# Cloud Cost Optimization Across AWS, GCP and Azure > How ZopNight finds and removes cloud waste: schedules for idle hours, evidence-backed recommendations, priced commitments and verified fixes on three clouds. Source: https://zop.dev/solutions/cost-optimization Published: 2026-07-01 · Updated: 2026-09-24 · Author: avinash-gaurav · Tags: zopnight, hub --- Optimization is a loop: see the spend, pick the lever that fits each resource, apply it, and check it stuck. ## Pick the lever that fits the resource No single tactic fits a whole estate. ZopNight sorts its findings into categories that each map to a different action: - **Schedule**: a resource that runs around the clock but is used on a pattern. The fix is a recurring start/stop window. - **Idle**: running but unused, such as a load balancer with no traffic. - **Rightsizing**: used, but provisioned larger than its measured load. - **Orphan**: left behind by something that no longer exists, such as a detached disk. - **Discount**: steady usage that a reservation, committed-use discount, Spot capacity or an owned licence would make cheaper. - **Compliance, security and reliability**: findings that are about risk rather than money but sit in the same queue. Each finding carries its current cost, the optimized cost and the evidence that triggered it. ## Savings that do not double count Two levers on the same resource cannot both be claimed in full. Stopping an instance at night and moving it to a smaller type overlap, so rollups count savings once per resource. The recommendations export states this, so a leadership summary is not inflated. Commitment recommendations are priced from the resource's real on-demand and committed rates, for one-year and three-year terms, and use measured uptime. If uptime cannot be measured, the saving is reported as unknown rather than assumed, and a commitment that would barely break even is not recommended. Schedule recommendations are priced from the hours the schedule would remove, at the resource's effective rate, including negotiated discounts for billing-connected accounts. ## Apply, then verify A finding can be acted on in the app through a remediation workflow, sent to Jira, or fixed by hand and marked applied. ZopNight keeps what you said (the status) apart from what it measured (the resolution). A recommendation marked applied is checked against a snapshot taken when it was raised; if the estate says otherwise, that is visible instead of silently trusted. Scheduling savings are measured the same way: the 24/7 cost of a resource minus what it actually cost with its schedule. ## Where to start The category pages go deeper: scheduling, idle detection, rightsizing and orphan cleanup. Commitment trade-offs are covered in scheduling vs reserved instances, and every rule family is listed under recommendations. ## Frequently asked questions ### Why is a savings figure missing on some recommendations? ZopNight does not invent numbers. If uptime or pricing cannot be measured, the saving is shown as unknown or the finding is marked advisory. ### Can recommendations run on a read-only account? Yes. Discovery, cost data and recommendations all work read-only. Applying a fix from ZopNight needs read + write access.