Ask a streaming operator what a subscriber costs them and you get an answer in seconds. Ask what one hour of viewing costs them to deliver, and the room goes quiet.

We have watched that silence in a lot of rooms.

Your cost metric divides by the wrong thing

Cost per subscriber takes everything you spend on delivery: origin, storage, transcoding, egress, DRM, software licences, the people who keep it running and divides it by how many accounts are on the books. Accounts do not consume infrastructure. Hours do.

A subscriber who never opens the app costs you almost nothing. One who watches six hours every evening costs you a great deal. Cost per subscriber averages them into a figure that describes neither.

The relationship is not complicated:

cost per viewing hour × hours per subscriber = cost per subscriber

One operational variable, one behavioural variable, blended into a single number where neither can be seen.

It cannot tell success from failure

Your cost per subscriber rises 15%. What happened?

Your infrastructure could be running away from you. Or your content investment worked, engagement climbed, and you are now serving a considerably more valuable audience for slightly more per head.

Opposite conditions. The metric reports them identically.

Split it and the ambiguity dies in a line. If hours grew faster than cost, your cost per viewing hour fell, that is operating leverage, proved. If CVH rose while hours stayed flat, you have a delivery problem and you know where to look.

You cannot plan with it

This is the part nobody argues about, because almost nobody tries.

Entering a new market? Cost per subscriber cannot price it. Hours per subscriber there will not match your existing base, and that is precisely the unknown.

Adding live sport? Accounts barely move. Peak concurrency triples. Every capacity decision you now face runs off a curve the metric does not track.

Nobody has ever sized a CDN contract against subscriber count. Nobody provisions storage read performance against subscriber count. Those decisions are made in hours and concurrency, because that is what infrastructure responds to.

So every operator already plans in one unit and reports in another. Engineering sizes for peak concurrent streams. Finance reports cost per subscriber. Neither number can be derived from the other without the missing variable, which is why the two argue past each other every budget cycle.

Forecast your hours, multiply by your CVH, and you have a delivery cost you can defend before you commit to the market, the rights or the tier.

And it changes what growth does to the number. With cost per subscriber, extra viewing lands in the numerator and nowhere else, so the figure can only ever get worse. Cost per viewing hour puts those hours on both sides, so it reports efficiency rather than volume. Not automatically, add capacity in a step and CVH rises until utilisation catches up but at least the metric is now capable of improving when your audience watches more. Cost per subscriber is not.

The obvious objection

You can improve CVH by dropping bitrate. True and true of every unit metric ever built. Cost per available seat mile improves if you remove the legroom, which is exactly why airlines report it against load factor and punctuality. CVH means something only against a fixed quality floor: rebuffering ratio, start-up time, VMAF.

They do move together. On one operator platform we run, concurrent capacity went from 30,000 to 100,000 viewers on the same 55 servers, while that country’s national satisfaction survey moved the operator from 6.2 to 10 out of 10. That did not come from buying cheaper inputs. It came from changing how the workload runs.

The calculator

Free and anonymous. No login, no email, nothing to hand over. Seven monthly figures: cloud and servers, CDN, storage, processing, software licences, operations, and hours actually watched. Content rights are excluded, they are a programming decision, and folding them in makes the number unactionable.

A high figure does not mean a supplier is overcharging you. Leakage usually sits in architecture, not in invoices.

Run it against your current platform. A metric that can only ever flatter the company publishing it is not a metric, it is a brochure.

→ https://morescreens.com/calculator/

IBC2026, 11–14 September at the RAI, stand 1.F11. Bring your number.