Cloud spend is growing faster than revenue and no individual team feels responsible. How should cost allocation be designed to actually change behaviour?

Why nothing changes without allocation

Unallocated cost is nobody's cost. Every team's individual contribution appears negligible against the total, no team is rewarded for reducing it, and the only actor with visibility — the platform or finance team — has no authority over the decisions that generate it.

This is a straightforward tragedy-of-the-commons structure, and it is not solved by exhortation, dashboards nobody owns, or an annual efficiency initiative. It is solved by attribution.

Making allocation possible

• Mandatory tagging at provisioning, enforced by policy so untagged resources cannot be created. Retro- actively tagging an existing estate is painful and partial, so this is the first and most important control.
• Ownership metadata on every resource: team, service, environment, cost centre.
• Shared cost allocated by a defensible key — the platform, the network, the observability stack, shared clusters — apportioned by usage where measurable and by an agreed proxy where not. The key matters less than it being agreed and stable, because arguments about the key consume the energy that should go into reducing the cost.
• Unit metrics, not just totals: cost per request, per active user, per transaction, per tenant. A team whose absolute cost rose 40% while traffic rose 80% is improving, and a total-only view punishes them for growth.

Showback versus chargeback

Showback — teams see their cost without being billed — is easier, uncontroversial, and produces real change because engineers respond to visibility. It is the correct starting point, and for many organisations it is sufficient.

Chargeback — the cost lands on the team's budget — is stronger and introduces real problems: teams optimise their own line item at the expense of the whole (declining to use a shared platform because its cost is visible while their own inefficiency is not), disputes over allocation keys consume time, and finance overhead grows.

Adopt chargeback only where budget authority genuinely sits with the teams, or it is a reporting exercise with extra friction.

What makes it work

• Cost visible in the tools engineers already use — in the dashboard next to latency and error rate, in the pull request that changes infrastructure. A separate cost portal is visited once.
• Cost as part of the design review, so the estimate exists before the commitment.
• Anomaly alerting on unexpected increases, routed to the owning team, which catches the accidental expensive change in days rather than at month end.
• Efficiency as a recognised engineering outcome, celebrated in the same way as a latency improvement. If cost reduction is never rewarded, it will never be prioritised regardless of how good the reporting is.
• A platform team that makes the efficient path the easy path — right-sized defaults, automatic shutdown of idle non-production resources, spot capacity available by default for suitable workloads.

The measurement makes the problem visible; the incentive makes it move. Organisations frequently build the first and omit the second, then conclude that FinOps does not work.