When should a dashboard show a median or a percentile instead of an average?

Average ticket is the classic offender

In most trades, jobs are not one population. Repairs cluster low and replacements cluster high, with very little in between. The average lands in the empty valley between the two clusters, describing a job that essentially never happens, and it moves whenever the mix moves even if neither type changed price.

The useful version is median ticket by job type plus the share of jobs in each type. Now a change is attributable: either the typical repair changed, the typical replacement changed, or the mix changed. Those are three different problems with three different owners, and revenue reporting should keep them separate.

Percentiles answer questions averages cannot

Operational promises are about tails. Average speed to answer looks fine while a tenth of callers wait long enough to hang up, because the abandoned calls are short and the answered ones are quick. The percentile version, the wait experienced by the slowest ten percent, is the number that predicts lost jobs.

The same holds for time to schedule, time from estimate to decision, and time from job completion to invoice. Ask what the worst tenth experienced, because that is where customers leave and where call intelligence finds recoverable revenue.

There is a practical reason tails matter more in service work than in most reporting: the tail is where the money and the churn live. The longest waits produce the abandoned calls, the slowest schedules produce the cancellations, and the largest jobs produce most of the revenue and most of the disputes.

When the average is still the right choice

• Anything you will multiply back out. Revenue per lead is a total divided by a count by definition; that is what makes it robust to attribution arguments.
• Unit economics. Cost per booked job, margin per truck day. These are ratios of totals, and medians would break the arithmetic.
• Small, tight distributions. When the spread is narrow the average and the median agree, and the average is easier to explain.

Show the shape once and the argument ends

A single histogram next to the headline number does more than any explanation. People immediately see whether they are looking at one population or two, where the mass sits, and how long the tail runs. Put it behind the tile, not in an appendix, and the debate about which average is correct usually resolves itself.

One caution on percentiles: they need volume to be stable. A ninetieth percentile computed on a few dozen records is essentially the second-worst value, and it will jump around week to week. Compute tail metrics on a trailing window wide enough to hold a few hundred records.