AI company accounting: token economics, end to end.

Subledger

A usage subledger built for tokens.

Import provider usage by CSV into a company-scoped provider/model taxonomy - input/output/cache tokens, requests, images, GPU-hours. Imports are idempotent and auto-create unknown providers and models, so the subledger is the operating layer under every cost and margin number.

• Anthropic, OpenAI, Google, Bedrock, Azure taxonomy - extensible.
• Idempotent CSV / manual import; dedupe across runs.
• An AI-SaaS chart-of-accounts template applied on demand.

Live capture

Real-time cost capture - bring your own key, no code change.

Skip the CSV. Route your provider traffic through the Quoining gateway with your existing bring-your-own-key - one base-URL change in your SDK and a gateway token instead of your provider key. Every call is captured into the usage subledger in real time, priced per request, and attributable to the customer it served - so gross margin is cost-to-serve per customer, not a monthly guess.

• Reuses your existing BYOK key - no admin key to provision.
• One base-URL change; your app never holds the provider key.
• Per-request capture with per-customer cost attribution.

Recognition

Cost and revenue, recognized into the GL.

Recognize usage cost (prepaid-asset relief or accrue-as-incurred) and customer revenue (bill-on-usage or prepaid-deferred) on configurable policies with no defaults - the engine refuses to post until you choose. Record customer prepaid credits as a contract liability that relieves into revenue as usage is consumed.

• Cost: DR COGS/R&D/G&A, prepaid-asset or accrued.
• Revenue (ASC 606): bill-on-usage or prepaid-deferred.
• Customer prepaid credits as deferred revenue.

Commitments & breakage

Committed minimums, overage, and breakage.

Link a customer's usage to a committed-minimum arrangement so it recognizes as a single stream, and recognize overage above the minimum. When a prepaid credit expires unused, the stranded remainder is recognized as breakage revenue under ASC 606-10-55 - by a closed form that can't double-recognize.

• Committed minimums with overage true-up (three methods).
• Prepaid-credit expiry → breakage (ASC 606-10-55).
• Single recognition stream - never double-counted.

R&D & tax

Capitalize R&D and carry the §174 tax layer.

Tag development-stage usage to capitalize it as internal-use software under ASC 350-40, place it in service as an amortizing intangible, and elect a §174 tax treatment per project - domestic §174A immediate expense or foreign §174 15-year amortization. The deferred-tax engine carries the as-incurred book-vs-tax timing difference, including the DTL→DTA flip.

Unit economics

Gross margin by model and by customer.

A read-only unit-economics report shows gross margin on AI usage overall and by model and by customer, with capitalized R&D excluded from margin and disclosed separately. Export to CSV or PDF. A one-click period close runs every recognition step in the correct order.

• Margin by model and by customer; recognized-vs-recorded.
• CSV / PDF export.
• One-click close: cost → revenue → overage → breakage → §174.

Why Quoining

AI-token accounting, native - not a spreadsheet bolted onto SaaS metrics.

There is no AI-token-specific GAAP standard, so every policy with a real fork - cost treatment, revenue treatment, §174 election, breakage method - is configurable with no default, applied under existing GAAP and tax frameworks. The accounting is grounded in ASC 606, ASC 350-40, and §174, not a hardcoded conclusion.