Enterprises run business transactions through multi‑agent AI systems that span clouds, large language models, orchestration harnesses, and systems of record. Finance only receives AI cost at settlement grain, aggregated by SKU and billing period, so the question “what did this order, claim, or case cost in AI?” cannot be answered. Providers never see the enterprise transaction boundaries.
The FinOps Open Cost and Usage Specification (FOCUS) gives a common schema for billing data, but it remains settlement‑grain even after adding token and requester attributes. We formalize this grain mismatch, derive seven requirements, and design the Two‑Ledger Model.
The model consists of: (1) a settlement ledger that stores FOCUS‑conformant billing records; (2) an attribution ledger that captures invocation‑grain usage via OpenTelemetry GenAI telemetry. A layered correlation spine links the two ledgers: trace context → A2A/MCP metadata → business‑key correlation → probabilistic matching. Each usage record is assigned to a specific business transaction with recorded method and confidence.
A versioned rate card prices usage in near real‑time. Period reconciliation aligns both ledgers to the invoice; the unexplained residual identifies “shadow AI”. Costs can then be reported per order, claim, or outcome, together with agentic cost metrics.
Simulation results show that when trace propagation drops to zero, the spine still maintains >99% cost coverage; reconciliation residual recovers planted unmetered spend within 0.24 percentage points and flags provider billing errors; mid‑flight gating reduces the p99 transaction cost by up to 61% at a 2‑4% false‑pause rate.
We propose the attribution ledger as an invocation‑grain FOCUS profile for standardization.
$$\text{Cost}= \sum_{i} \text{Usage}_i \times \text{Rate}_i$$
Review: By marrying fine‑grained attribution with a unified billing ledger, the Two‑Ledger Model delivers end‑to‑end AI cost visibility, giving enterprises a solid foundation for FinOps decision‑making.