Should you measure average response time?

Averages fail here for a specific reason rather than a general statistical one. Response times are not distributed symmetrically around a middle: most enquiries are answered reasonably quickly by whoever happened to be at a desk, and a minority wait enormously longer because they arrived through a path nobody watches. A single number covering both groups describes neither, and it moves in the wrong direction when the business gets busier, which is exactly when the tail grows.

The published average B2B first response is roughly 42 hours, and that figure is more useful as an illustration of the problem than as a benchmark. Nobody's business consists of enquiries that consistently wait a day and a half. It is composed of many answered in minutes and some answered never, and a mean placed over the top of that mixture produces a number that matches no actual case in the sample.

The ninetieth percentile fixes this by reporting a case that genuinely happened. If the P90 is nineteen hours, one enquiry in ten waited at least that long — a sentence anyone in the business can act on, argue with, or test by looking. It also has the useful property of being hard to game: hiring one more person to answer the easy queue faster moves an average and leaves a P90 where it was, because the tail is produced by a routing gap rather than by capacity.

Non-responses are counted separately and kept out of the percentiles entirely, and that is not a technicality. An enquiry that never received a reply has no duration. Folding it in as an arbitrarily large number lets the choice of that number decide the result, in either direction, which is how a response-time report gets quietly rigged by someone with no intention of rigging anything. Reported honestly it is a count, sitting next to the sample it came from.

The practical version of all this fits on one line of a dashboard: P90 by source, and the number of enquiries with no reply, both by week. Two figures, neither flattering, both actionable. Anything with the word average in it can go.