Vivek Acharya’s Two-Ledger Model links AI costs to business transactions

Vivek Acharya’s Two-Ledger Model links AI costs to business transactions

Vivek Acharya argues that enterprise billing data cannot answer a practical question: what did a particular order, claim, or case cost in AI? Providers bill at settlement grain—by SKU and charge period—while businesses need costs associated with individual transactions. Acharya proposes joining a settlement ledger of FOCUS-conformant billing data with an attribution ledger of invocation-grain usage records built on OpenTelemetry GenAI telemetry. A layered correlation spine uses trace context, A2A/MCP metadata, business-key correlation, and probabilistic matching to assign records to transactions, recording the method and confidence. A versioned rate card prices usage near real time; period reconciliation compares both ledgers with the invoice, with the unexplained residual identifying possible shadow AI. The model also proposes reporting costs per order, claim, or outcome and using mid-flight gating to limit transaction costs. In a simulation with known ground truth, Acharya reports cost coverage above 99% even when trace propagation falls to zero, recovery of planted unmetered spend within 0.24 percentage points, and p99 transaction-cost reductions of up to 61% with a 2-4% false-pause rate. These are simulation results, not evidence of performance in deployed enterprise systems. Acharya proposes the attribution ledger as an invocation-grain FOCUS profile for standardization.

Publication: abf09e61-e171-4682-8e8c-5e8291a8c86f

attribution — pass: The source lists a single author, Vivek Acharya, and the reading attributes the argument, proposal and simulation results to him. One claim says 'the authors report,' which is a minor wording slip for a single-author paper; it does not misattribute the work. The smallest repair is to change it to 'Acharya reports.'

Authors:Vivek Acharya

summary — pass: The reading leads with the actor and his central problem: billing cannot show the AI cost of an order, claim or case. It explains settlement grain and keeps the technical terms exact; terms such as FOCUS, OpenTelemetry, A2A/MCP, p99 and shadow AI suit tooltips. The proposal and its simulation caveat are faithful. Small fidelity repairs: mention that these are multi-agent AI systems spanning clouds and models, and note that the residual also flags provider billing errors. 'Possible shadow AI' slightly softens the source's 'identifies shadow AI.'

Enterprises run business transactions through multi-agent AI systems spanning clouds, large language models (LLMs), orchestration harnesses, and systems of record.

date — pass: The originalPublished date, 2026-11-04, matches the arXiv v1 submission date. The feed date of 2026-11-06 is later and correctly not used.

[Submitted on 4 Nov 2026]

support — pass: Each claim quotes the abstract verbatim. The reading frames the figures as simulation results rather than evidence from deployed systems, and the question about real-world performance follows from that gap without implying an answer. No unsupported benefits or costs are added.

A simulation with known ground truth shows the spine sustaining cost coverage above 99% as trace propagation falls to zero

media — pass: The selected media is the verified arXiv abstract page for this paper, which is the direct original source.