The savings number nobody can defend
Cost tools are generous. They sum every finding they produce, count the same dollars twice when two detectors reach them from different directions, and recommend actions the reader cannot execute — resizing a node an autoscaler owns, for instance. The resulting figure looks impressive until an engineer picks at it, and then the whole exercise loses credibility.
What the work involves
- Partition the bill so every line has an owner: usage types, clusters, namespaces, workloads.
- Reconcile findings against each other, so money counted by two detectors is claimed once. the audit where reconciling removed three quarters of the headline →
- Discard advice the reader cannot act on, and rewrite it as the action they can.
- State the range rather than the maximum, with the basis for each end of it.
- Write down what we could not evaluate and why — the gap list is part of the deliverable, not an embarrassment.
How the work goes
Collect
Read-only access to billing data and a metrics agent in the clusters. Fourteen days of measurement gives usable percentiles; less is guessing.
Reconcile
Overlaps removed, each exclusion recorded with its reason. This step usually shrinks the headline figure substantially, which is the point.
Report
A document with the findings, the sequence to act in, what each step is worth, and what it depends on. Yours to keep and to forward.
After the report
- Implementation of the ranked plan at a fixed price — each change as a pull request, measured before and after.
- A monthly cost review inside ongoing operation, measured by the MetrixForge agent where you keep it, or by the billing export where you do not.
Is this the right call?
A good fit when
- AWS estates with Kubernetes where the bill has grown faster than the traffic
- Anyone who has been handed a savings number they cannot defend internally
Not us when
- Estates without measurement access — we will not estimate an environment we cannot observe
- Clients who want the largest possible number rather than the defensible one
What we need, and what we will never ask for
What we need
- A read-only role in the billing account.
- Our agent with read-only RBAC and no privileged containers, where you keep it.
- A branch or fork we may open pull requests from.
- An engineer who can answer “why is this here?” — about an hour in total during the inventory, as on the diagnostic, and whoever reviews pull requests afterwards.
What we never ask for
- Administrator credentials in the cloud account or the cluster.
- Merge rights on your repositories.
- Your source code.
- A migration onto a platform of ours.
Questions
How is the audit priced?
Fixed price, quoted against a written scope before start. Anything outside that scope is a new short agreement rather than a variation.
How long does the audit take?
Fourteen days of measurement sit inside it, so the percentiles describe a real fortnight. The report follows the readout, and the date is agreed before start.
What access does the audit need?
A read-only role in the billing account, and our agent where you keep it. No administrator credentials and no merge rights — the block above lists it in full.