Unattributed spend
A growing share of the AI bill maps to no team, no application, and no cost center. It cannot be forecast, cannot be allocated, and cannot be governed.
AI Cost Intelligence
Your AI bill is growing faster than your ability to explain it.
The AI Token Optimization Assessment maps every AI API dollar across Bedrock, Azure OpenAI, Vertex AI, and direct APIs to the team, application, and environment that generated it. No tagging prerequisite. No self-reported spreadsheets. Evidence from your live environment, delivered in days.
Read-only, approval-gated scan
Findings in days, not months
No credentials stored centrally
No tagging required to start
38%
Of AI spend had no owner in a recent sample engagement
+312%
Six-month AI spend growth in the same environment
0
Agents with enforced token budgets before the assessment
Days
From connection to a board-ready deliverable
The Problem
Token spend grows quietly, then all at once.
Teams adopt LLM APIs the way they once adopted cloud: fast, without a provisioning process, and without a budget model. By the time finance notices, the bill has tripled and nobody can say which team, application, or experiment is driving it.
Shadow AI experiments
Workloads launched outside any formal process are often the fastest-growing spend category, and the next runaway workload is usually hiding among them.
Non-linear growth
Token spend does not grow like infrastructure spend. One team scaling an agent from 2 users to 34 can multiply a line item by ten in a quarter.
No budget model
Provider quotas sit at the account level, not per workload. A runaway or compromised workload can consume the whole quota and generate unbounded spend.
No alerting thresholds
Without per-team and per-application thresholds, the first alert most organizations get is the monthly invoice.
Forecasts nobody trusts
When 38 percent of spend has no owner, every projection handed to the board is a guess. The CFO knows it, and so does the audit committee.
The Attribution Gap
Five questions your CFO is already asking.
These are the questions this assessment answers with evidence from your live environment, not estimates.
1
What is our total AI spend this month, across every provider and every account?
2
Which teams and applications generate it, and how much has no owner at all?
3
What will the number be in Q4 if nothing changes?
4
Which workloads have token budgets or spending ceilings, and which have none?
5
Where should budgets, quotas, and alerts be set first?
What the Assessment Finds
This is what the deliverable looks like.
Sample findings below are drawn from an illustrative assessment of a fictional industrial enterprise, Astro Mining International, with six AWS accounts and two Azure subscriptions. The structure is exactly what you receive for your environment.
Sample finding. Illustrative environment.
AI spend grew 312 percent in six months, from $23,000 to $94,800 per month, with 38 percent of it ($36,024/mo) mapped to no team or application and zero enforced token budgets across 23 discovered agents.
The unattributed category was the fastest-growing spend line in the environment, up 1,616 percent in six months, reflecting AI experiments launched outside any formal process. At the observed growth rate, monthly spend would cross $198,000 by Q4 with no intervention. One provider project carried $8,200 per month in spend whose owner had left the organization 60 days earlier with no documented handoff. None of the critical remediations required architectural change: they were budget, quota, tagging, and IAM scope corrections.
$94,800
Monthly AI token spend across all providers
38%
Of spend with no team or application owner
+1,616%
Six-month growth of unattributed shadow AI spend
$198K
Projected monthly spend by Q4 with no governance
How It Works
Two scoped phases. Deterministic results. Human approval at every gate.
The assessment runs as two scoped, deterministic, approval-gated phases: inventory discovery through approved connectors, then deterministic policy checks against that inventory. Oscar never performs open-ended network scanning, and no action executes without explicit human sign-off.
- 1
Connect with read-only credentials
Oscar connects through approved connectors using read-only access and the permissions your engineers already hold. Setup takes under 30 minutes. No credentials are stored centrally and no changes are made to your environment.
- 2
Oscar builds the spend picture from live sources
Cost APIs, provider-native token metering, IAM policy analysis, and runtime logs are correlated in the Cloud Intelligence Graph. Every dollar is resolved to a team, application, and environment where the evidence supports it, and flagged as unattributed where it does not.
- 3
Deterministic checks run against the inventory
Budget coverage, quota placement, alerting thresholds, and attribution completeness are checked deterministically against the discovered inventory. Same inputs, same findings, every time. Findings are labeled Live, Inferred, or Gap so you can see the evidence basis for each one.
