Blended cost-per-contact hides the customers who actually cost you

Ask a support leader what a ticket costs and you'll get one number, usually to the cent, usually wrong. Not wrong because the arithmetic slipped — total support spend over total contacts is easy — but wrong the way any average of wildly different things is wrong. Blended cost-per-contact is the mean of a distribution so lopsided that the mean describes almost no customer in it.

Start with the channel spread, because it's the easiest to see. Gartner put the gap starkly:

That's not a rounding difference; it's two orders of magnitude. A more recent benchmark tells the same story in dollars a controller would recognize:

Voice sits at the top of the assisted range:

So a blended figure isn't measuring efficiency. It's measuring channel mix. Move 5% of volume from phone to chat and your "cost per contact" falls — nobody got faster, nobody got better, the arithmetic just re-weighted. Two teams with identical unit costs in every channel can post blended numbers that differ by half, purely because one skews to voice and the other to self-service. Comparing those two figures, or celebrating a quarter-over-quarter dip in your own, tells you almost nothing about the work.

The channel blend hides a second trick underneath it. Cheap channels are cheap only when they succeed, and they mostly don't:

A self-service session that fails doesn't cost ten cents. It costs ten cents plus the phone call it becomes — and the customer arrives at that call more annoyed for having tried. Deflection accounting that books the cheap attempt and ignores the expensive escalation is how a "cost-saving" channel quietly raises your true cost-to-serve. The honest unit isn't cost per contact; it's cost per resolved issue, which folds the retries back in.

Then comes the segment problem, the one blended cost buries deepest. Contacts are not spread evenly across customers. A minority of accounts — on the wrong plan, never onboarded, running an integration that breaks weekly — generate a wildly disproportionate share of the volume. This is where the classic customer-profitability research lands hardest:

Read that against a flat blended cost and the distortion is obvious. Your average is a single point standing in for a curve where the top decile of contacters costs many times the median while a long, cheap tail barely touches support at all. Manage to the average and you'll under-invest in the accounts quietly funding everything and over-tolerate the ones quietly draining it.

None of this argues against measuring cost. It argues for measuring it where it actually varies. Three cuts get you most of the way. By channel, per resolved issue rather than per contact, so the self-service win is netted against its escalations. By segment or plan tier, so you can see which customers cost what they pay and which don't. And against revenue, always — a high cost-to-serve on a high-margin enterprise account is an investment; the same number on a free-tier user is a leak. The real cost of a ticket only becomes legible once you stop demanding one number for it.

The floor keeps dropping, too. AI now handles a slice of contacts at:

which makes the blend a moving target again: every point of automation drags the average down whether or not the underlying work got cheaper. That is one more reason to distrust the single figure. Blended cost-per-contact is a headline for a slide. Cost-to-serve — cut by channel and segment, and set beside what each cohort pays you — is the thing you can actually manage.

For the definition and current spreads, see cost per contact and the cost-per-contact benchmarks; the full benchmark library covers the wider operating context.