Acme AI Platform Sample Report

Observed GPU spend

$104,000From the latest report

Recoverable, estimated

$9,960/mo9.6% of spend · low bound $8,020

Approved for change

$2,760/mo1 decision

Bill-verified to date

$6,400/moConfirmed on the April invoice

Where the $9,960 stands

Estimated $1,800
Owner review $5,400
Approved $2,760

Next: Ship A100 research pool so the next invoice can verify it.

Decisions

Select a finding to inspect its evidence. Estimates require review.
Owner	Est. / mo	Stage
Maya Chen	$2,760	Approved
Maya Chen	$3,100	Owner review
ML platform	$2,300	Owner review
FinOps	$1,800	Estimated
Inference	—	Needs evidence
Research	—	Needs evidence

Spend by source

May 2026 · USD

AWS$68,000

Google Cloud$36,000

Bill-verified savings

monthly rate, by invoice

DecJanFebMar$6,400AprpendingMay

Recent reports

Executive Summary Optimization Plan

Illustrative report · not customer evidence

From finding to owner decision.

Inspect the complete assumptions, prepare an engineering handoff, and explore a separate prior verification example.

Review a sample decision

Selected decision

A100 research pool idle windows

Estimated opportunity

$2,760/mo

Provider: AWS
Owner: Maya Chen

Illustrative evidence and assumptions, provided to demonstrate the review workflow.

Make the change reviewable

Medium · Maya Chen

Proposed change

Confirm that the observed idle windows are representative.
Test a bounded schedule or allocation change on the named non-production workload.
Record the implementation date and retain the previous configuration.

Validate and recover

Compare queue delay and completed jobs with the baseline; then reconcile a complete later billing period for the same resource scope.

Rollback: Restore the previous schedule and allocation if jobs miss their performance target.

Check before acting

Idle capacity is required for a scheduled experiment or latency target.
A commitment keeps the billed cost unchanged after reducing usage.

From observation to outcome

Baseline: 5/1/2026 – 5/31/2026. a100-research-pool, experiment-runner, checkpoint-eval

What would make this decision stronger?

Performance evidence
Attach queue depth, latency, and the workload performance target.
Enables: Check the operational risk before changing capacity.

More evidence can confirm or invalidate an opportunity.

Is this worth the engineering effort?

Your assumptions stay in this browser. Compare one-time engineering cost with the lower end of this report's modeled monthly range over a horizon you choose.

Engineering hoursCost per hour (USD)Decision horizon (months)

Enter positive hours and hourly cost to compare.

Loading sample evidence…

Savings ledger

Make the change reviewable

Proposed change

Validate and recover

Check before acting

From observation to outcome

What would make this decision stronger?

Performance evidence

Make the change reviewable

Proposed change

Validate and recover

Check before acting

From observation to outcome

What would make this decision stronger?

Performance evidence