." In the quality agreement, the same expectation becomes a shall with a number attached.

The clauses that do the work are the ones with a verifiable predicate. A retention period in years, not "as long as appropriate." A notification window in days, not "promptly." A defined trigger list for re-validation — cell change, BMS revision, a process change at the weld station, a site move. Records available for review on request, and a right to audit that survives a change of ownership at the supplier. I'm not interpreting what any of that means legally; that belongs to the people who draft it. What I care about is that each sentence can be answered yes or no by pulling a document.

Living system, laminated system

The tell isn't in the manual. It's in the friction. A living system has revision history with reasons attached, control plan changes dated to the program that caused them, NCRs closed inside a stated window, calibration records pulled from a database rather than a filing cabinet, and an operator who can explain a step without opening the binder. A laminated system has a pristine manual, a control plan referencing a part number that isn't mine, a CAPA log whose most recent entry predates the current year, and a quality engineer who answers every technical question by saying the person who knows is in another building.

One caveat I've had to learn the hard way: demanding the full package from a small, genuinely capable assembler can stall an RFQ for weeks or push you toward a worse supplier who happens to have prettier documents. Rank the items. Scope, traceability, and change notification are never optional — if those three aren't there, nothing else is going to save the program. The rest can be an improvement plan with dates, as long as the dates are in the agreement and someone is actually checking them.

The certificate gets the supplier through the door. The records decide whether they stay.