GVGanivra

Free tool · No login · OpenAI invoices

AI Invoice
Reconciliation

Reconcile your AI invoices against recorded usage. Compare costs, identify model-level differences, and uncover gaps in your usage data.

Files stay in your browserNo API key neededDownload your findings
Start with your bill, or explore an example.

One account · One billing period · USD only

01 / Recorded usage

Add your usage costs

Prepare a CSV with record_id,date,cost_usd. Use unique IDs and dates in YYYY-MM-DD format. Costs must be before invoice credits and adjustments. Optional columns: model and customer_id. Use exact matching model names across both files.

Up to 2 MB / 50,000 rows. Processed locally; contents are not uploaded.

No file selected yet.

record_id,date,cost_usd,model,customer_id
usage-001,2026-08-01,24.50,model-a,customer-a
usage-002,2026-08-12,35.25,model-a,customer-b
usage-003,2026-08-28,40.25,model-b,customer-a

Use pseudonymous record and customer IDs. Do not include prompts, API keys, or customer contact details. The report contains totals and checks, not individual records.

02 / OpenAI invoice

Enter the invoice figures

When selected, your next real-file comparison saves usage and invoice totals, differences, period length, coverage counts, and missing-data flags to our research database. No raw files, dates, account labels, model names, or customer IDs are sent. Sample financial figures are never saved.

Files and form entries stay in this page unless you share a summary or submit the pilot form. We record usage actions, comparison mode, and error stages with a random browser-session ID in our database. Google Analytics receives categorical actions only, not financial summaries or contact details. Reloading clears the form, not previously submitted records. See our Privacy page for details.Research session reference: available after loading

Illustrative example / USD

How to reconcile an OpenAI invoice with usage costs

For the same account and August 2026 billing period, the sample CSV contains three usage records totaling $100.00. The invoice lists $108.00 in usage charges, $5.00 in credits, and a positive $2.00 adjustment.

Expected subtotal compared with the supplied invoice
ItemAmount
Recorded usage costs$100.00
Credits to subtract$5.00
Adjustment to add$2.00
Expected net subtotal: usage − credits + adjustment$97.00
Invoice net subtotal$105.00
Difference: invoice minus expected$8.00

The invoice arithmetic is consistent: $108.00 − $5.00 + $2.00 = $105.00. The $8.00 gap comes from invoice usage charges exceeding the recorded usage costs. Check account scope, missing or delayed records, and cost assumptions to investigate it.

This is sample data, not a discovered overcharge. Use “Try sample data” above to explore the same figures interactively. Matching totals would still not prove that the supplied records are complete.

What reconciliation tells you

A clear comparison, with the gaps visible.

Same account. Same period.

Use USD usage costs for the same OpenAI account, projects, and services as your invoice. Dates are UTC and both selected dates are included. Prepare your export using the template; raw token exports and invoice PDFs are not supported yet.

Costs, not token estimates.

Each row supplies an existing usage cost before credits. This tool sums those costs; it does not calculate token prices or verify negotiated rates. Use either request-level records or daily totals, never both for the same usage.

Differences need context.

Late records, different account scopes, rate changes, rounding, or invoice adjustments can explain a mismatch. Missing days do not prove missing usage. Matching totals do not establish completeness or validate every charge.

Expected net subtotal = recorded usage costs − credits + signed adjustments. Compare against the invoice subtotal after these items, excluding tax, prior balances, and payments. A $0.01 tolerance is applied to each comparison. This tool does not assess revenue recognition or compliance.

AI invoice reconciliation FAQs

What is AI invoice reconciliation?

AI invoice reconciliation compares recorded usage costs with provider invoice charges for the same account, services, and billing period. It helps identify differences that need investigation. This tool compares supplied amounts; it does not independently verify provider rates or prove every usage record is present.

Which providers and file formats are supported?

This version supports OpenAI invoices in USD. Enter invoice figures manually and prepare a usage CSV with record_id, date, and cost_usd columns. Optional model and customer_id columns enable more detail. Raw token exports, invoice PDFs, and direct provider connections are not supported.

What is the difference between quick and detailed reconciliation?

Quick mode compares total usage costs, credits, adjustments, and the invoice net subtotal. Detailed mode also accepts an invoice breakdown CSV with model,cost_usd columns and one row per model. Use exactly matching model names and the same scope and period across both files.

Why might my invoice differ from recorded usage?

Possible explanations include different account or project scopes, late records, missing usage, estimated rather than provider-reported costs, rounding, or invoice adjustments. The tool separates usage differences from invoice arithmetic differences. A mismatch is not proof of overcharging.

How should I enter credits, adjustments, and tax?

Enter credits as positive amounts to subtract. Other adjustments are signed: positive adds a charge and negative reduces it. Compare the invoice subtotal after these items but before tax. Exclude prior balances, payments, and prepaid credit top-ups. Expected net subtotal equals recorded usage minus credits plus signed adjustments.

How are duplicates and missing data handled?

Identical records with the same ID are counted once; conflicting repeated IDs block the comparison. Usage outside the inclusive UTC billing period is excluded. Days without records are flagged but may indicate inactivity or aggregated records. In detailed mode, missing model labels and incomplete invoice breakdown totals are shown explicitly.

Does matching mean my invoice is correct?

No. Amounts within $0.01 are treated as matching, but matching totals do not establish completeness, confirm negotiated rates, or validate every charge. Review the underlying records and invoice scope before drawing conclusions.

Does this calculate customer profitability?

Optional customer IDs show how recorded costs are concentrated across customers. This is a cost view, not a profitability calculation. Customer revenue and other delivery costs are needed to evaluate margins.

Are my files or financial figures uploaded?

CSV files are processed locally in your browser. Usage actions and comparison mode are recorded with a random session identifier. If you opt in to research sharing, a real-file comparison sends aggregate financial summaries and coverage flags to Ganivra; raw files and customer IDs are not included. Submitting the pilot form saves your contact details and links them to the session. Google Analytics receives categorical actions, not financial summaries or contact details.

Is AI Invoice Reconciliation free?

Yes. No login or API key is required. You can try illustrative sample data and download a report. The tool does not perform an audit or assess revenue-recognition compliance.

Read our privacy policy for research storage and deletion information. Planning your business economics? Explore the AI Startup Financial Model and Runway Calculator or calculate cost per outcome.