Cloud cost optimisation for the enterprise estate.
Gartner estimates 30–35% of cloud spend is wasted. The organisations that recover it are not simply deleting idle instances — they are working seven distinct savings levers and wrapping them in FinOps governance so the savings hold. This note sets out the levers, quantifies each, and prioritises the sequence for a 25–45% total reduction.
There is no single lever that fixes cloud cost. Roughly 40% of the opportunity is technical — rightsizing and waste — and 60% is commercial: commitment coverage, licensing leverage and negotiation. Organisations that work all of them under FinOps governance reach 25–45% total reduction; those that run a one-off cleanup watch costs drift straight back.
01 Key findings
Waste is structural, not incidental. Gartner puts wasted cloud spend at 30–35% — idle resources, oversized instances and workloads on pay-as-you-go when committed-use pricing would halve them. On a $20M estate that is $6–7M a year with no business value.
The largest savings are commercial, not technical. Rightsizing and waste elimination capture perhaps 40% of the opportunity; commitment structuring, licensing leverage and negotiation deliver the other 60%. The 40%+ performers work both halves at once.
Commitment coverage is the master metric. The average enterprise commits only 55–65% of eligible compute; active quarterly management reaches 85–92%. That gap alone is 15–20% of total compute cost.
Default-off beats find-the-waste. Requiring every resource to justify its existence — decommission after a 14-day grace period — removes 10–18% of the estate within 90 days at near-zero performance risk.
Without governance, savings decay. Technical and commercial optimisation are one-time events; only FinOps — showback, tagging enforcement, anomaly alerting — stops the "optimise once, drift forever" pattern.
02 The savings levers
Cloud cost reduction is not one project but seven distinct levers, each with a different savings ceiling, effort profile and risk. Treating them as a portfolio — rather than fixating on rightsizing — is what separates a 15% cleanup from a 40% transformation. This note is the companion to our Cloud Contract Negotiation Guide.
| Lever | Mechanism | Typical saving | Effort | Risk |
|---|---|---|---|---|
| Commitment instruments | RIs, Savings Plans, CUDs against stable workloads | 35–72% on committed spend | Medium | Low |
| Licensing & BYOL | Azure Hybrid Benefit; existing Windows / SQL licences | 40–55% off compute | Medium | Low |
| Storage tiering | Lifecycle rules to infrequent-access / archive classes | 20–40% of storage | Low | Low |
| Rightsizing | Match instance size and family to real utilisation | 15–25% of compute | Medium | Medium |
| Idle & waste removal | Decommission orphaned resources, auto-shutdown dev/test | 8–15% of total | Low | Low |
| Egress economics | Negotiated egress waivers; architecture to cut transfer | 5–15% of network | Medium | Low |
| Architecture | Spot / serverless for interruptible and bursty workloads | Up to 60–90% on eligible | High | Medium |
03 Savings potential
Savings ceilings differ by an order of magnitude across levers. The chart shows the typical maximum discount each lever delivers against the spend it addresses — a reminder that commitment and licensing structure, not instance tuning, move the largest dollars.
Applied together across a $10M compute estate, disciplined commitment sizing alone typically returns $1–3M annually. See our AWS EDP, Azure Committed Use and Google Cloud CUD guides for provider-specific structuring.
04 The false-savings trap
The most expensive mistake in cloud optimisation is booking a saving that never materialises — or reverses within two quarters. Three patterns account for most of it.
Committing to consumption you cannot sustain. Aggressive commitment purchases look like savings on day one, but a Reserved Instance or CUD sized to a workload that later shrinks becomes stranded spend at full price. The same logic applies to migration credits that mask weaker long-run unit economics: the headline discount is real, the net position is not. Always model commitments against a conservative, not optimistic, utilisation baseline — and lock unit pricing before introductory credits expire.
Two further traps recur: rightsizing that is quietly reversed by application owners fearful of degradation, and waste "eliminated" on paper but never decommissioned because no owner will authorise deletion. Both are governance failures, addressed in section 06.
