For finance teams
Veqen prepares the journal entries for your company's AI spend each month. Costs are classified by product, project and team as cost of revenue, R&D or G&A, matched to provider invoices, netted against credits and accrued, ready to review, approve and import into QuickBooks or Xero.
An engineer connects it once. Finance owns the policy and the approval.
| Account | Description | Debit | Credit |
|---|---|---|---|
| AI-2026-09 · September 30 · AI inference costs | |||
| 5100 AI inference | Cost of revenue | $8,912.40 | |
| 6110 AI inference, R&D | Research and development | $1,204.77 | |
| 1350 Prepaid AI credits | Provider credits used | $2,500.00 | |
| 2150 Accrued AI costs | Billed in arrears | $7,617.17 | |
| AI-2026-09-R · October 1 · Reverse the accrual | |||
| 2150 Accrued AI costs | Reverse accrual | $7,617.17 | |
| 5100 AI inference | Reverse accrual | $7,617.17 | |
An example close. Account names and codes are yours to set.
How it works
The close is built from the same ledger engineering uses for cost per customer, and each pack states the method used to produce it.
Spend is classified by the project, feature or environment that caused it: the support bot is cost of revenue, the drug-discovery program and the staging environment are R&D, the claims-intake agent is G&A. Anything unclassified goes where you say.
| Rule | Books to | September |
|---|---|---|
| Feature: support-bot | Cost of revenue | $6,104.22 |
| Feature: summaries | Cost of revenue | $2,808.18 |
| Environment: staging | R&D | $1,204.77 |
With a provider cost-access key, Veqen imports what OpenAI and Anthropic billed, day by day. For Together, Bedrock, fal, a GPU cloud or any other provider, enter the invoice total each month. Veqen books the bill where there is one and its own exactly priced ledger where there isn't. Months can't be approved until every connected provider's costs are in.
| Provider | Billed | Tracked | Matched |
|---|---|---|---|
| OpenAI | $4,212.09 | $4,190.55 | 99.5% |
| Anthropic | $5,820.31 | $5,782.02 | 99.3% |
| Together, entered | $3,140.00 | $3,101.26 | 98.8% |
| Own GPUs, entered | $2,880.00 | $2,705.40 | 93.9% |
Record the prepaid and promotional credits you hold with providers. Usage draws down the grant that expires first; prepaid usage is booked against the asset, promotional usage costs nothing. You're warned 30 and 7 days before credits expire unused.
| Grant | Used | Left | Expires |
|---|---|---|---|
| OpenAI, prepaid | $2,500.00 | $7,500.00 | Mar 2027 |
| Anthropic, startup program | $1,840.00 | $160.00 | In 12 days |
Usage billed in arrears is accrued at month end and reversed on the 1st, so the expense sits in the month it was incurred. Approve the month to freeze it, then download the entries for QuickBooks or Xero.
| Entry | Debit | Credit |
|---|---|---|
| Sep 30 · 5100 AI inference | $7,617.17 | |
| Sep 30 · 2150 Accrued AI costs | $7,617.17 | |
| Oct 1 · reversal | $7,617.17 | $7,617.17 |
Controls
Each figure traces back to the calls behind it, the rates that priced them and the person who approved the month.
Every call stores the version of the price book that priced it. Rates follow each provider's published rules, including caching, batch and long-context tiers.
Each close states how it was built: sources, classification rules, credit treatment and accruals, in plain sentences you can hand to an auditor.
Approved months don't change. Reopening rebuilds the entries from current data and is recorded, with who and when.
Keys, roles, connections, budgets, billing and every approval are logged for two years. A finance role sees the books without touching engineering settings.
Margin
Connect a read-only Stripe key and revenue sits next to AI cost for every customer. The ones that cost more than they pay are listed first, before the renewal conversation.
| Customer | Revenue | AI cost | Margin |
|---|---|---|---|
| Hooli | $49.00 | $71.80 | −46.5% |
| Globex | $299.00 | $97.31 | 67.5% |
| Initech | $899.00 | $268.94 | 70.1% |
| Acme | $1,990.00 | $412.20 | 79.3% |
Setup
The first two steps are a one-time task for an engineer. Send them the quickstart.
An admin adds read-only OpenAI and Anthropic keys, and the last 30 days of billed cost appear right away. Invoices from other providers are entered once a month.
An engineer wraps the AI client so each call carries its customer, feature or project. That's what makes classification and margin possible.
Finance maps projects, features and environments to cost of revenue, R&D or G&A, names the accounts and records any credits.
Review the entries, approve, and import the CSV into QuickBooks or Xero.
No. It prepares the AI-cost part of your month-end: classified costs, credit drawdown, accruals and the journal entries. Your accounting system stays the book of record, and you or your accountant approve what goes into it.
QuickBooks Online and Xero, through their journal-entry CSV import. The account names and codes are yours: start each with its code, such as 5100 AI inference, and the Xero export uses it as the account code.
Prepaid credits are an asset, so usage they cover is booked against your prepaid credits account. Promotional credits cost you nothing, so usage they cover isn't booked as expense. Credits about to expire are flagged 30 and 7 days ahead.
Where a cost connection exists, Veqen books what the provider billed. Otherwise it accrues its own priced ledger and reverses the accrual on the 1st. A month can't be approved until every connected provider has synced past its last day.
Approved entries are frozen. Reopening a month rebuilds them from current data, and approvals and reopenings are recorded in the audit log.
No. It records the model, token counts, the price and the tags your engineers attach, such as a customer id. Never prompts or responses.
The close is included in Team and available on Pro as an add-on.
