Cloud cost work has a bad reputation as endless fiddling for marginal savings. Done in the right order, the first pass usually finds 20–40% with little risk.
Look at the shape of the bill first
Before touching anything, get a breakdown by service and by environment. Almost always a handful of line items — compute, managed databases, data transfer, and idle non-production environments — are most of the total. Optimise those; ignore the long tail.
Data transfer in particular is the cost that hides: cross-region traffic, chatty services, and egress to the internet add up quietly.
The cheap wins, in order
Turn off what nobody is using. Non-production environments running nights and weekends, orphaned volumes, old snapshots, over-provisioned instances.
Right-size, then commit. Match instance sizes to real utilisation, then buy savings plans or reserved capacity against the stable baseline — not before, or you lock in the waste.
Fix the architecture-level costs. Caching to cut database load, a CDN to cut egress, and lifecycle rules to move cold data to cheaper storage.
Make it someone's job
The reason bills creep is that cost is everyone's concern and no one's responsibility. A monthly fifteen-minute review of the breakdown, owned by one person, keeps the savings from silently coming back.