AI Workforce · Module
AI FinOps
Know what every AI call costs, keep budgets in hierarchy and degrade gracefully
AI FinOps allocates every model call's cost from tenant down to department, team, agent and run, enforces hierarchical budgets, and links cost to outcomes. When budgets tighten, lower-priority work queues first so customer-facing service keeps running.
Cost visibility down to the run
AI costs are hard to predict and harder to explain. AI FinOps, planned for W2, records cost per model span with allocation keys from tenant to department, team, agent and run, and by feature such as Copilot, agent, embedding or render. Shared costs such as utility agents, embeddings and indexing are allocated by usage share.
Budgets that nest
Hierarchical AI budgets mean a child budget can never exceed its parent's remaining budget. Alerts fire at configurable thresholds, with a forecast at the current run-rate. Spend anomaly detection compares spikes against a rolling baseline. Per-run limits on steps, seconds and cost are reported with breach counts per agent.
Degrade, don't fail
The graceful degradation policy assigns priority classes: P0 customer-facing SLA work, P1 revenue-impacting work, P2 internal work and P3 batch work. When budgets or capacity tighten, lower classes queue first. P3 work such as evals, backfills and indexing can be scheduled off-peak. Approvals and R4 paths are never blocked mid-flow.
Credits and BYOK
The AI credit meter maps cost to credits with balances and soft and hard limits. Where a tenant brings its own provider keys, the provider cost is estimated and shown but not billed as credits.
Cost against outcome
Cost per outcome joins cost records to outcome events, and the same data feeds the Investment & ROI Ledger. Routing savings analysis finds cheaper model routes that pass the same evals and proposes them for approval. Unit cost views show AI cost per active tenant per month. Showback and chargeback exports go by cost centre, energy and carbon estimates per AI workload are planned, and enterprise tenants can use capacity-pool budgets across their Digital Employees.
Features
What AI FinOps includes
Cost per model span
Allocation keys from tenant to department, team, agent and run, and by feature.
Hierarchical budgets
Child budgets cannot exceed the parent's remaining budget.
Budget alerts and forecasts
Configurable thresholds and run-rate forecasts.
Cost per outcome
Cost records joined to outcome events.
Graceful degradation
P0 to P3 priority classes; lower classes queue first.
Routing savings analysis
Cheaper routes that pass the same evals, proposed for approval.
Shared cost allocation
Utility agents, embeddings and indexing by usage share.
Per-run limits reporting
Steps, seconds and cost limits with breach counts.
AI credit meter
Cost to credits with soft and hard limits; approvals never blocked mid-flow.
Spend anomaly detection
Spikes against a rolling baseline.
Off-peak batch scheduling
Evals, backfills and indexing at P3.
BYOK cost visibility
Tenant-paid provider cost estimated and shown.
Showback and chargeback
Exports by cost centre.
Energy and carbon estimates
Per AI workload.
Workflow
Budget breach handling
- 1
Spend approaches a budget threshold
- 2
An alert goes to the budget owner with a forecast
- 3
At the cap, P3 and P2 work queues first
- 4
Customer-facing P0 work continues
- 5
The owner raises the budget through approval or accepts the queue
AI
AI skills and Digital Employees
Spend anomaly detection
Flags unusual spend against baselines.
Routing savings proposals
Finds cheaper routes that pass the same evals.
Digital Employees that use this module
- AI FinOps Analyst (proposed)
They act through the same governed Actions as people; risky steps wait for approval.
Frequently asked questions
Will an agent stop talking to a customer when credits run out?
The degradation policy queues lower-priority work first so customer-facing work continues, and conversations can move to a human queue.
Can I bring my own model keys?
Where the plan allows, yes. The provider cost is estimated and shown but not billed as credits.
When is AI FinOps planned?
W2, inside the Year-1 boundary.
See AI FinOps in your workflow
Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.