The capacity fade warranty clause that leaves your battery cost model exposed

A capacity-fade warranty is a financial input, not a quality statement. What moves your cost model is the retention threshold, the proration schedule, and the replacement remedy, read together against the duty cycle you actually run — not the headline percentage on the spec sheet. Change any one of those three and the same pack lands at a different cost per cycle, sometimes by more than the price gap between two competing bids. Most bid comparisons never get that far. They set dollars per kilowatt-hour next to a warranty percentage and stop there.

I keep one tab in the workbook that I open before anything else in a bid package. Three columns across the top: nameplate cycle life, warranted retained capacity, and the cycle count at which I expect to pull the pack off the robot. Price goes in the fourth column, and I fill it in last. Not because price doesn't matter — because until those first three are filled in, price doesn't mean anything.

The last bid I worked through, the warranty page was one page long. A short coverage paragraph, a small table underneath it, and a line of exclusions in type small enough that I had to zoom in. I read the paragraph the way everyone does, in about nine seconds. Then I spent the next forty minutes on the exclusions line, because that's where the model actually breaks.

The threshold is your replacement date, not a score

"70% retained capacity at 2,000 cycles or 3 years, whichever comes first." That last clause is the one that sets your capital schedule for the next five years. If the fleet runs roughly 250 cycles a year, the cycle limit doesn't bind until year eight. The calendar limit binds at year three. Two very different maintenance plans, one sentence apart.

The next question is when the calendar clock starts. Ship date is not installation date. If the packs sit in your receiving dock for a month because a line install slipped, that's a month of warranty you paid for and never used. It shows up in the model as a shorter effective coverage window, and if you're modeling a five- or seven-year horizon on a fleet, that month is real money amortized across every unit.

Proration is a discount rate on your recovery

A prorated warranty is a declining balance. Full coverage in the first year, then a schedule stepping down as the pack ages. The number that matters is not the coverage in month one — it's the coverage at the moment you'd actually be standing there with a dead pack in your hand, which is dictated by your end-of-life threshold, not by the warranty's.

So the recovery line in the model reads like this: replacement cost equals pack price plus freight plus labor plus downtime plus requalification, minus whatever the proration schedule returns at your swap point. If you model replacement at year four and the proration at year four returns forty percent of a pack that has itself risen in price since the original PO, you're carrying most of the event. That's fine. What's not fine is a model that quietly assumes full replacement because the warranty section said "covered."

The exclusions are your duty cycle in disguise

Fast charging. Over-temperature operation. Improper installation. Out-of-spec cycling. Every one of those words is doing work, and the work it's doing is defining the difference between the pack you bought on paper and the pack your fleet actually runs.

Walk the floor with the exclusion list and it stops being boilerplate. Where are the chargers, and what rate do they push? If the fleet opportunity-charges over a shift break at a rate the warranty caps lower, every pack on the floor is operating outside coverage from day one. That's not a legal question. That's an assumption in your cost model that just went to zero recovery.

Temperature is the same trap. A robot in an unconditioned warehouse in a cold-weather state, or a cell next to a heat source, is generating cell temperatures the warranty excludes above a threshold. And here's the part that catches people: the threshold is usually on cell temperature, which you cannot observe without a BMS log. You own an exclusion you cannot prove or disprove unless you were recording the right channels the whole time. I've watched buyers discover that the data they'd need to support a claim was never captured, which converts an exclusion from an argument into arithmetic. Zero recovery, no discussion.

How the warranty counts a cycle

Two warranties can both say "2,000 cycles" and mean wildly different things. One counts every charge event as a cycle. The other counts equivalent full cycles. If your model assumes 80% depth of discharge and the fleet actually tops up twice a day at 40% each, the first definition burns through the allowance at roughly double the rate. Same number on the page, half the coverage in practice.

This is why I treat the cycle definition as worth more than a percentage point of retained capacity. It's also the ask most likely to be granted, because it's not really a concession — it's an alignment. Buyers can put the intended depth of discharge, charge rate, ambient range, and cycles per day into the RFQ and ask the warranty to cover that profile specifically rather than a generic one. Suppliers who build packs for the duty cycle can usually write to it. Suppliers who don't, can't, and that tells you something useful before you've spent a dollar.

What the remedy leaves with you

A replacement pack is not a replacement cost. It's a component of one.

When the warranty remedy is a bare pack, the recovery covers the cell, the enclosure, the BMS, and nothing else. Freight in and out, the technician's hours, the requalification, the BMS pairing, the firmware version match, the safety check before that unit goes back into service — none of that is offset. And if the unit is down for half a shift, the downtime line doesn't get a warranty discount at all. It's a full-cost event.

So the model should carry two separate lines: the hardware cost, net of recovery, and the service cost, at full rate. In fleets where the machine is sitting in a revenue path, the second line usually dwarfs the first, and the warranty — however generous it looked in the bid tab — barely touches the total. That's the sentence that starts the argument in the conference room, and it's the right argument to have before the PO, not after.

Matching the end-of-life definition to the model

Your model probably replaces the pack at a capacity threshold driven by machine performance: run time falls short, cycle time slips, the charger can't keep up. Say that threshold is 80%. The warranty warrants to 70%. Between 80 and 70 sits a band of degradation you pay for and recover nothing on. It's small per unit and not small across a fleet.

There are two honest ways to handle it. Cost the band as un-warranted life and move on. Or ask for the warranty threshold to be set where your model actually swaps, and expect to pay for that in unit price. Both are defensible. What isn't defensible is a model that assumes a swap at 80% while the recovery calculation quietly runs to 70% — because that model will look correct for two years and then surprise you in year three, when the fleet's replacement wave hits and the recovery checks are smaller than the spreadsheet promised.

What I ask before the bid closes

Not contract language — that's for the people who do that work. These are commercial questions, and I want answers in the proposal, not in a side email, because a side email doesn't survive a change in account manager.

State the duty cycle in the RFQ — depth of discharge, charge rate, ambient range, cycles per day — and ask for the warranty to be written against that profile. Ask how a cycle is counted. Ask for the retention behavior as a curve or a small table rather than a single endpoint, because a curve lets you model a swap at 80% instead of guessing at it. Ask what the remedy includes besides the pack: freight, labor, advance replacement, core return credit, BMS pairing support. Ask when the calendar clock starts. Ask which data channels you're required to log to keep coverage intact, and make sure the BMS build actually records them.

And keep one thing straight in your head while you're doing it. A weaker warranty is not automatically a worse buy. I've priced small packs where the swap is twenty minutes, the spare machine is on the floor, and downtime genuinely costs nothing — in that case warranty coverage is worth very little and the cheaper pack without the paperwork is the right call. I've also priced packs buried in sealed enclosures on machines in a revenue line, where coverage that pays out on your actual profile is worth real money and worth paying for. The warranty isn't good or bad in the abstract. It's a value that depends on your downtime cost, your swap labor, and your ability to produce a claim.

Which is why I stopped comparing warranty terms side by side a while ago. I compare cost per cycle, net of expected recovery, at the swap point my own end-of-life threshold sets. That's the number that survives contact with the fleet. I've never seen a retention percentage win an argument in year three. I've seen a net-of-recovery cycle cost hold up.