The debt born at the till

Every product sold with a warranty carries a promise: if it fails within the covered period, repairing or replacing it is on you. That promise is not sales rhetoric — it is a debt. A debt of uncertain amount on the single unit, but surprisingly predictable on volume: if 4% of units will request service and each claim costs 90, every unit sold leaves 3.60 of future obligation in your pocket.

The warranty reserve is the accounting translation of this statistical truism: set aside, at every sale, the expected cost of future claims, so that the income statement of the year in which you sell also carries the full cost of what you sold. It is standard practice in structured accounts — accounting principles require it when the obligation is estimable — and it is almost always absent exactly where it would hurt most: in small companies, where a year of above-average failures lands on the accounts without warning.

This calculator does the maths with four numbers every manufacturer knows or can measure: price, warranty claim rate, average cost per claim and how much you accrue today. It returns the expected annual cost, the required rate in points of price and a dimensionless coverage index that says whether your reserve is uncovered, balanced or overabundant.

Warranty cost accrues at the sale, not at the failure

The heart of the model is a shift of perspective: the cost of the warranty is not born the day the customer calls support. It is born the day you collect the price. From that moment the obligation exists, it has a computable expected value — and the revenue you booked already contains it, whether you recognize it or not.

Whoever sets nothing aside is not saving: they are selling below cost without knowing it. The margin they see in the accounts is inflated by exactly the unrecognized warranty rate — 0.90 points of price in the reference case — and the difference will show up, one or two accounting years later, as a «bad year for service». It was not a bad year: it was the true price of the product, paid in instalments.

The number to know is w = f·C_r/P: the share of the price the product owes to its own quality. It is a rate, not an amount — which is why it can be compared across products, price lists, even companies of different sizes: 0.9% is a light weight; 3% is an industrial tax that deserves engineers, not just accountants.

The same factory, five failure rates

The table holds everything fixed — price 400, cost per claim 90, 20,000 units a year — and varies only the warranty claim rate. The product and the factory are the same in every row: only how often it comes back changes.

The linearity is perfect — every point of claim rate costs 18,000 a year — and this is precisely what makes the table a decision tool: any investment that removes one point of failure rate is worth 18,000 a year, forever, at these volumes. An extra test station, a better component, a packaging revision finally have a yardstick to be judged against.

There is also a reverse reading, a less comfortable one: the row you sit in is a measure of your quality expressed in money. Two manufacturers with the same price list and rates of 2% and 8% have, all else equal, 108,000 a year of structural cost difference — invisible in the quotes, inexorable in the accounts.

How this calculator works

The unit selling price is the net price at which you sell the unit covered by the warranty: it only serves to express the reserve in points of price, so use the actual average price, not the full list. The warranty claim rate is the share of units sold that generates at least one intervention within the covered period: your service system knows it; if one unit can generate several interventions, fold the subsequent ones into the average cost.

The average cost per claim is the most underestimated item: parts and labour, but also returns logistics, diagnosis of «no fault found» cases, case handling and — for replacements — the full cost of the new unit. The current accrual is the percentage of the price you set aside today at every sale: zero is an allowed value, and indeed the most instructive case to compute.

The units sold per year turn rates into money: they feed the expected annual cost and the shortfall, and it is on volume — not on the single unit — that the statistics become reliable.

How to read the results

The expected annual warranty cost is the debt one year of sales accumulates toward future failures: it is the number to compare with what you actually set aside. The required reserve rate is the same debt expressed in points of price: the part of the list price that was never yours.

The annual shortfall is the difference between what is needed and what is current: positive is a hole that will surface in future accounts, negative is a surplus of caution that has its own cost too. The coverage index ρ = r/w is the dimensionless verdict: below 0.5 the reserve is uncovered; between 0.5 and 1 it covers at least half of what is needed but not all; between 1 and 1.5 it is in balance; beyond 1.5 you are over-accruing, and the capital locked in caution should be returned to profit and investment.

The consumer twin

This site also hosts the extended warranty calculator, which answers the opposite question: is it worth buying a warranty? There the protagonist is the consumer, and the verdict compares the price asked with the expected value of reimbursements. Here the protagonist is the one who promises the warranty, and the maths is the same seen from the other bank.

The two perspectives illuminate each other: the margin that makes an extended warranty a bad deal for the buyer is exactly what makes it a profitable product for the seller. And indeed the natural step, for a manufacturer who has tamed their reserve, is to turn the warranty from cost into product: the paid extension is a policy whose industrial cost — f·C_r over the extended period — you know better than anyone else.

The boundary between the two questions is also the boundary of the model: this page computes the expected cost of the promise included in the price; the price of an extension must be built on top, adding margin and a loading for risk.

