How to Build a Battery Pack Landed Cost Model That Wins Approval

Build the battery pack landed cost model bottom-up in seven buckets — cells, BMS, housing, connectors, assembly, testing, and freight — price each bucket from your own assumptions instead of accepting the supplier's lump sum, normalize every quote to the same volume, Incoterm, warranty term, and test scope, and put the two or three assumptions that actually swing the total on page one. That is the version that gets approved. A model that reproduces the supplier's quote format gets argued with, line by line, and handed back.

The version I remember building wasn't the polished one. It was the Tuesday afternoon before the review, three supplier PDFs open on a laptop at the end of a conference table, a half-empty coffee, and a workbook with seven tabs I'd renamed so many times the file said "FINALv4final." You're sitting next to me in that room, so look at the screen. Column A is the bucket. Column B is what the supplier quoted. Column C is what I think it actually costs. Column D is the gap, and Column D is the only column anyone in the review meeting will care about.

Here's what I learned building that thing, and the thing I'd tell you before you open Excel.

The quote in your hand is a price, not a cost structure

A pack quote arrives as a package: one unit price, maybe a tooling line, maybe a footnote about testing. That number is a margin structure wrapped around an unknown cost structure. Two suppliers can quote you almost the same unit price and be nowhere near each other on what's inside it — one buries the harness in assembly labor, one bills connectors separately at markup, one amortizes the test fixture across the order and the other charges it up front as NRE.

The fastest tell I know: ask the same supplier to price the same pack at two volumes. Watch which lines move and which lines vanish. Lines that vanish at high volume were NRE wearing a unit-price costume. Lines that barely move are material. Lines that move more than they should are where the margin lives.

So the model gets built bottom-up. You are not checking the supplier's math. You are reconstructing the pack from parts and minutes, then comparing.

What actually goes in each bucket

Cells

Cell price is the anchor, but the bucket is bigger than the cell. Add incoming inspection and grading, matching and sorting labor if the application needs it, attrition through pack assembly (a real percentage, not zero), first-build scrap, and the cell's share of the dangerous goods documentation. Model the cell as delivered-to-the-assembly-line, not as quoted-at-the-port. That distinction alone changes the ranking between a domestic distributor and a direct import in most models I've built.

BMS

The board, yes. Also the sense harness and its connectors, the end-of-line programming step, the firmware NRE if you're specifying behavior, the current-sense element, and — on higher-voltage packs — the contactor and precharge hardware that somehow never appears on the BMS line. This is also where supplier lock-in enters the cost model quietly: if the BMS is designed around that supplier's enclosure, your second source is theoretical, and theoretical second sources don't hold price.

Housing

Material and process, plus tooling. Injection mold, die-cast tool, extrusion die, however it's made. Add first-article inspection and any fixture the housing needs. Tooling is the line where a supplier's quote most often says "TBD" and where your model most needs a number, even a rough one.

Connectors

This is the bucket most frequently missing from a "pack price." High-voltage connectors, terminal lugs, busbars, service disconnect, IP-rated glands, the little sealed pass-throughs. On some packs this bucket is second only to cells. If a supplier's quote doesn't have a connector line, the connectors are in the unit price and you just can't see them — which means you can't compare them.

Assembly

Don't ask for the labor rate. Ask for the labor content — minutes by station — and multiply by a rate you believe, including burden. If a supplier won't give you minutes, that's not a data gap, that's an answer. Compare minutes across suppliers and you'll usually find the difference sits in how much of the pack is hand-terminated versus busbar-and-bolt.

Testing

End-of-line functional test, hipot, insulation resistance, capacity and IR grading. Then the part people forget: the test time is factory capacity. Whether the supplier prices it as a line or not, you're paying for the minutes on the tester. Model it as minutes.

Freight

Inbound cells and outbound packs are two different freight problems, and the outbound one is regulated. Class 9, UN 3480 for standalone cells and UN 3481 for packs with equipment, state of charge at or below 30% for air, UN specification packaging where it applies, and the hazmat fee that shows up per shipment rather than per unit. Ocean versus air is not a rounding difference on a pack with a high energy density. Neither is the surcharge structure, which changes by carrier and by lane and is worth a call to your forwarder rather than a guess in a cell.

