AWS Trusted Advisor vs Compute Optimizer vs Cost Explorer: How Much Recoverable Waste Each One Actually Finds

Ask AWS how to cut your bill and you get pointed at three tools with overlapping names and different jobs: Trusted Advisor, Compute Optimizer, and Cost Explorer. Most comparisons list features. The question that actually matters is coverage: of the money you could realistically recover this quarter, how much can each tool even see? That framing changes the conclusion. The tools are good at what they look at. The problem is the waste categories none of them looks at, which for most accounts include the largest one. What each tool actually is Trusted Advisor is a checklist engine. Its cost-optimization category runs point-in-time checks: low-utilization EC2 instances, idle load balancers, unassociated Elastic IPs, underutilized EBS volumes, idle RDS instances, and a handful more. The catch everyone learns at signup: the full cost checks require a Business or Enterprise support plan; the free tier exposes only a sliver. It answers "what known-bad patterns exist right now", one check at a time, with no dollar rollup you can hand to finance. Compute Optimizer is a rightsizing recommender. It applies ML to utilization history (14 days by default, up to about three months if you enable enhanced infrastructure metrics, which is paid) and names target configurations: this m5.2xlarge should be an m5.large, this gp2 volume should be gp3, this Lambda is over-provisioned on memory, this autoscaling group is oversized. It's free, it's specific, and its blast radius is exactly compute-shaped: EC2, ASGs, EBS, Lambda, some container workloads. It will never tell you about an idle NAT gateway or an unattached volume, because that's not its question. Cost Explorer is spend analytics with two recommendation surfaces bolted on: reservation and Savings Plans purchase recommendations (its genuinely unique job: commitment coverage math from your usage curve) and a basic rightsizing view that overlaps Compute Optimizer's territory with less depth. It answers "where is the money going and what should we commit to", not "what's wasted". So the honest one-liner: Trusted Advisor finds a fixed list of hygiene problems, Compute Optimizer finds oversized compute, Cost Explorer finds commitment opportunities. Three different questions, all real. The coverage map, category by category Put the tools against a full waste taxonomy and the gaps get visible. Eight categories cover most recoverable spend: - Idle resources (running, zero consumers): Trusted Advisor covers the classics (EC2, RDS, load balancers) behind the support plan; the others largely don't. - Rightsizing (running, oversized): Compute Optimizer's home turf; Cost Explorer shallowly; Trusted Advisor a little. - Scheduling (running at 3am for no one): none of the three. There is no native check that says "this non-production fleet runs 168 hours a week and is used for 50". For most companies this is the single largest controllable category. - Orphans (billing with no parent): partial at best. Unassociated EIPs and underutilized volumes appear in Trusted Advisor; orphaned snapshots, abandoned multipart uploads, forgotten dashboards, and node-less clusters appear nowhere. - Commitment coverage (paying on-demand for steady load): Cost Explorer, genuinely well. - Storage class (hot storage for cold data): S3 has its own analytics; none of the three surfaces it as "recoverable waste with a dollar". - Data transfer (NAT for traffic that has a free path): nobody. NAT gateway processing charges hide in plain sight. - Kubernetes workload waste (requests double the usage): invisible to all three, because it lives below the instance boundary. Read down that list and the pattern is clear: native tooling sees maybe half the taxonomy, weighted toward the easier half, and the categories with the biggest dollars (scheduling, orphans at scale, K8s) are structural blind spots because each tool was scoped to one question. How to measure it for your own account Don't take coverage claims (including this post's) on faith; the audit is an afternoon: - Write down the eight categories above. - Run all three tools, export every finding, and bucket each into a category with its monthly dollar value. - Separately, do a manual sweep of two categories the tools skip: pull weekend-vs-weekday spend by account (scheduling opportunity) and run the classic orphan queries (unattached volumes, aged snapshots, idle gateways). - Compare: dollars surfaced by tools versus dollars surfaced total. That percentage is your personal coverage number, and it varies wildly by architecture. Teams that run this usually find the native trio nails commitment math and basic rightsizing, and that the majority of recoverable dollars sat in the unchecked categories. Whatever the split, you'll have a defensible answer to "the provider says we're already optimized", which is the sentence that usually triggers this whole investigation. FAQ Is AWS Trusted Advisor free? Partially. A small set of core checks is free for everyone; the full cost-optimization category (idle instances, idle RDS, underutilized volumes, and the rest) requires a Business or Enterprise support plan. If you're on basic support, the free tier's cost coverage is thin enough that it shouldn't anchor any conclusions about your waste. What's the difference between Trusted Advisor and Compute Optimizer? Trusted Advisor runs a fixed checklist of known-bad patterns (idle this, unassociated that) across many services, point in time. Compute Optimizer analyzes utilization history with ML and recommends specific target sizes for compute resources. One is hygiene, the other is sizing; they overlap only at the edges (low-utilization EC2). Should I use Cost Explorer's rightsizing or Compute Optimizer? Compute Optimizer for the analysis (longer lookback options, more resource types, risk scoring, specific targets); Cost Explorer for the commitment side (RI and Savings Plans purchase recommendations), which Compute Optimizer doesn't do. They're complements, not competitors. Do any AWS native tools find scheduling opportunities? No. None of the three flags always-on non-production resources as schedulable waste, which for most organizations is the largest single controllable category. Weekend-versus-weekday spend analysis per account is the quickest DIY proxy: a flat weekend curve on a business-hours company is the signal. How often do these tools refresh? Trusted Advisor checks refresh on the order of hours to a day (and can be refreshed manually); Compute Optimizer recomputes daily against its lookback window; Cost Explorer data lags the meter by roughly a day like all billing surfaces. None is real time, and none needs to be for this job. Top comments (0)