Warranty Claim Rate and Approval Rate Read Together
Why this matters
Warranty numbers are the only quality signal most shops get that somebody else verifies. That is their value and it is also the trap. A claim measures what a manufacturer agreed to consider, not what failed: every failure you absorbed as goodwill, every failure just out of term, and every failure on a unit you did not supply is invisible to it. So the claim count moves for two unrelated reasons, and one of them is your own filing discipline. Hire an office person who files properly and the count doubles in a quarter with nothing whatsoever happening in the field.
Read on its own, either number will mislead you. Read as a pair, they separate a failing part from a failing practice, which is the one distinction that changes what you do on Monday.
What the two numbers are actually made of
The claim count is claims filed inside the window, anchored on the filing date. Not on the failure date, not on the install date. A failure the customer reported in March and the office filed in May sits in the second quarter.
The approval share is claims approved over claims resolved, where resolved means approved or denied. Claims still open are in neither side. That is the correct construction, and it is also the one that creates the problem in the last section.
Days to resolution is an average over resolved claims only, by the same logic and with the same consequence.
Three silent exclusions are worth having in mind before you read any of them. Goodwill repairs never appear. Failures on equipment you did not supply never appear even when you serviced them. And a claim the manufacturer never answers sits open indefinitely, so it stays out of the approval share and out of the resolution average for as long as the manufacturer wants it to.
Why the claim rate is not really a rate
The standard construction divides claims filed this quarter by installations completed this quarter. Those are two different populations. The claims come from work you did months or years ago; the denominator is work you did in the last thirteen weeks. In a growing shop the denominator inflates and the rate falls while nothing improves; in a flat quarter after a busy year it spikes the same way.
Two honest alternatives:
Read the count as a trend against your own trailing quarters. Four quarters of history and a plain count is more informative than a ratio of two unrelated populations, and it is what the cases below do.
Or build a real cohort rate, if your records carry install dates on the equipment. Take the units of a given type installed in one quarter, and count how many of that same cohort have claimed by a fixed age - 90 days, one year. Always compare cohorts at the same age, because a cohort's claim count only ever goes up, and comparing a two-year-old cohort against a six-month-old one will tell you the old work was worse every single time.
The pair, and what each combination points at
| Claim count | Approval share | Most likely cause | The cut that confirms it |
|---|---|---|---|
| Up | Holding or high | A part, a batch, or a supplier change | Component name, then install date range |
| Up | Falling | Installation or commissioning practice, or claims going in thin | Denial reasons, split into excluded-cause and missing-evidence |
| Down | Falling | Filing discipline: only the obvious ones get sent, and even those are weak | Claims filed against comebacks logged on covered work |
| Down | Holding | Genuinely fewer failures, or you quietly stopped filing | Same cut; the answer is whether comebacks fell too |
The two middle rows are the ones this card is about, because they share a direction on the count and point at opposite causes.
The floor before you read the share. A quarter with fewer than about 20 resolved claims will not support a trend read on approval share; at 15 resolved, two denials move it by more than 13 points. Below that, roll two quarters together and say you did. Note that this floor is counted in resolved claims, which is a different unit from the completed-job floors used by the sibling cards on repeat visits and checklist coverage; do not carry one across to the other.
Case one: claims up, approvals holding
A shop's claims filed by quarter over the trailing year: 9, 11, 10, 12. Average 10.5 a quarter. This quarter: 27, about 2.6 times the trailing average.
Approval share has run in an 84 to 89 percent band of resolved claims across those four quarters. This quarter, 24 of the 27 filed are resolved: 21 approved, 3 denied, 87.5 percent of resolved claims approved. Twenty-four resolved clears the 20-claim floor, so the share is readable.
The count jumped and the manufacturer is still paying. That combination says the claims are good claims - the failures are real, they are covered, and the paperwork is fine. The problem is upstream of the shop.
The confirming cut. Nineteen of the 27 claims name the same component. Of those 19, 17 were installed inside one six-week stretch. The shop installed 96 of that component in the same six weeks, so 17 of 96 units from that cohort, 17.7 percent, have already claimed, against roughly 2 percent for the same component in earlier cohorts read at the same age. That is a batch.
What the shop does. Stop installing from that lot today. Go to the supplier with the 19 claim numbers and the install date range rather than with an impression. Then work the list of customers who have a unit from that cohort and have not called yet, because the failure rate says a number of them are about to, and a planned visit costs a fraction of an emergency one and keeps the review.
Case two: claims up, approvals falling
Same shop, a later year. Claims filed by quarter over the trailing year: 12, 14, 13, 15. Average 13.5. This quarter: 26, about 1.9 times the trailing average.