The scope sheet is the actual normalization tool

Before the model is worth building, every quote has to be on the same basis. I use a one-page scope sheet: a row for each of the seven buckets, a column for Included, Not Included, and Priced Separately, and I send it back to every supplier with the request that they fill the third column. What comes back is usually the most useful document in the whole package.

Then I force four things to a common basis. Same annual volume and same order release pattern. Same Incoterm — DDP versus FOB moves freight, duty, and the DG paperwork, and comparing across Incoterms is comparing nothing. Same warranty term, because a two-year and a five-year warranty are not the same product. Same test specification, because one supplier's "tested" is a continuity check and another's is a full capacity sort.

Payment terms are the fifth thing, but they don't belong in landed cost. A 50% deposit is a working capital cost. Put it in the TCO sheet, not the landed cost sheet, or you'll double-count it when your finance reviewer does the carry calculation.

Certification: the cost that isn't in anybody's quote

No supplier quotes you the full certification bill, because most of it isn't theirs to quote. UN 38.3 T1 through T8 is the baseline for shipping. IEC 62133-2 for consumer. UL 2054, or UL 2271 for e-mobility, or UL 1973 and UL 9540A for stationary storage, depending on where the pack lands. The lab fee is the smallest part. The rest is samples consumed, engineering time to chase a failure, and schedule.

You can model this without a single live quote by treating it as a parametric block: a fixed certification cost per pack design, a sample count priced at your modeled unit cost, and a retest contingency expressed as a percentage of the block. That percentage should reflect your design change rate, not optimism. Change the cell vendor six months in and you may be back at T1 with a whole new sample set. The cert block belongs in the model with a note that says what triggers a retest, because that note is what turns a number into a risk.

Volume, tooling, and the amortization trap

Tooling is a step function wearing a straight line's clothing. The same mold that adds an uncomfortable amount per unit at five hundred pieces is almost invisible at fifty thousand. If you compare a low-volume quote against a high-volume quote, you've compared the amortization, not the pack.

So model volume bands — pilot, low-rate initial production, steady state — and amortize NRE over the volume you actually expect to take, not the volume printed in the RFQ. Then add a column for what happens if you take half. That column is the one the approvers read first, because it's the risk they're already carrying in their heads.

One more line in this section: tool ownership. If you paid for the tool and the supplier owns it, you've bought an option you can't exercise. Ownership, transfer, and maintenance responsibility belong in the model notes, not in a side conversation after award.

Warranty and field failure: where the award actually gets decided

These are two different numbers and conflating them is the most common mistake I see.

Warranty cost is expected claims rate times cost per claim — replacement pack, labor, inbound freight, diagnosis — plus whatever reserve you're required to carry. It's modelable. Ask suppliers for their field return rate by application; the ones with real data will share a shape of it even if they won't share the exact figure.

Field failure cost is the other thing. The failures that never become a claim because they become a customer escalation, a line-down, a redesign, or a recall. You can't put a point estimate on it credibly, so don't. Name it, weight it with a judgment call, and show the sensitivity.

Here's the comparison that ends meetings. Two suppliers, near-identical unit price, one at a fraction of a percent annual claims and one several times that. Put the claims delta into the TCO alongside the price delta. In most models I've built, the warranty line moves the ranking more than the unit price does, and almost nobody put it in the original spreadsheet.

The page that gets signed

One page, in this order.

A bucket table: the seven lines, plus certification, plus warranty reserve. Total landed cost at the target volume.

A "quoted versus normalized" column showing exactly what you added and what you pulled out, with a one-line reason for each.

A sensitivity block: the three assumptions that move total cost most. In most battery pack models, cells and volume are two of them before you even look at the third.

A short list of what you don't know and how you'd find out. Approvers trust a model that admits its blanks far more than one that's uniformly confident.

And keep the assumptions on the page, not in a hidden tab. A model whose seams are visible gets interrogated. A model that looks finished gets discounted.

One caveat, because it matters: a parametric model will be wrong on nearly every individual line. Its value is the ranking and the sensitivity, not the absolute number. If someone in the review asks whether the total is accurate to the dollar, the honest answer is no — and the useful answer is that it's accurate enough to tell you which supplier is cheaper once everything is inside it, and which assumption you'd need to be wrong about for that to flip. That's the question the approval is really asking.