Cloud Cost Recommendations: How ZopNight Finds Waste
Every finding names a resource, the measured evidence that fired it, and the lever that fixes it, or it does not ship.
A ZopNight recommendation is the output of one audit rule evaluated against one discovered resource in your AWS, GCP or Azure estate. Rules are pre-computed on a schedule and stored, so the Recommendations page reads instantly rather than querying your clouds while you wait. Each record carries a title, the current monthly cost, the optimized cost, the potential saving, the cloud account, and an evidence block holding the monitoring values that triggered the rule.
How many rules
As of 6 October 2026 the rule registry holds 684 rules: 273 for AWS, 209 for Azure, 181 for Google Cloud, 20 for Snowflake and 1 for Microsoft 365. Of these, 113 are wired to the remediation workflow behind the Remediate button; schedule findings open the schedule creator instead, and the rest are advisory. The count changes as rules are added and retired, so treat it as dated.
The categories
Findings are filed into categories so you can work one kind of problem at a time:
- Idle: a resource is running but nothing uses it.
- Rightsizing: a resource is provisioned beyond what it measurably needs.
- Schedule: a resource has a clear off-hours pattern and should follow a start/stop window.
- Orphan: a resource is detached, unreferenced, or has lost its parent.
- Discount: a commitment, Spot capacity or an owned licence would price the same usage lower.
- Compliance: configuration misses a best practice such as encryption, MFA or backups.
- Security and Reliability: posture and workload-health findings, most of them from the Kubernetes workload rules.
Concrete or abstain
The engine’s standing rule is that a recommendation either names a real fact or says nothing. A resize with no nameable target, a saving with no priced rate, or an idle finding with no resolvable lever abstains instead of emitting a guessed number. Schedule recommendations go further: they are generated only from the latest 21 days of metrics and refuse stale or gapped data, and when a saving cannot be honestly claimed they stay visible as a no-dollar advisory rather than inventing one.
From finding to fix
Where a rule supports it, a Remediate button runs a wizard: a precondition check, an admin approval step for destructive operations, the cloud action itself, then validation. A central safety gate demotes risky levers, such as a one-shot stop on a VM or a delete without proof the resource is unused, to advisory, and advisory findings show a numbered “How to Fix” playbook instead. Schedule-class findings open the schedule creator with the resource prefilled. Any finding can also become a Jira ticket that stays in sync with the recommendation’s status. Remediation needs a read + write cloud account; a read-only account still gets every finding.
Adoption is measured, not assumed
Marking a recommendation Applied records your claim. ZopNight separately re-reads the resource against a baseline frozen when the finding was raised and records whether the change actually happened, abstaining when no honest verdict exists. Each finding keeps a history timeline of when it was generated, viewed, applied, dismissed or reopened.
By cloud
The full rule-by-rule reference lives in the integrations directory: AWS rules, GCP rules and Azure rules.
Every one of the 684 rules is also covered in the ZopNight rule catalogue: 447 check pages, grouped by category and by cloud, each with the reason it fires and the fix.
Questions we get a lot.
Does ZopNight need write access to produce recommendations?
No. A read-only cloud account gets discovery and every recommendation. Start/stop, scheduled execution and remediation require the account to be connected with read + write permission.
Why did a recommendation disappear?
Recommendations are re-evaluated on every sweep. A finding closes when its rule no longer fires, and schedule findings on an estate whose metrics have gone stale close automatically and reopen when fresh data arrives.
Can I export the list?
Yes. The filtered list exports to an Excel workbook with an executive summary sheet and a detail sheet. Summary savings are de-duplicated per resource so overlapping levers are not double-counted.