practice

Cost Governance

The guardrails, budgets and approval structures that prevent cost surprises without becoming a bottleneck.

governancebudgetsguardrailsanomaly-detectionpolicy

Definition

Cost governance is the set of controls that bound spend structurally, rather than relying on people noticing. Its design problem is the same as any governance: preventive controls that are too strict get routed around, and detective controls alone arrive too late.

Preventive controls

  • Policy blocking the most expensive mistakes: instance types above a threshold requiring approval, approved regions only, no resources without an owner tag, no public data transfer without review.
  • Quotas per account or team, so a runaway process is bounded by construction rather than by vigilance.
  • A paved road of pre-approved, correctly-sized templates that are the easiest path — so the default choice is also the sensible one.
  • Expiry by default on non-production resources.

Detective controls

  • Anomaly detection on spend, alerting within a day. Month-end discovery is far too late; a misconfigured job can spend a quarter's budget in a weekend.
  • Budget alerts at thresholds, to the team, not only to finance.
  • Continuous unused-resource detection, reported to owners.
  • Commitment coverage and utilisation, so purchased capacity is actually used.

The balance

The failure at the strict end is real and common: approval required for every resource produces a queue, and teams under deadline pressure build shadow infrastructure that is both more expensive and invisible. The failure at the loose end is a bill nobody predicted.

The workable position is preventive controls on the expensive and irreversible; detective controls on everything else; and a fast, credible path to an exception.

What must be in place first

Allocation. Governance without attribution cannot target anything — budgets cannot be set, anomalies cannot be assigned, and every conversation stalls on ownership. Mandatory tags enforced at creation are the precondition for all of the above.

Failure scenarios

  • Approval gates on everything, producing shadow infrastructure.
  • Alerts to finance only, so the people who can act do not see them.
  • Budgets with no consequence, so they are ignored.
  • Anomaly detection tuned so loosely that a doubling of spend is within tolerance.
  • Governance without a paved road, so compliance is an obstacle rather than the default.

Interview question

"A misconfigured job spent a quarter's budget over a weekend. What controls would have prevented or caught it, and what would each have cost in developer friction?"