Why Quotes for the Same Robot Battery Pack Never Match

Quotes for the same robot battery pack never match because the RFQ leaves the expensive assumptions open, and each supplier closes them differently. Sample quantity, tooling amortization, test burden, and the duty cycle the pack is actually expected to survive move a quote far more than cell selection or margin does. An underspecified RFQ does not produce comparable bids. It produces several different products wearing the same part number.

I learned to sit still in front of that bid tab instead of reaching for the phone. Five PDFs open, one spreadsheet, a spread wide enough that the top and bottom numbers cannot be describing the same object. The first time it happened I assumed somebody was padding. Then I opened the RFQ I had sent and read it the way a stranger would read it, and the spread stopped looking like dishonesty and started looking like a mirror.

You are at that same table now. Scroll past the price column and land on the Notes field, the little grey box every supplier treats as optional. Two are empty. One says "standard pack, see attached" and attaches forty pages of datasheet for a product family. One lists a cell part number but no BMS model. One has four handwritten lines about test fixtures and a question about whether the enclosure is vented. Before you read a single number, the notes already tell you these five people are quoting five different scopes.

The spread is your ambiguity, priced five ways

A quote is a sum of decisions, and the RFQ decides which decisions the supplier gets to make alone. Every assumption you left open gets resolved in the supplier's own favor of effort — not dishonestly, just quietly, because nobody told them otherwise. The cell is rarely the reason two numbers differ by that much. The reason is that one supplier priced a pack that gets shaken, thermally cycled, and run to end of life on a bench, and another priced a pack that gets assembled and shipped.

That is why I stopped treating a wide bid tab as a negotiation problem. It is a document problem.

What has to be fixed before a number means anything

The electrical and mechanical envelope is the easy part, and it is also where most RFQs are vaguest in exactly the places that cost money.

Voltage window, not just nominal. A 48 V nominal pack is 13 series Li-ion, and the window runs from a full-charge ceiling down to a cutoff that has to be defined under load, not at rest. If I write "48 V nominal" and stop, one supplier sizes the BMS around a conservative low-voltage cutoff and another sizes it around a generous one, and the two packs hold different usable energy.

Continuous and peak current at a stated temperature. Forty amps continuous at 25 °C in open air is not forty amps continuous inside a sealed aluminum enclosure sitting over a drive motor at 45 °C ambient. Peak current needs a duration attached — a two-second boost and a ten-second boost pull different FET stacks, different busbars, and different fusing.

Duty cycle. I write it as a sentence a human can act on: this AMR runs two shifts, opportunity-charges between tasks at partial state of charge, and hits full depth roughly twice a week. A supplier reading that will size thermal margin and a BMS that can actually count coulombs through partial charging. A supplier reading "industrial use" will not.

Charge rate and charge profile. Whether the pack has to accept a fast charge during a break, and what the charger actually does at the top of the curve, changes cell choice and BMS behavior.

Communications, ingress rating, connector, enclosure. These are four lines in the RFQ and four different cost centers in the response. "BMS with CAN communication" is not a requirement. The message set, the baud rate, which values get broadcast, and whether cell-level voltages are readable over the bus are the requirement. IP54 in a vented box with a standard connector is a different build from IP67 potted with a sealed connector, and the enclosure material question usually shows up in the notes column three days later.

The three assumptions that quietly price themselves

When I audit a divergent bid tab now, I go looking for the same three things, in this order.

Sample quantity. If the RFQ says "provide samples" and nothing else, one supplier prices two units hand-built by an engineer and another prices a pilot lot off the line with fixturing. Those are not the same deliverable and they are not the same cost, and neither supplier is wrong.

Tooling amortization. If the pack needs a custom enclosure, a welded busbar fixture, or a potting nest, someone has to pay for it. Whether that cost is folded into the unit number, quoted separately as a one-time line, or absorbed because the supplier already owns the fixture, changes the comparison completely. Two quotes can be identical per unit and nowhere near identical in total.

Test burden. This is the one buyers forget to specify and then get surprised by. Does the quote include incoming cell screening, in-process hipot, an end-of-line capacity check, an abuse test on a first article, a transport test summary on the production configuration? Suppliers price what they normally do. If their normal is lighter than what your program needs, the quote looks competitive and the gap shows up later as a schedule problem.

