The bill that grows by itself
There's a pattern that repeats in nearly every company that migrates to the cloud: the first few months the bill is reasonable, but from then on it grows quarter after quarter even when the business doesn't grow at the same pace. When someone finally asks "why are we paying this?", nobody has a clear answer.
The reality is uncomfortable but hopeful: in most infrastructures we audit, between 25% and 35% of the spend is pure waste. And the best part: cutting that share doesn't require rewriting anything. These are configuration, governance, and discipline decisions. Let's break it down.
The five holes where the money leaks out
1. Oversized resources
The absolute classic. Machines and databases provisioned "just in case" at two or three times what they actually use. The occasional-peak rule gets applied to the size of every day. Telemetry usually tells a different story: CPUs averaging 8%.
2. Non-production environments… running 24/7
Development, staging, and demos left on nights, weekends, and holidays. A pre-production environment used only during business hours is wasting more than 70% of its hours.
3. Forgotten storage
Disks that outlived their machines, snapshots from two years ago, copies of copies, eternal logs sitting in the most expensive tier. Storage looks cheap per gigabyte… until you add up terabytes of things nobody will ever read again.
4. Data transfer
Data egress from the cloud is one of the least understood and most expensive line items. Architectures that cross regions or download data daily pay a constant toll that nobody consciously decided to pay.
5. Orphaned resources
Reserved IPs with nothing attached, load balancers without backends, test services someone spun up in March. Each one costs little; together, a serious figure over a year.
Savings levers that don't touch code
These actions are purely operational and typically recover that 25-35% within weeks:
- Data-driven rightsizing: review 30 days of actual usage and adjust sizes to what's observed, not what's feared. Start with the 10 most expensive resources — that's where 80% of the savings live.
- Scheduled shutdown of environments: development and staging on only during working hours. It's literally programming a calendar, and the savings are immediate and recurring.
- Usage commitments (reserved instances / savings plans): for stable production capacity, a 1-year commitment usually discounts 30 to 40%. One caveat: rightsizing first, reservations second — never the other way around, or you commit to the waste.
- Storage lifecycle policies: automatic rules that move old data to cold tiers and delete what has expired (snapshots, logs). Configure once, save forever.
- Orphan cleanup: a quarterly pass over IPs, detached disks, and empty load balancers. Half a day's work, permanent savings.
- Budget alerts: notifications at 50, 80, and 100% of the monthly budget. They don't save money by themselves, but they turn the quarterly surprise into a weekly conversation.
What does require engineering (and also pays off)
To be fair: the next level of savings does involve technical work — real autoscaling instead of fixed capacity, spot instances for interruption-tolerant workloads, serverless architectures for irregular loads, or simply shutting down services that no longer deliver value. It's not where you start, but it's where long-term savings consolidate.
A one-week plan to audit your bill
- Days 1-2: enable cost breakdown by service and by tag. Without visibility there's no savings, only intuition.
- Day 3: identify the 10 most expensive resources and compare their size against their actual 30-day usage.
- Day 4: inventory non-production environments and their real usage schedules. Program the shutdowns.
- Day 5: storage and snapshots — lifecycle policies and deletion of expired data. Review orphans.
- Following week: with consumption stabilized, evaluate reservations for the stable baseline and configure budget alerts.
So it doesn't happen again: minimum viable governance
One-off savings evaporate within two quarters if nobody governs the spend. You don't need a FinOps department: mandatory tagging (every resource with an owner and an environment), a 30-minute monthly review of the bill covering the top movers, and budgets with alerts per team or project are enough. Light discipline beats the grand plan that never gets executed.
Want a second opinion on your bill?
Cloud cost audits are among the most direct-payback work we do: in one or two weeks you get a report with every source of waste identified, how much it costs, and how to eliminate it — starting with the levers that don't touch code. If your bill has been growing for months without explanation, let's talk: there's very likely a 30% waiting to be recovered.