A logistics operator running AWS across more than forty countries had a cost problem it could see and not reach: a multi-account estate too sprawling for its lean platform team to interrogate, and cloud spend growing faster than anyone's ability to explain it. In Meridian's first month, conversational FinOps surfaced seven figures of annualized waste — every finding with its evidence attached, and no data pipeline built to get there.
The organization
The client is a global logistics operator with freight, warehousing, and last-mile operations spanning more than forty countries. Its AWS footprint grew the way logistics networks grow — by acquisition, by regional autonomy, and by speed-over-tidiness decisions made under seasonal peak pressure. The result is a sprawling multi-account estate: production workloads, regional applications, integration layers, and the long tail of accounts that every large organization accumulates and few can fully inventory.
Governing it all was a deliberately lean central platform team — strong engineers, but a headcount sized for enablement, not for manually auditing an estate of this surface area.
The challenge
The team knew waste existed; every estate this shape has it. What they lacked was throughput. Answering a question as simple as "which volumes are orphaned across the estate?" meant assuming roles account by account, running queries region by region, and assembling spreadsheets that were stale before they were finished. Cost Explorer showed the totals; it could not tell them which specific resources, in which accounts, owned by which teams, were the waste.
The conventional remedies each carried a cost the team did not want. A traditional FinOps platform meant standing up a data pipeline — exporting billing and inventory data into a vendor's warehouse, maintaining the feed, and waiting out an onboarding measured in months. Hiring a dedicated FinOps function contradicted the lean operating model. Doing nothing meant continuing to fund idle infrastructure across forty countries.
The deployment
CAELION deployed Meridian — its agentic FinOps, CloudOps, and security platform built on Amazon Bedrock AgentCore — inside the client's own AWS boundary. Meridian's fifty-plus integrations are read-only and use the same native APIs the team's engineers already trusted: Cost Explorer, CloudWatch, Compute Optimizer, EC2, and their peers across the service surface. Because Meridian reasons over live state through those APIs, there was nothing to build: no export pipeline, no data warehouse, no copy of the estate to secure. Connection to the multi-account estate was a permissions exercise, not an engineering project.
The operating change was the interface. Instead of writing queries and assembling spreadsheets, the platform team asked questions in plain English — where is my waste? which instances have been idle for thirty days? where are we paying on-demand for steady-state load? — and received evidence-backed answers in roughly five seconds, resolved to specific resources in specific accounts, with the utilization and billing data behind each conclusion attached.
The results
The first month was a systematic sweep of the estate's waste classes. Meridian surfaced seven figures of annualized savings, concentrated in four categories.
| Dimension | Before | After |
|---|---|---|
| Waste visibility | Totals only; no resource-level answers | Seven figures of annualized waste surfaced, with evidence |
| Time to answer a cost question | Days of manual account-by-account work | ~5 seconds, conversational |
| Data infrastructure required | Pipeline and warehouse assumed necessary | None built — live, read-only API reasoning |
| Optimization cadence | Occasional manual reviews | Continuous, standing capability |
The findings were unglamorous and lucrative, as estate-scale waste usually is: orphaned EBS volumes persisting long after their instances were terminated; idle compute fleets running well below any defensible utilization threshold; unused Elastic IPs billing quietly across dozens of accounts; and commitment gaps — steady-state workloads running on on-demand pricing that Savings Plans or Reserved Instances should have covered. Because every finding arrived with its evidence — the resource identifiers, the utilization history, the billing lines — the team could route remediation to workload owners without a re-verification cycle, and owners could not dismiss findings as tooling noise.
After the first-month sweep, the engagement settled into what it was designed to be: continuous optimization. New waste is caught as it appears rather than discovered at invoice time, and cost questions from leadership are answered in the meeting where they are asked.
Why it worked
Three decisions drove the outcome. Reasoning over live APIs instead of a replicated dataset meant time-to-value was measured in days and the answers were never stale. Read-only scope meant the security review was short and the platform team kept full control of every change — Meridian finds and proves; the team decides and acts. And the conversational interface removed the specialist bottleneck: an estate too large for any human to hold in working memory became interrogable by a lean team at the speed of a question.
If your AWS estate has outgrown your team's ability to interrogate it, CAELION can run Meridian against your live environment — read-only, inside your boundary — in a private briefing.
Client identity withheld by agreement. Figures anonymized and rounded from engagement reporting.