Azure FinOps for Real Teams: Cutting Cloud Spend 20-35% Without Slowing Delivery

Cloud cost optimization works when treated like engineering: measure, change one variable, confirm impact, and prevent regressions. This article focuses on Azure, but the FinOps mechanics apply anywhere: allocate costs accurately, target the highest-impact services first, and put guardrails in place so savings do not disappear next sprint.

Start with allocation. If your bill is one shared subscription with no owners, you cannot manage spend. Implement a tagging standard and enforce it via Azure Policy. Minimum tags that work in practice: cost_center, app, env, owner, data_classification.

Example baseline from a two-subscription setup at ~$28.4k/month: Compute $12.6k, Data (SQL/Cosmos/Redis) $9.1k, Storage $3.8k, Network/egress $2.2k, everything else $0.7k. Your numbers will differ, but the pattern is common.

• Right-size VMs using 14-30 days of metrics; target 40-60% average CPU as a starting point.
• Schedule non-prod shutdown (nights/weekends) and measure weekly savings.
• Enable storage lifecycle rules (hot -> cool -> archive) for logs and exports.
• Remove orphaned public IPs, unattached disks, old snapshots on a cadence.

Realistic savings: a shutdown schedule (10 hours/day weekdays + weekends off) commonly reduces dev/test compute cost by 45-65% when teams previously ran everything 24/7.

Commitments are powerful but must be data-driven. Use them for stable workloads: baseline AKS nodes, steady SQL compute, or long-running app tiers. Avoid committing on spiky workloads or architectures in flux.

resource "azurerm_consumption_budget_subscription" "monthly" {
  name            = "prod-monthly-budget"
  subscription_id = data.azurerm_subscription.current.id
  amount          = 30000
  time_grain      = "Monthly"

  time_period {
    start_date = "2025-01-01T00:00:00Z"
  }

  notification {
    enabled        = true
    operator       = "GreaterThan"
    threshold      = 80
    contact_emails = ["finops@company.com"]
  }
}

In the $28.4k/month example, an 8-week plan achieved ~$7.1k/month reduction (~25%): $3.4k from non-prod scheduling/right-sizing, $2.1k from compute commitments, $1.0k from DB tier adjustments after query/index tuning, and ~$0.6k from storage lifecycle + cleanup. The durable win is preventing rebound via tagging + budgets + monthly review.

• Run a 30-minute monthly FinOps review: top movers, anomalies, and upcoming commitments.
• Track unit cost when possible: cost per customer, per 1k requests, per GB processed.
• Treat cost regressions like performance regressions: find the change, fix it, add a guardrail.