Cost Management and Governance
The economics
Cloud converts capital expenditure (buying hardware up front) into operating expenditure (paying for consumption). Total cost of ownership must include migration, training, operations effort, licensing and - the classic omission - data egress and inter-region transfer. FinOps runs in three repeating phases: inform (visibility and allocation), optimise (right-size, discount, eliminate waste), operate (budgets, anomaly detection, continuous accountability).
Purchasing instruments
On-demand pricing is flexible and dearest. Committed-use discounts (reserved instances, savings plans, committed use) trade a one- or three-year commitment for a large discount; convertible/flexible variants give less discount for the freedom to change family or region. Spot/preemptible capacity offers the deepest discount but can be reclaimed at short notice, so it suits batch, CI and stateless work only. Hybrid benefit schemes let you bring existing licences. Know that reservations are usually a billing discount, not a capacity guarantee - capacity reservations are a separate product.
Waste and allocation
Right-sizing (matching instance size to measured utilisation), autoscaling, scheduling non-production shutdowns, lifecycle tiering, and deleting unattached disks, aged snapshots and idle load balancers are the recurring savings levers. Tagging (labels) drives cost allocation; unenforced tag policies produce untaggable spend, so enforce them at creation with policy. Distinguish showback (reporting costs to teams) from chargeback (actually billing them internally).
Governance guardrails
Organise accounts/subscriptions/projects under folders or an organisation, apply service control policies or Azure Policy to restrict regions, SKUs and services, set quotas and budgets with alerts, and consolidate billing to capture volume tiers centrally.
Trap: assuming a stopped instance costs nothing (attached storage still bills), or that spot capacity suits a stateful database.
Rule of thumb: steady, predictable load → commit; spiky or interruption-tolerant → spot; unknown → on-demand plus a budget alert.