Nobody decides to overspend on cloud infrastructure. It happens through reasonable decisions made in a hurry: a large database instance for a launch, a staging environment that never gets switched off, logs retained forever because nobody chose otherwise.
FinOps is the practice of making those decisions visible and owned. For a growing product, you don't need a dedicated team — you need a few habits and one person who cares.
Visibility
- Tag every resource with an owner, environment and product area. Untagged resources are the first place waste hides.
- Set budgets and alerts per environment, so a runaway job pages someone before the invoice does.
- Review a cost dashboard in the same meeting where you review product metrics, once a month.
Quick wins
- Schedule non-production environments to shut down outside working hours.
- Delete unattached storage volumes, old snapshots and unused load balancers.
- Set retention policies on logs and backups instead of keeping everything indefinitely.
- Move infrequently accessed files to cheaper storage tiers automatically.
- Right-size instances using actual utilisation data rather than launch-day guesses.
Architecture decisions that matter
Match the compute model to the workload
Spiky, event-driven work suits serverless functions. Steady, predictable load is usually cheaper on reserved or committed capacity. Mixing them thoughtfully beats choosing one for everything.
Watch data transfer
Moving data between regions, availability zones and out to the internet is easy to overlook and can become a major line item. Keep chatty services close together and cache aggressively at the edge.
Question managed-service defaults
Managed databases, queues and search clusters often default to generous sizes and high availability everywhere. Production may need that; development and staging usually don't.
Make cost part of engineering
The most effective change is cultural: include expected cost in design discussions, show engineers the cost of what they own, and celebrate savings the way you'd celebrate a performance improvement. When the people building the system can see what it costs, most waste never gets created.
Written by
DevTechnology Engineering