How do you reduce cloud costs without hurting reliability?
Short answer
Reduce cloud costs by first attributing spend to services and owners, then fixing the largest items: idle and oversized resources, inefficient architecture, missing commitments and data you no longer need. Track service levels throughout so savings never come at the expense of reliability, and put budgets, alerts and ownership in place so costs stay down.
Step by step
- Baseline spend by service, environment and, where possible, customer
- Name an owner for every significant line of the bill
- Remove idle resources, forgotten environments and unused data
- Rightsize compute and databases against real usage
- Fix architectural waste: chatty services, over-replication, expensive data transfer
- Buy commitments or savings plans for the stable baseline only
- Set budgets and alerts, and review spend with owners monthly
Measure uptime, latency and error rates before and after each change, and roll back any saving that costs reliability.
What savings are realistic?
It depends on how the platform grew. Platforms built fast during hypergrowth often carry large savings. At Remo I cut infrastructure and vendor spend by more than 80% over 14 months, normalised for user-hours in events, while holding ~99.99% uptime during event hours; and at Deepnote I have delivered substantial savings in both server infrastructure and AI token costs while usage grew. A baseline of usage and spend helps establish a realistic estimate for your platform.
At AdsDroid and Adigy.ai, where I am Fractional CTO, the changes I drove in my first couple of months roughly halved the AWS bill. As Ivan Finn put it, the annual saving effectively paid for a year of my involvement.
Within his first couple of months, James identified and drove changes that roughly halved our AWS bill. The annual saving alone effectively paid for a full year of his involvement with us.
Related questions
Should we move off AWS to save money?
Rarely as a first step. Most savings come from how the platform uses the cloud, and a migration is expensive and risky. Fix usage first, then decide.