What Your Own Warranty Is Actually Underwriting

Why this matters

When you write a workmanship warranty you are not offering a courtesy, you are writing an insurance policy and naming yourself as the carrier. The customer pays a premium inside your price whether or not anyone calculated one, and you pay every claim out of production capacity you cannot sell twice. Shops that have never counted the exposure are not running without insurance. They are running an insurance book with no reserve, no incidence data, and no idea how large the exposed population is on any given morning.

The premium is already in your price

There is no version of this where the warranty is free. If your labour rate covers the work and nothing else, then every warranty visit is paid out of the margin on some other job. That is still a premium, just an unpriced one collected at random from customers who did not have a failure, spent on customers who did.

The reason to compute it is not to add a surcharge. It is that an uncounted cost cannot be compared to anything. You cannot tell whether your warranty book is normal, improving or quietly doubling, and you cannot evaluate a longer term against a shorter one, because you have no unit to state either in.

Exposure is a stock, not a flow

This is the piece most shops miss, and it changes how the whole thing reads.

Your warranty cost is usually described as a flow: how many callbacks came in last month. But what you are carrying is a stock, the count of completed jobs still inside their warranty term at this moment. A shop finishing a steady twenty warrantable jobs a month on a twelve-month term is carrying two hundred and forty live obligations, every one of which can call. New work does not replace that stock, it adds to it, and the stock only stops growing when jobs start aging out at the same rate they are completed.

Two consequences fall straight out of the stock view.

First, a growing shop's warranty load grows twice: once because it does more work, and again because its stock has not reached steady state yet. A shop that doubled its volume nine months ago is carrying a warranty book that is still filling, and its callback count next quarter will rise even if its workmanship improves.

Second, term length multiplies the stock directly. Doubling the term doubles the number of live obligations. Whether it doubles the cost is a separate question, and the answer is usually no, for reasons the worked example below makes concrete.

Three numbers turn the stock into a cost

Exposed job-months. The stock times the months it sits there. Twenty jobs a month on a twelve-month term produces two hundred and forty jobs times twelve months, or two thousand eight hundred and eighty job-months of exposure across a year. This is your denominator and it is the number nobody has.

Incidence per hundred job-months. Warranty events divided by exposed job-months. Stated this way, incidence is comparable across shops of different sizes and across terms of different lengths, which a raw callback count is not.

Severity in hours per event. Total hours consumed per warranty event, counting travel, diagnosis, the repair, the parts handling and the office time to open and close it. Use the median and the mean separately. The median tells you what a typical event costs. The mean, which is dragged up by the tail, tells you what the book costs.

Severity is not one number, and the tail is where it lives

Most warranty events are small. A visit, a diagnosis, an adjustment, done. The book is not priced by those.

The tail is made of events with a multiplier attached: the fault that damaged something else, the one that needed a return trip because the part was not on the truck, the one where the original tech has left and the diagnosis starts from nothing, the one where the customer's schedule turns a two-hour job into three visits. A tail event running four to six times the median is ordinary. Two of them in a quarter will erase the arithmetic of a whole year of small ones.

That is why the mean matters for pricing and the median matters for scheduling, and why quoting a single average severity is a way to be wrong in both directions at once.

Worked example: a twelve-month term on a steady book

A shop completes twenty warrantable jobs a month, steady for two years, so the stock has reached steady state at two hundred and forty live jobs. Average productive labour per job is 8.0 hours, giving annual productive capacity of 1,920 hours.

Over the last twelve months the shop logged 36 warranty events consuming 130 hours in total. Mean severity is 3.6 hours per event; the median is 2.5, with the gap made entirely by four events that ran past 8 hours.

Where the events landed by age of job:

Two-thirds of the events, 24 of 36, arrive inside the first ninety days. That shape is the single most useful fact a shop can know about its own warranty, and it is invisible until events are dated against job completion rather than against the calendar.

The cost: 130 hours against 1,920 hours of productive capacity is 6.8 percent. That is the embedded premium. If the labour rate does not carry roughly 7 percent for this shop, the warranty is being funded out of the margin on jobs that did not fail.