The quality-cost family

The warranty reserve is the everyday member of a wider family: the costs of non-quality, the ones the product generates after leaving the factory. The catastrophic member of the same family is the recall: the product recall cost calculator shows what happens when the defect does not arrive one unit at a time but all at once, and the scope is decided not by the defect but by how much you know about what you shipped.

The two tools complete each other: the warranty prices the statistical defect, the physiological one no process fully removes; the recall prices the systematic defect, and the traceability that limits its scope. A serious company watches both, because they move together: the same point of failure rate that inflates the reserve is often the first signal of the drift that, ignored, ends in a recall.

There is finally a close accounting relative: commercial returns. There too a percentage of what is sold comes back at a cost — but with no failure, out of second thoughts or wrong size. The accrual mathematics is identical; only the nature of the rate changes.

Where the claim rate comes from

The whole model rests on f, and the good news is that f is in your data: every service system records the cases, and it is enough to divide them by the units sold of the corresponding cohort. The important word is cohort: this year's claims belong to the sales of one or two years ago, and in a growing company the raw claims-to-sales ratio of the year systematically understates the true rate.

The second important word is disaggregation: the catalogue-wide average mixes mature products and recent launches, robust families and fragile ones. The reserve is computed per product family — and the comparison across families is often the first free industrial diagnosis: if one line has triple the rate of the others, the reserve knew it before the production manager did.

Finally the cost per claim deserves the same care as the rate: the average hides the tail of full replacements, and a mix sliding from repairs to replacements can double C_r at constant rate. Rate and cost are the two hands of the same lever — and the table on this page shows they multiply, not add.

Limits of the estimate

The model is deterministic on averages: it computes the expected value, not the distribution. On large volumes the law of large numbers defends it; on small volumes a single year can deviate noticeably from the mean, and the reserve should be topped up with a prudential cushion the calculator does not estimate.

It also assumes a stationary rate: it holds for products at steady state, not for launches — where the first-year rate can be a multiple of the mature one — nor for end-of-life runs with components running out. And it deals with the contractual warranty, not the legal conformity obligations that in many jurisdictions overlap and extend it: the duration relevant for f is the effective one of the promised cover, wherever it comes from.

A parametric estimate for information purposes: it supports the managerial sizing of the reserve and the comparison of scenarios, it replaces neither the applicable accounting standards nor your auditor's judgement on the liability to book.

The cost-per-claim ladder

The first table holds the rate fixed — 4% on 20,000 units at 400 — and raises the average cost per claim: from the walk-in repair to the full replacement with return logistics.

The failure rate is the same in every row: what moves the bill is only how much fixing costs. It is the half of the lever product engineers forget most often, because quality has a department and repairability does not: a product that rarely fails but repairs badly can cost more in warranty than one that fails twice as often and is fixed in ten minutes.

Design for service — replaceable modules, remote diagnosis that avoids the return, a refurbished circuit for replacements — acts on this column. And it has a precious property: while lowering the failure rate requires touching the product, lowering the cost per claim often only requires reorganizing the service process, and it works on the fleet already sold.

The coverage ladder

The second table photographs five accrual policies on the same product — a heavier case than the reference: a required rate of 2% — from those who accrue nothing to those who accrue one and a half times what is needed.

The most interesting row is the first, because it is the most widespread: no accrual does not mean no cost, it means 160,000 that will arrive anyway — but with no date, no line in the accounts and without the price ever having covered them deliberately. The difference between the first and the fourth row is not the money spent, which at steady state is identical: it is who decides when to recognize it.

The last row recalls that caution is not free: 80,000 a year of excess accrual is capital that finances no development, no discounts, no margin. The right reserve is not the biggest: it is the one that balances the debt, with the cushion your volume volatility justifies — and nothing more.

The accounting view: where the obligation comes from

The logic of this page is not a management invention: it is the same one accounting standards impose on structured accounts. Under international standards, warranties granted at the sale generate a provision when the obligation is probable and estimable — and the expected value over a portfolio of units is exactly the estimator required. Under US standards, product warranty follows the same road: the cost is recognized at the sale, based on historical claim experience.

On top of this contractual layer, many jurisdictions overlay a legal consumer protection — the European two-year conformity is the best-known example — which exists regardless of what the manufacturer promises. For the model only one number changes: the effective duration of the cover, and therefore the window over which f is measured. A manufacturer selling in several markets has, for the same product, different effective rates per country.

The practical point is one: the rate w is not a prudential whim but the photograph of an obligation that already exists. The calculator estimates it managerially; the accounts book it under the applicable rules; and the two figures, if the data are honest, must tell the same story.

Turning the warranty from cost into product

Whoever knows their own rate can take a step that competitors who do not accrue cannot afford: selling cover. The paid warranty extension is a policy whose industrial cost — f·C_r over the additional period — the manufacturer knows better than anyone else: the price is built on top of that cost, with a margin and a risk loading, and it is typically one of the highest-margin products in the catalogue.