- 4
You receive the report and remediation roadmap
Spend trend and forecast, full attribution map, provider and account breakdown, and a prioritized roadmap where every item is an Oscar-ready task with an owner and a time horizon attached.
What You Get
Five deliverables, all from live infrastructure.
Every number in the report is traceable to a live API response, a policy analysis, or an explicitly labeled inference. Nothing is assembled from interviews.
Spend Trend & Forecast
Month-by-month spend across all providers with growth attribution by application, plus a forward forecast showing where the number lands with and without controls.
Attribution Map
Every dollar mapped to team, owner, application, and environment. The unattributed share is called out exactly, not approximated, so you know precisely what cannot currently be governed.
Provider & Account Breakdown
Spend by provider, model, account, and environment across Bedrock, Azure OpenAI, Vertex AI, and direct APIs. Concentration risks and orphaned projects surfaced explicitly.
Token Budget Recommendations
Recommended budgets, quotas, and alert thresholds by team and application, sized from observed usage rather than guesswork, with the highest-leverage controls first.
Prioritized Remediation Roadmap
Every finding delivered as an Oscar-ready task with owner, priority, and time horizon. Items marked NOW require no architectural change and are addressable within 30 days.
Ongoing Spend Monitoring
The assessment is a baseline. Monitoring keeps it current: new workload detected, alert fires. Spend crosses a threshold, alert fires. Growth trend shifts, you know first.
Why OpsCanvas
Attribution without a tagging project.
The hard part is not reading the bill. It is connecting the bill to your organization.
Provider cost dashboards tell you what you spent by service. They cannot tell you which team owns the spend, which application generated it, or what happens to the number next quarter, because that requires correlating cost data with IAM identities, runtime logs, and organizational topology.
The usual fix is a tagging initiative that takes two quarters and decays the moment it ships. OpsCanvas skips the prerequisite: the Cloud Intelligence Graph resolves ownership from what is actually running, so attribution arrives in days and does not depend on tag discipline.
✓
No tagging prerequisite and no instrumentation project✓
Multi-provider coverage in a single engagement✓
Findings labeled Live, Inferred, or Gap with evidence attached✓
Every recommendation is an Oscar-ready remediation task✓
Deterministic checks: same inputs, same findingsCapability
OpsCanvas
Discovery method
Live APIs via approved connectors
Tagging required
No
Providers covered
Bedrock, Azure OpenAI, Vertex AI, direct APIs
Attribution granularity
Team, owner, application, environment
Forecast model
With and without controls
Continuous monitoring
Available as add-on
Common Questions
Frequently asked.
Does the assessment require us to tag our resources first?
No. Attribution is resolved from the Cloud Intelligence Graph, which correlates IAM roles, resource relationships, execution logs, and organizational topology. Where tagging exists it is used as one signal among several. Where it does not, ownership is resolved from what is actually running.
Which AI providers and APIs are covered?
AWS Bedrock, Azure OpenAI, and GCP Vertex AI through their native cost and metering APIs, plus direct API usage such as OpenAI and Anthropic where it is observable from your environment through runtime logs and egress patterns.
Does Oscar scan our network to find AI workloads?
No. Oscar performs no open-ended or autonomous network scanning. The assessment runs as two scoped, deterministic, approval-gated phases: inventory discovery through approved connectors using existing permissions, then deterministic policy checks against that inventory.
How is unattributed spend calculated?
The unattributed share is the exact portion of spend where the graph found no consistent identity or ownership linkage sufficient to assign an owner with confidence. It is not an estimate, and the report shows the evidence basis for every attributed dollar as well.
What happens after the report is delivered?
Every finding arrives as an Oscar-ready task with owner and priority attached. You can execute the roadmap yourself, or run Oscar in co-pilot mode where Oscar proposes each action, an engineer approves it, and every change is logged with a full audit trail. Ongoing monitoring is available to keep the baseline current.
Can this be combined with the other AI assessments?
Yes, and it usually should be. The Context Graph is built once, so the Agent Risk Assessment and GPU Optimization Assessment stack on the same foundation at reduced incremental effort. Token spend and agent risk in particular are two views of the same inventory.
Get Started
Know what your AI costs before the CFO asks.
Read-only connection through approved connectors. No tagging required. No changes to your environment. An evidence-backed spend picture in days.
Read-only access | No changes to your environment | Setup under 30 minutes