The after-sales metrics worth tracking (and the ones that mislead)

Most after-sales dashboards measure activity. Claims opened this month, claims closed this month, average days to close. The numbers move, someone remarks on them, and nothing changes as a result.

A metric earns its place only if you can name the decision it changes. If the number goes up and nobody does anything differently, you are not measuring — you are reporting.

What follows is a working set of after-sales metrics: the formula, the decision each one drives, and — the part usually left out — how each one gets gamed once people know it is being watched.

What makes an after-sales metric worth tracking?

Three tests. Apply them before you add anything to a dashboard.

1. Is the denominator right? Most bad warranty metrics are counts pretending to be rates. "Claims this month" tells you nothing unless you know how many units are in the field.
2. Can someone act on it within their authority? A metric owned by nobody is decoration.
3. What behaviour does it reward? Every metric is an incentive. Assume it will be optimised, and check that optimising it produces the outcome you actually want.

The third test is the one that gets skipped, and it is why a dashboard can look healthy while the operation degrades.

Claim rate per units sold

Formula: claims received for a product line ÷ units of that line sold in the same cohort.

The word doing the work is cohort. Dividing this month's claims by this month's sales compares two unrelated populations: claims arriving now are mostly against units sold months or years ago. If sales are growing, that ratio flatters you; if sales dip, it makes quality look like it collapsed.

Track it by sale cohort instead. Take all units sold in a given month or batch, then count claims against those specific units, however long after the sale they arrive. The number matures over time — the January cohort's claim rate is not final until January's warranties expire — and that is fine. You are comparing cohorts against each other at the same age.

Decision it drives: whether a product line, supplier, or production batch is behaving differently from its peers. This is the metric that catches a bad batch.

How it gets gamed: by discouraging claims from being logged at all. If a front-line team is measured on claim rate, the cheapest way to improve it is to resolve things informally — a goodwill part handed over without a record. The rate falls, the cost stays, and you lose the failure data entirely.

Time to first failure

Formula: for each claimed unit, claim date minus sale date. Report the distribution, not the average.

This is the most diagnostically useful metric in after-sales and the one fewest businesses track, because it needs the sale date and the claim date joined by serial number. Without serial-level records you cannot compute it at all.

The shape tells you the failure type:

• Clustered in the first days or weeks — manufacturing defect, transit damage, or installation error. The unit was broken before the customer used it.
• Spread evenly across the coverage window — ordinary component wear-out, or a mixed population.
• Clustered near the end of the window — either a genuine wear characteristic, or customers noticing the expiry date and getting in before it closes.

That third pattern is worth sitting with: a spike in the final month of coverage is a fact about customer behaviour as much as about the product.

Decision it drives: where to intervene. Early failures point at the factory, the packaging, or the installer. Late failures point at design life and at how you price extended coverage.

How it gets gamed: by backdating. If registration happens late and someone enters an approximate sale date, time-to-first-failure becomes noise. Capture the sale date at the point of sale or accept that this metric is unreliable.

First-time-fix rate

Formula: claims resolved without a second visit or a second shipment ÷ total claims resolved.

Decision it drives: parts stocking, technician training, and diagnostic quality. A low first-time-fix rate almost always traces to one of three causes: the fault was diagnosed remotely and wrongly, the part was not on the van, or the technician had not seen this model before.

How it gets gamed: by redefining what counts as a second visit. A follow-up logged as a new claim rather than a continuation makes the rate look excellent and quietly inflates claim volume. Define the rule once, in writing: any return to the same fault on the same unit within a stated window is the same job.

Claim cycle time: why the mean lies

Formula: claim closed date minus claim opened date. Report the median and the 90th percentile. Do not report the mean.

Cycle-time distributions in warranty work are strongly right-skewed. Most claims are routine and close quickly. A minority stall — waiting on a part, on a manufacturer decision, on a customer who does not answer the phone — and those run for weeks.

An average blends those two populations into a number that describes neither. Worse, it can be improved by closing easy claims faster while the stalled ones get no attention at all.

Consider a deliberately simple illustration. Suppose nine claims close in 2 days and one takes 60:

• Mean: (9 × 2 + 60) ÷ 10 = 7.8 days
• Median: 2 days
• 90th percentile: 60 days

The mean of 7.8 describes no claim that actually happened. The median tells you the normal experience; the p90 tells you the worst experience a meaningful number of customers had. Those are two different problems with two different owners, and the mean hides both. The mechanics of where claims actually stall are covered in why warranty claims take so long.

Decision it drives: median drives staffing and process design. P90 drives escalation rules — what happens to a claim that has been open longer than your stated threshold, and who is told.