The same knowledge allows the opposite move: lengthening the included warranty as a commercial weapon. If your rate is 0.9%, promising one extra year costs fractional points of price — and a competitor with double the failure rate cannot follow you without bleeding. The long warranty is advertising only quality can afford, and it is credible precisely because it is expensive to imitate.

In both cases the sequence is the same: first measure the reserve, then decide the strategy. Doing it backwards — promising durations the failure rate does not sustain — turns the marketing department into the company's main hidden creditor.

The industrial levers to lower the rate

The rate w = f·C_r/P has two attackable numerators, and the table above showed they multiply: halving one is worth as much as halving the other. On the failure rate work the factory testing calibrated on infant defects, the qualification of critical components, the packaging — a surprising share of first-month claims is transport damage — and the systematic analysis of returns, which turns every service case into a design datum.

On the cost per claim work the remote diagnosis that filters «no fault found» cases before they generate a return, the modularity that turns an hour of labour into five minutes of swapping, the refurbished circuit that cuts the cost of full replacements, and the choice of network — repairing close to the customer often costs less than repatriating.

The calculator gives each lever its price: at the reference, one point of rate is worth 18,000 a year and ten units of cost per claim are worth 8,000. These are the numbers to put next to the quotes of every improvement project — and the reason the warranty rate deserves a fixed line in the industrial review, not an annual remembrance.

Mistakes that inflate the bill

The first is the raw ratio while growing: dividing this year's claims by this year's sales. The claims belong to the past, smaller cohorts: the true rate is higher than what you see, and precisely when you grow fastest you understate it most.

The second is the catalogue-wide average: a single rate across families with different rates silently subsidizes the worst lines with the margin of the best, and hides the diagnosis. The reserve is computed per family, and the differences are free industrial information.

The third is the frozen cost per claim while the mix slides from repairs to replacements: C_r can double at constant rate, and the reserve ages without any indicator blinking. Rate and cost must be reviewed together, at least once a year.

The fourth is using the reserve as an earnings shock absorber: inflating it in good years and thinning it in lean ones. It is the fastest road to losing its meaning — and to finding yourself, the year the failures really arrive, with a reserve that existed only on average.

Frequently asked questions

What is a warranty reserve?

The accrual that recognizes, at every sale, the expected cost of future warranty claims: cost per unit e = f·C_r, i.e. claim rate times average cost per claim. With 4% and 90 per claim it is 3.60 per unit.

What is the required reserve rate?

The expected cost expressed in points of price: w = f·C_r/P. In the reference case it is 0.90% — the part of the list price that was never yours, because it belongs to future failures.

What is the coverage index?

ρ = r/w: what you accrue today divided by what would be needed. Below 0.5 the reserve is uncovered; between 0.5 and 1 it covers at least half but not all; between 1 and 1.5 it is balanced; beyond 1.5 you are over-accruing.

Why accrue if the failures have not happened yet?

Because the obligation is born at the sale: the revenue collected already contains the expected cost of the promise. Deferring recognition inflates today's margin and dumps the cost, as a surprise, on future accounting years.

What goes into the average cost per claim?

Parts, labour, returns logistics, diagnosis of no-fault-found cases, case handling and — for replacements — the full cost of the new unit. It is the most underestimated item: the average must be weighted on the real repair/replacement mix.

How do I measure the warranty claim rate?

Service cases divided by the units sold of the corresponding cohort, not of the current year: in a growing company the raw ratio understates the true rate. And it should be computed per product family, not catalogue-wide.

One unit can generate several claims: how do I handle it?

The model counts at most one claim per unit: subsequent interventions belong in the average cost per claim. Alternatively you can use expected interventions per hundred units as f, as long as it stays below 100%.

Can the annual shortfall be negative?

Yes: it means you accrue more than needed. The surplus is not free — it is capital taken from profit and investment to cover a risk already covered — and an index above 1.5 flags it as over-accrual.

How is this different from the extended warranty?

This page computes the cost, for the maker, of the warranty included in the price; the extended warranty calculator judges whether a consumer should buy an extra one. Same mathematics, opposite banks.

How does it relate to recall cost?

The warranty prices the statistical, physiological defect; the recall prices the systematic one, arriving all at once. They move together: a rising warranty rate is often the first signal of the drift that ends in a recall.

Does the model work for freshly launched products?

With caution: the first-year rate can be a multiple of the steady-state one, and the model assumes a stationary rate. For launches, use the rate of the most similar family and revisit the reserve at the first cohort data.

Is the reserve computed here the one to book in the accounts?

It is the managerial sizing of the expected value: the applicable accounting standards and your auditor may require different assumptions, horizons and cushions. The tool supports the estimate, it does not replace it.

Updated July 2026