A platform's cloud spend grows faster than traffic for six quarters, with no single team responsible. Why do cost-reduction sprints fail to hold, and what is the reframing that works?

Why sprints do not hold

Optimisation removes accumulated waste; it does not change the rate at which new waste accumulates. Six months later the spend is back, because the mechanism that produced it was untouched.

The mechanism is that the default direction of every individual decision is upward, and nothing pushes back: instances sized for a peak that has moved, environments outliving their projects, retention set to the maximum "just in case", cross-zone transfer nobody attributed to a design, storage accumulating because deletion requires someone to decide it is safe, and redundancy added incrementally by different people after different incidents.

None of these is a mistake. The aggregate is a cost curve that outruns growth.

The reframing

Unit economics rather than absolute spend. Cost per request, per active user, per gigabyte processed.

This matters because absolute spend should grow with the business, so "spend less" conflicts with growth and loses. "Improve efficiency" does not conflict, and it is measurable. The reframing is what makes the conversation productive rather than adversarial.

The controls, in dependency order

1. Attribution enforced at resource creation. Every resource tagged to a team and product; untagged resources blocked or flagged for deletion. Without this, cost is one number nobody owns and everything downstream is impossible.

2. Cost visible to the creating team, daily, in their own space. Not a monthly finance report. Most reduction comes from engineers seeing a number they did not expect.

3. Budgets with alerts owned by engineering. A team learning in week two that it is 40% over can act; learning next quarter, it cannot.

4. Automated waste reclamation — unattached volumes, idle load balancers, orphaned snapshots, non-production environments overnight. Automate it rather than raising tickets, because tickets do not get closed.

5. Cost as an explicit input to architecture review. New designs state expected cost at projected volume, which catches the expensive decision before it ships — orders of magnitude cheaper than after.

6. Retention defaults that expire, especially for telemetry, which grows with both traffic and cardinality and is the line item that most often doubles without anyone deciding it should.

The principle

Cost is a design property, like latency and availability. The objective is not to spend less; it is to make the cost of a decision visible to the person making it, at the moment they make it.