Most organisations get one good round of cloud savings from tidying up: idle resources, oversized instances, forgotten environments. The second round has to come from architecture, because that is what sets the shape of the bill.
Where the cost is actually decided
Data movement, retention choices, chatty service boundaries and always-on components designed for peak load usually account for more spend than instance sizing. None of them are visible in a cost dashboard as decisions — only as line items.
That is why cost work owned solely by finance plateaus. The levers sit with the people who choose storage classes, partitioning, caching and service granularity.
Make cost a design constraint
Treat unit cost — per tenant, per transaction, per report — as a non-functional requirement alongside latency and availability. Engineers optimise what they are asked to measure.
Attribute spend to the team that can change it. Shared, unallocated cost is nobody's problem.
Know which savings are structural
Commitments and discounts reduce the bill without changing the system; they are useful but finite. Architecture changes reduce consumption and keep reducing it as volume grows.
Sequence the two deliberately: take the commercial wins now, and put the structural changes on the roadmap with named owners.