05 Waste categories
Waste is the least controversial lever: no performance risk, no architectural debate. Idle and orphaned resources typically represent 8–15% of total spend. A default-off policy — any resource not attributable to an active owner is tagged for decommission after 14 days — reverses the burden of proof and clears 10–18% of the estate in the first 90 days.
| Waste category | Typical % of total spend | Effort to eliminate | Risk level |
|---|---|---|---|
| Idle compute instances | 4–8% | Low | Low–Medium |
| Unattached storage volumes | 2–4% | Very low | Low |
| Orphaned snapshots | 1–3% | Very low | Very low |
| Oversized databases | 3–6% | Medium | Medium |
| Dev/test (no auto-shutdown) | 2–5% | Low | Low |
| Unused reserved capacity | 2–4% | Low | Low |
06 FinOps governance
Technical and commercial optimisation deliver one-time savings events. Sustaining them requires governance that fixes three structural failures which otherwise push costs back up.
- Showback and chargeback to the teams generating cost
- Cost visibility at workload and team level
- Make consumers feel the consequence of their choices
- Real-time cost visibility, not month-end reporting
- Automated tagging enforcement at provision time
- Anomaly alerting within hours of a spike
- Continuous optimisation, not periodic cleanup
- Quarterly commitment review as commitments expire
- Crawl → Walk → Run maturity progression
Commitment coverage rate is the single most predictive metric of FinOps maturity. Reaching the Walk stage — consistent tagging, workload-level attribution, quarterly commitment review — takes 6–9 months and delivers 20–30% reduction; the Run stage adds a further 10–15%. Organisations above 80% coverage of eligible compute run total costs 25–35% below peers stuck at 50%.
07 Prioritisation framework
With seven levers competing for the same programme capacity, sequencing matters. Weight each candidate action across four dimensions before committing effort.
Savings magnitude
How many dollars does the lever address, and at what discount? Commitments and licensing move the largest absolute spend; instance tuning rarely does.
Speed to capture
Waste removal and storage tiering realise in days; commitment restructuring and architecture take a procurement or engineering cycle.
Execution risk
Deleting an orphaned snapshot is near-zero risk; rightsizing production databases or moving to spot capacity is not. Weight risk against the owner anxiety it triggers.
Durability
Will the saving hold? A commitment sized to conservative utilisation and enforced by governance persists; a manual cleanup with no default-off policy decays.
08 Where to start
No coverage baseline yet and pressure for visible results. Impose a default-off policy and auto-shutdown dev/test — 10–18% of the estate clears in 90 days with negligible risk while you build the commitment case.
Eligible compute sits below 65% coverage. Baseline it, model RI, Savings Plan and CUD options against conservative utilisation, and open the account-team conversation using the commitment opportunity as leverage.
You have optimised before and watched costs return. Stand up showback, automated tagging and anomaly alerting, and move deliberately from Crawl to Walk to Run rather than running another one-off cleanup.
09 90-day roadmap
The sequence that consistently delivers fastest — 15–25% immediate savings while laying the governance foundations:
Days 1–30 — Visibility Foundation
Implement consistent tagging, deploy native cost tooling (AWS Cost Explorer, Azure Cost Management, GCP Recommender), and generate a first-cut workload cost attribution model. This alone surfaces the largest waste items.
Days 31–60 — Waste Quick wins
Decommission idle resources, auto-shutdown dev/test, and begin rightsizing the top 20% of compute spend. Low risk, no architecture debate, immediate realised savings.
Days 61–90 — Commercial Largest dollars
Baseline commitment-eligible compute, model RI and Savings Plan options, and open account-team discussions — using the commitment opportunity as the opening position. Add egress waivers via our egress negotiation guide.
Beyond 90 days — Sustain Governance
Institutionalise quarterly commitment review and continuous optimisation. Graduate from native tools to third-party platforms (Apptio Cloudability, Spot.io, Densify) only once the process to act on them exists.
Quantify your savings opportunity
Our Cloud & FinOps practice sizes the recoverable spend across every lever in a structured two-week assessment.
The Licensing Edge
Weekly cloud and licensing intelligence for enterprise IT leaders. 3,000+ subscribers.