Approval share has run in an 82 to 88 percent band of resolved claims. This quarter, 22 of the 26 are resolved: 12 approved, 10 denied, 54.5 percent of resolved claims approved. Twenty-two resolved clears the floor.
Same direction on the count, opposite direction on the share, and the cause is now inside the shop rather than inside the supply chain.
The confirming cut, on the denials. Of the 10 denials, 7 cite an excluded cause tied to how the unit was installed or commissioned - a clearance not met, a start-up record absent, a condition the manufacturer's terms put outside cover. 3 cite missing evidence: no serial recorded, no photograph of the failed part, no dated record of the original installation.
Those are 7 of 10 denials and 3 of 10 denials, not shares of the 26 claims, and they need two different fixes. The seven are a field practice problem: something in how the work is being done or recorded at commissioning puts the failure outside cover before it ever happens. The three are a paperwork problem, and paperwork is the cheaper of the two to fix - a required field at install and a photograph at failure. Fixing the paperwork first is tempting because it is easy, and it will move the approval share by three claims while leaving the actual defect running.
What the shop does. Read the seven excluded-cause denials and write down the condition each one names. If they all name the same condition, it is a missing commissioning step, and the fix is to put that condition on the install checklist as an item that records a measured value rather than a tick, so a year later you hold a dated record of what was actually measured, which is the evidence a claim gets argued on rather than proof the condition was met, and it does not replace whatever start-up documentation the manufacturer's own claim terms require, so record both. See related: A Checked Box Is Not a Passed Check. If the seven name several different conditions, it is more likely that commissioning is being rushed as a whole, and the fix is in how long that job type is scheduled for rather than in the template. Handle the three evidence denials in the same week, because they are quick, but do not report the improved approval share until the seven have an owner.
A denial is worth reading even when you accept it. Somebody outside your shop, with no reason to be kind, has looked at one of your jobs and named what was wrong with it. Most shops file, get denied, absorb the cost and never open the reason code again, which throws away the only independent audit of field practice they get for free. Note that absorbing it is usually the obligation rather than a gesture, because a manufacturer's denial settles only what the manufacturer owes, while what you owe the customer is set by your own workmanship warranty and your state's implied-warranty law, so a denial is never something to hand a customer as not covered. Reading the reason codes once a quarter, in a batch, is cheaper than any inspection programme and it is the half of this subject that nothing else in the numbers can reach.
Days to resolution only describes the claims that closed
The resolution average is computed over resolved claims, and the claims still open are the hard ones. That is survivor bias, and it means the figure looks best exactly when your queue is worst. See related: Time to Invoice Only Counts the Invoices You Sent, which owns this pattern.
The specific consequence here is worth printing rather than describing. Take case two's quarter:
- 22 resolved claims at an average of 19 days: 22 times 19 is 418 claim-days.
- 4 claims still open, sitting today at 74, 61, 55 and 48 days: 238 claim-days already accumulated, and none of them finished.
- A floor on the eventual average across all 26 claims: 418 plus 238 is 656, over 26 claims, 25.2 days.
So the published 19 days over 22 resolved claims understates the eventual figure over all 26 by at least 6.2 days, and that is a floor rather than an estimate, because the four open claims can only get older. The gap is not a rounding issue; it is a third again on top of the number being reported.
Note also that the resolution average and the approval share are computed over the same resolved set, so they degrade together. A quarter where the manufacturer is stalling on the difficult claims shows a fast resolution average and a high approval share at the same time, which reads as a good quarter and is the opposite.
The pair worth publishing
Publish two figures side by side, every period, and the second one is what stops the first one being gamed:
- Average days to resolution, over resolved claims, labelled as such on the face of it.
- The age of the oldest open claim, in days, as of the reporting date.
The second number cannot be improved by leaving anything open. It gets worse every day a hard claim sits, which is precisely the behaviour the first number rewards. In case two those read 19 days and 74 days, and a shop looking at that pair asks the right question immediately: what is happening with the claim that has been open two and a half months, and is anybody chasing it.
Add the floor calculation above whenever the open count is more than about a fifth of the claims filed, since at that point the resolved-only average is describing a minority of the quarter's work.
References
- See related: Tracking Callbacks to Find Your Real Warranty Cost, and Warranty Management, for what to track and where the money actually sits
- See related: Capturing the Evidence a Warranty Approval Requires, and The Warranty Claim Paper Trail, for the three missing-evidence denials above
- See related: Time to Invoice Only Counts the Invoices You Sent, which owns the survivor-bias pattern in a resolved-only average
- See related: A Checked Box Is Not a Passed Check, for recording a commissioning condition as a value rather than a tick
- Manufacturer warranty terms and claim submission requirements, which define the excluded causes behind the practice denials