I sit on the finance side of the table, and here is the uncomfortable truth about cloud costs: by the time the invoice reaches finance, the money was spent weeks ago. The bill is a lagging indicator. Engineers made hundreds of small decisions in the meantime, each sensible on its own, and nobody saw the total until it was too late to change it.
Most mid-market companies don’t overspend because cloud is expensive. They overspend because nobody owns the spend. FinOps is how you fix that, and it doesn’t require a large team or an expensive platform.
Untagged spend is nobody’s spend.
Start with ownership
Untagged spend is nobody’s spend. Agree on a short tagging standard, owner, team, product and environment, and make it a rule that new resources aren’t deployed without it. Then show every team what they spend each month. In my experience, showback changes behavior faster than any policy, because people rarely waste money they can see.
Learn to speak both languages
Finance talks in budgets and variance. Engineering talks in instances, throughput and latency. Unit cost is the bridge: cost per customer, per transaction or per million requests. A bill that grows 20% while the business grows 40% is good news, and only unit cost makes that visible.
Take the quick wins in order of risk
- Delete what’s idle: unattached storage, orphaned snapshots, forgotten test environments
- Schedule non-production: a test environment that runs only during working hours is switched off for roughly two-thirds of the week
- Rightsize what’s consistently underused
- Move old data to cheaper storage tiers
None of these touch production, and together they usually fund the rest of the program.
Buy commitments against your floor
Reserved capacity and savings plans offer real discounts, but only on usage you’re sure will still be there. Commit against your steady baseline, ladder purchases over time rather than buying everything at once, and leave the spiky part of your usage on demand.
Look for the hidden line items
Compute gets the attention, but some of the fastest-growing costs hide elsewhere: data transfer, log retention, and especially monitoring and observability tools, which grow with every host and every log line. When our engineers reviewed the monitoring estate for a global ad tech client, retiring unused dashboards and consolidating three overlapping tools into one cut monitoring spend by about 15%. Nobody had looked, because it wasn’t “the cloud bill.”
Make it a monthly rhythm
One-off cost-cutting pushes fade within a quarter. What lasts is a cadence: anomaly alerts reviewed weekly, a 30-minute monthly review with finance and engineering in the same room, and commitment decisions each quarter. Tie the forecast to the product roadmap, so a planned launch shows up in the budget before it shows up on the invoice.
Your first 30 days
- Week 1: agree the tagging standard and enforce it on new resources
- Week 2: clean up idle resources and schedule non-production shutdowns
- Week 3: set budgets and anomaly alerts for each team
- Week 4: hold the first joint finance and engineering review
The goal isn’t the smallest possible bill. It’s a bill that everyone understands, someone owns and nobody is surprised by.