The broadcaster paying twice for an audience it no longer had

The situation

Streaming economics are unforgiving in a way broadcast never was. A linear channel costs the same whether one person watches or a million. A streaming platform costs more or less exactly in proportion to how much it is used, which means the finance function and the engineering function are now looking at the same number from opposite ends.

Most media organisations moved to the cloud for elasticity during live events, where demand genuinely swings by orders of magnitude within minutes. Elasticity was the correct reason. The trouble is that it justifies almost any level of provisioning if nobody revisits it.

The problem

This group's cloud spend had roughly tripled over three years. Viewing hours over the same period were broadly flat.

Finance had asked for an explanation twice and received two technically accurate answers that did not help: that costs had risen because usage patterns had changed, and that reducing spend would put resilience at risk. Both were true in places. Neither was actionable.

The obstacle was that nobody owned the bill. Cost sat with finance, who could not read it, and consumption sat with about forty engineering teams, none of whom saw what they spent. A team that cannot see its own consumption has no way to make a trade-off, so the rational choice is always to over-provision.

There was also a straightforward reporting problem. Roughly a third of spend landed in a shared account with no tagging, which meant it could not be attributed to anything. It was known internally as "platform" and functioned as a place costs went to stop being discussed.

What we did

We declined to start with an optimisation exercise, which is what had been requested, and started with attribution instead.

The argument was that cost reduction without ownership is a one-off. You can find savings by inspection, present them, and watch the run rate return within two quarters, because the conditions that produced it are untouched. The group had in fact done precisely this eighteen months earlier with a different consultancy.

Eight weeks went into tagging, account structure and showback, so that every team could see what it consumed. No savings were claimed in that period, which required some patience from a finance director who had been promised numbers.

What attribution exposed was more interesting than waste. Three full-scale environments were running continuously for services that had been decommissioned, retained because a launch had been planned and then cancelled, and nobody had connected the cancellation to the infrastructure. Their combined cost was close to a million pounds a year.

Encoding was provisioned for peak concurrent live events, permanently, although the peak occurred a handful of times a year. Storage held every intermediate rendition of every asset indefinitely, because deletion had never been anyone's job.

None of this was foolish. Each decision had been made by a competent person solving a real problem, and then the problem had gone away quietly.

The outcome

Annual run rate fell about thirty-four per cent with no reduction in service and no change to live event headroom, worth roughly £2.1 million a year.

More durable is that spend is now visible to the teams that cause it, reviewed monthly against consumption rather than against budget, with an owner for every account. The previous exercise saved money once. This one changed who is holding the question.

A cloud bill is not a bill for what you use. It is a bill for what you asked for, and organisations are much better at asking than at remembering they asked.