Now extend the term to twenty-four months. The stock doubles, from 240 live jobs to 480, and annual exposure doubles from 2,880 job-months to 5,760. If cost scaled with exposure, the warranty would go from 6.8 percent of capacity to 13.6.

It does not, because incidence falls with age. Months 13 to 24 add 1,440 job-months per year. Taking the months 7 to 12 incidence of 0.28 per hundred job-months as the ceiling for the new band gives about 4 additional events a year, which is 11 percent more events for a doubled term.

Severity is where the naive version of this goes wrong. The 3.6-hour mean was derived from a population that was two-thirds inside ninety days, when the job is fresh, the tech who did it is still on staff, the parts are still on the shelf and the record is still detailed. Carrying that constant into month-twenty events is applying a coefficient outside the conditions it was measured under. Late events reliably cost more: the diagnosis starts cold, the original crew may be gone, and a fault that took eighteen months to surface is more often a real defect than an adjustment. Priced at 5.0 hours rather than 3.6, the four added events cost 20 hours, taking the book from 130 to 150 hours, or 7.8 percent of capacity.

So doubling the term raised the warranty cost by about a point of capacity, not by seven. That is a defensible commercial decision, and it is only defensible because the numbers behind it were separated.

What flips it. The whole result rests on incidence continuing to decay after month twelve, and that assumption fails in two situations. Where the equipment only runs in one season, a job completed in spring is not genuinely exercised until months eight through twelve, so the shop's own decay curve is measuring calendar age rather than run hours and the late band will be busier than it looks. And where the dominant failure mode is wear rather than installation error, incidence rises with age instead of falling, which inverts the model completely. Check which shape your own event dates make before you borrow this conclusion.

What the model does not underwrite away

The arithmetic above prices the promise you chose to make. It does not bound what the law may require of you, and the two are separate instruments.

Written warranties on consumer products are reached at the federal level by the Magnuson-Moss Warranty Act, 15 U.S.C. 2301 and following, administered by the FTC, which among other things prohibits a supplier who gives a written warranty on a consumer product from disclaiming the implied warranties that arise under state law. Implied warranties on goods come from Article 2 of the Uniform Commercial Code as enacted by each individual state, including merchantability at 2-314 for a merchant seller. Services on real property are usually governed by state common law rather than Article 2, and many states recognize some form of implied duty to perform in a workmanlike manner. Outer limits on old claims come from state statutes of limitation and, for improvements to real property, statutes of repose that commonly land somewhere in the six to ten year range depending on the state.

Which of those reach your shop depends on your state, on whether the customer is a consumer or a business, and on whether what you sold is characterized as goods or services. That is a question for your own attorney, and it is worth asking once rather than after a claim. The practical point for this article is narrow: your stated term sizes your service book, and your legal exposure is a different and longer-lived thing, so do not read a twelve-month number as a twelve-month horizon.

When the model disagrees with your gut

The most common disagreement is a shop convinced its warranty cost is trivial while the model says otherwise. Usually the shop is counting only visits that got coded as warranty. The uncoded ones are the tech who swung by on the way home, the adjustment made during a maintenance visit, the office hours spent on a dispute that never became a visit. Those are claims, and until they are coded the incidence number is a curated one rather than a rate.

The second disagreement runs the other way: a shop convinced warranty is eating it alive, whose events turn out to be concentrated in one job type or one crew. A book-level percentage answers a book-level question. If your incidence per hundred job-months splits by job type and one type carries most of it, the instrument you need is not a shorter term, it is a fix on that job type, and the sibling article on mining callbacks covers how to run that.

References

• Magnuson-Moss Warranty Act, 15 U.S.C. 2301 et seq., the federal statute governing written warranties on consumer products, administered by the Federal Trade Commission
• Uniform Commercial Code Article 2, including implied warranty of merchantability at 2-314, as enacted by the individual state whose law applies
• See related: The Warranty Reserve Most Shops Never Set Aside, Mining Your Callbacks for What They Reveal About Your Systems, How to Set a Warranty Period You Can Live With