Recommendations

An always-on engine that scans your AWS, GCP, Azure and Kubernetes estate against 18 production-grade rules — each finding shipped with a dollar-denominated savings estimate, a one-click remediation plan and a confidence score.

18 rules, multi-cloud
USD impact per finding
Confidence scoring

Start free

← All features

lumicost.com — recommendations

By the numbers

Recommendationsin 4 numbers

downsize · terminate · delete · RI · schedule · rightsize · cleanup · grant-perm · enable-svc

recommendation types

IaaS · K8s · serverless · LLM

detection rules

CRITICAL · HIGH · MEDIUM · LOW

severity tiers

AWS · GCP · Azure · OCI

clouds analysed

How it works

Three steps. No agents. No surprises.

Connect read-only

Add an AWS account through a cross-account IAM role (read-only). GCP and OCI use the same pattern. We never gain write access to your accounts.

read-only IAM · STS:AssumeRole

Scan & rank

Twenty-one rules (idle, oversized, unused-disk, old-snapshot, RI opportunity, K8s, serverless, LLM …) score each resource against provider metrics + cost data, producing recommendations sorted by `severity × estimatedMonthlySavings`.

21 rules · 4 clouds

Apply or hand off

Apply or dismiss in-app — both events go to the audit log. Or generate a Jira ticket proposal: each ticket carries the runbook (problem · manual steps · CLI commands · expected impact · rollback).

apply · dismiss · Jira ticket

What's inside

An always-on engine that scans your AWS, GCP, Azure and Kubernetes estate against 18 production-grade rules — each finding shipped with a dollar-denominated savings estimate, a one-click remediation plan and a confidence score.

Rules engineered by people who got paged

Idle compute, oversized DB, orphan disks, unattached IPs, idle load balancers, low-utilisation commitments, expired snapshots, zombie clusters, leaked NAT gateways and more — every rule has a real-world incident behind it.

Confidence scoring stops false positives

Every finding ships with a 0–100 confidence score that factors in observation window, signal stability and cross-signal corroboration — so 'rightsize this' isn't based on a 7-day spike.

Remediation, not a backlog item

Each finding offers a generated remediation plan: Terraform diff, gcloud / aws CLI script, or a Jira ticket draft pre-populated with the impacted ARN/IDs.

Capabilities

Everything you need on day one.

Idle / always-on / low-CPU instance rules
Unused-disk + old-snapshot cleanup
Managed-DB rightsizing
Idle load balancer detection
Commitment opportunity (RIs / SPs / CUDs)
Kubernetes rules (5)
Serverless rules (cold-start · oversized memory · idle)
LLM rules (idle provisioned · expensive tier · high error rate)

Before vs after Lumicost

Night and day.

Without Lumicost

Trusted Advisor's generic checks
'$$$' priority — pick what you can
Recommendations rot in a backlog

With Lumicost

8 rules tuned per resource type
Sorted by severity × monthly impact
Hand off as Jira ticket with runbook

Where teams use it

Three ways teams put it to work.

First 24h after onboarding

Net-new accounts typically surface USD 8K–80K of low-risk savings on their first scan — orphan disks and idle load balancers lead the pack.

Quarterly waste review

Filter recommendations by team, sort by USD impact, freeze a snapshot for the quarterly business review.

Pre-renewal commitment audit

Before signing a new 1-yr Savings Plan, see which existing commitments are under-utilised and re-baseline accordingly.

From the field

“Eight focused rules with a transparent priority score beat a hundred opaque ones. You should be able to read why a recommendation exists.”

The Lumicost teamdesign principle

Plays nice with your stack

AWSGCPAzureOracle CloudKubernetesTerraformPulumiSlackJiraLinearGitHubGitLabDatadogSplunkOpenTelemetryPagerDutyNetSuiteSAPAWSGCPAzureOracle CloudKubernetesTerraformPulumiSlackJiraLinearGitHubGitLabDatadogSplunkOpenTelemetryPagerDutyNetSuiteSAP

FAQ