How it gets gamed: by stopping the clock. Claims put "on hold" pending parts, or closed and reopened, drop out of the measurement. If you allow a hold state, report held time separately rather than excluding it.

Cost per claim

Formula: total warranty cost over a period ÷ claims closed in that period.

The number is only as honest as the cost lines you include. At minimum: parts, labour, shipping both ways, and any replacement unit issued. The line most often left out is internal handling — the administrative time between the claim arriving and the work starting.

Decision it drives: whether to repair, replace, or refund by product line. For low-value, high-volume goods, the handling cost alone can exceed the value of the item, which is an argument for a replace-on-sight policy rather than a diagnostic one.

How it gets gamed: by excluding admin time, which makes repair look cheaper than it is and biases the repair-versus-replace decision.

Warranty cost as a percentage of revenue

Formula: total warranty cost over a period ÷ revenue over the same period.

This is the executive-level number, and it is genuinely useful for one purpose: trend across your own history. It is nearly useless for comparison against other businesses, because the cost lines people include vary so widely that two published figures rarely mean the same thing.

Decision it drives: pricing, accrual, and whether a product line is profitable after service.

How it gets gamed: by the denominator. A strong sales quarter reduces the percentage without anything improving in quality or service. Look at the numerator on its own alongside the ratio.

Repeat-failure rate

Formula: units with more than one claim ÷ units with at least one claim.

Note that the denominator is units, not claims. You are asking: of the products that failed once, how many failed again?

Decision it drives: two very different things depending on what the repeat looks like. Repeats of the same fault point at repair quality. Repeats of different faults point at a product that was marginal from the start, and at whether a replacement would be cheaper than a second repair.

It is also a fraud signal rather than a fraud verdict — a unit with an improbable claim history is worth a human look, for the reasons set out in how warranty fraud actually happens.

How it gets gamed: by not linking claims to the unit. If claims are filed by customer name or by date, repeats are invisible and the rate reads as zero.

Rejected-claim rate

Formula: claims rejected ÷ claims received.

This is the metric most likely to be misread, because it has no good direction. A rising rate could mean tighter verification is working, or that genuine customers are being turned away. A falling rate could mean quality improved, or that staff stopped checking.

Split it by reason and the ambiguity disappears:

Rejection reason	Rising rate means	Where to look
Out of coverage window	Expiry checking is being enforced	Fine, if the reason reaches the customer clearly
Serial not found	Registration is failing at the channel	Point-of-sale process, not the customer
Excluded by terms	Terms and expectations are misaligned	Marketing copy, packaging, terms wording
Damage not defect	Diagnostic disagreement	Technician guidance, evidence requirements
Duplicate claim	Duplicate detection is working	Confirm these are genuine duplicates

"Serial not found" rising is a warning about your own process, not about your customers. That distinction is invisible in a single headline number.

How it gets gamed: by leaving claims unresolved instead of rejecting them. An open claim that nobody closes is a rejection with better optics and a worse customer experience.

Leading and lagging: which metrics warn you early?

Lagging metrics tell you what already happened. Leading metrics tell you what is about to. You need both, but you should know which is which.

Metric	Type	Tells you about
Claim rate by cohort	Lagging	Quality of units already sold
Time to first failure	Leading, once the early cohort reports	A defect entering the field now
First-time-fix rate	Leading	Cost and repeat volume next month
Cycle time median	Lagging	Current process capacity
Cycle time p90	Leading	Complaints and escalations arriving soon
Cost per claim	Lagging	Margin already spent
Warranty cost % revenue	Lagging	Financial exposure over a period
Repeat-failure rate	Leading	Repair quality problems building
Rejected-claim rate by reason	Leading	Channel and terms problems

The most valuable early warning available to most businesses is time-to-first-failure on the newest cohort. If units sold three weeks ago are failing at a different pace than units sold three months ago at the same age, you know something now rather than at the end of the quarter.

A starting set, if you track nothing today

Do not build all nine. Four metrics computed correctly and reviewed monthly beat nine that nobody trusts:

1. Claim rate by sale cohort, per product line.
2. Time to first failure, as a distribution.
3. Cycle time median and p90.
4. Rejected-claim rate, split by reason.

Everything else is refinement. The prerequisite for all four is that claims are linked to specific units with known sale dates — which is the real reason most businesses cannot report on after-sales, and the subject of the warranty data you are not collecting.

---

Warranlytics computes these from the records it already holds: claims are linked to the unit and its registration date, so claim rate by cohort, time to first failure, and cycle-time percentiles come out of the same data you use to run the claim. See how a claim runs end to end, or look at the plans.