Sample quantity is a plan, not a line item

This is where I see the most money wasted in both directions. Teams either buy one round of samples and try to validate everything on them, or they buy production-representative units too early and pay for tooling before the design has stopped moving.

I plan three tiers and say so in the RFQ.

Functional prototypes exist to prove the electrical design and the firmware conversation. Two to four units, hand-built, matching cell part number and BMS model but not necessarily matching the production assembly process. What you can validate here: voltage window under load, communications, current handling, thermal behavior at the duty cycle you specified. What you cannot validate: anything about the manufacturing process, and usually nothing meaningful about cycle life.

Pre-production samples are built on the intended process, with the intended fixtures, in small quantity. This is where ingress testing, shock and vibration, and connector retention become real questions. It is also the tier where a supplier's process problems surface — the ones that would have become your field failures.

Production-representative samples come last and exist for two purposes: a cycle-life run on the actual duty cycle, and a frozen configuration you can point at when the first production lot arrives. If cycle life is a threshold you have to defend, that run needs to start early enough that it is not finishing after the purchase order. Which means the sample plan has to be in the RFQ, so the supplier prices the right tier instead of guessing.

Duty cycle is where the cycle-life claim gets decided

A pack cycle-life number is not a cell cycle-life number. The cell datasheet figure assumes a depth of discharge, a temperature, a charge rate, and a cutoff that probably do not match your machine. When a supplier says a pack will last a certain number of cycles, that claim is only as good as the duty cycle you handed them.

So I hand them one. Continuous load as a percentage of capacity, the average discharge depth between charges, the ambient the pack lives in, the charge rate during opportunity charging, and how often it goes to full. Then I ask a direct question in the response form: what cycle life will you stand behind for this profile, and what test did you base it on. The answers separate the suppliers who have a cycle rig and have run industrial profiles from the ones reading a cell datasheet back to me.

The same logic applies to the charge profile. A BMS tuned for a full charge every night behaves differently from one managing partial charges all day, and if the RFQ never said which, both are defensible.

Tell them what the answer has to look like

The response format is not paperwork. It is how I keep five documents comparable when the suppliers are not.

I send a form, not a paragraph, and require it back filled in. Per line: cell manufacturer and part number, BMS make and model, firmware version quoted, the exact enclosure and connector, the ingress rating as tested, and the compliance evidence attached to the response rather than promised later. I want the transport test summary for the configuration being quoted, and I want to know which safety standard the pack was evaluated against and whether the file covers this build or a sibling.

Firmware version matters more than people expect. Two suppliers quoting the same BMS family with different firmware can have different behavior on charge taper, low-temperature cutoff, and what gets broadcast on the bus. If I do not capture the version at RFQ stage, I find out during integration.

I also ask two commercial questions in the RFQ itself, not to negotiate them there but to make them visible in the comparison: how the one-time engineering effort is treated — a single charge, tied to a revision, or bundled — and who owns the tooling and fixtures built for this pack, and whether they can move. Those answers tell me whether two quotes are actually comparable. The binding terms for both belong in the supply agreement that comes after selection, and that is where I leave them.

When a datasheet comes back instead of a filled form

It happens on almost every RFQ, and the temptation is to just read the datasheet and score it anyway. I do not. A datasheet describes a product family, not the pack I asked about, and scoring it against four completed responses corrupts the comparison for everyone.

I send it back once, with the specific lines that are unanswered highlighted, and a note that the response will be evaluated on the form. Sometimes it comes back filled in properly, which usually means the first submission was a sales habit rather than a refusal. Sometimes it does not come back, and that is information too — a supplier who will not complete a response form at bid stage is telling you something about how they will handle a change notice later.

What the second-round RFQ looks like

When I reissue, the spec section barely changes. What changes is that the open assumptions get written down. Sample tiers and quantities by tier. The duty cycle as a paragraph and as a test profile. The list of evidence that must be attached to the bid. The response form, mandatory. The tooling and one-time-effort questions stated plainly so nobody has to guess whether they are allowed to quote them separately.

The spread usually narrows on the second round, and not because anyone got cheaper. It narrows because five suppliers are finally pricing the same thing. The rest of the work — the validation, the audits, the agreement — gets easier once the numbers mean something. But the numbers only ever mean something because the RFQ made them mean something first.