This question-and-answer walkthrough is for finance, FinOps and engineering teams closing a month for an AI SaaS product. It follows a fictional October 2026 review from a $40 net usage-to-invoice difference to $900 of customer contribution and a specific follow-up for a loss-making account.
Sample scope: all providers, customers, evidence IDs and approvals are fictional. The pack contains aggregated evidence summaries, not raw transaction exports or original invoices. It illustrates operational contribution after included costs, not net profit, an audited statement or a real Ganivra customer result.
Download the sample close pack (printable HTML) · Download the same sample inputs and review as JSON. Open the HTML in a browser and use Print / Save PDF for a review copy.
What should an AI monthly close pack contain?
Keep the scope, source totals, adjustments and approval together. A reviewer should be able to trace a customer’s contribution back to both revenue and delivery-cost evidence without reconstructing the analysis.
| Record | Sample reference | What it establishes |
|---|---|---|
| Scope and period | SCOPE-10 | October 1–31, 2026 inclusive, UTC; USD; production accounts A and B |
| Usage snapshots | USG-A-01 / USG-B-01 | $1,000 and $200 initially recorded before supplier credits |
| Provider invoices | INV-A-10 / INV-B-10 | $1,040 and $200 net subtotals, excluding tax |
| Corrections and credit | EX-01 / EX-02 / CM-A-10 | Missing usage, duplicate removal and a supplier service credit |
| Delivery records | TOOLS-10 / INFRA-10 / REVIEW-10 | $120 tools, $90 shared infrastructure and $50 human delivery review |
| Revenue schedule | REV-A-01 / REV-B-01 / REV-C-01 | Period allocation, refund and consumed-credit basis |
| Review record | CLOSE-10-v1 | Scope, exceptions, preparer, reviewer, decision and follow-up |
For a real close, retain the actual exports, invoice copies, event IDs, price versions and mapping rules behind these summaries. The FinOps Foundation’s invoicing and chargeback guidance emphasizes coordination with Finance, data quality, allocation and correction procedures. This walkthrough applies that discipline to a small illustrative AI workload.
Fix the timezone, currency and account scope before comparing numbers. Exclude payments, prior balances and taxes from this example’s usage-cost comparison. Convert currencies consistently and document any conversion differences in a real review.
How do you reconcile recorded usage with provider invoices?
Build a bridge from the original usage snapshot to the invoice’s comparable net subtotal. Provider A’s initial usage is $1,000, but its net invoice is $1,040. The $40 net gap hides three different issues.
| Provider | Initial usage | Missing events | Duplicate correction | Supplier credit | Adjusted cost | Invoice net | Residual |
|---|---|---|---|---|---|---|---|
| A | $1,000 | $100 | −$20 | −$40 | $1,040 | $1,040 | $0 |
| B | $200 | $0 | $0 | $0 | $200 | $200 | $0 |
Provider A corrected usage = $1,000 + $100 − $20 = $1,080
Provider A net cost = $1,080 − $40 supplier credit = $1,040
Combined provider cost = $1,040 + $200 = $1,240First compare corrected usage with invoice usage charges before credits. Then apply the credit to reach the net amount. Keep those comparisons separate so a missing batch cannot be hidden by a supplier discount.
What caused the discrepancies?
| ID | Issue | Impact | Evidence / resolution | Owner | Status |
|---|---|---|---|---|---|
| EX-01 | Missing batch events | +$100 usage | Batch-export reconciliation; $60 Acme / $40 Beacon | Engineering | Resolved in v1 |
| EX-02 | Duplicate telemetry | −$20 usage | Same event ID recorded twice; remove Acme duplicate, retain real retries | Engineering | Resolved in v1 |
| EX-03 | Supplier service credit | −$40 cost | Credit memo CM-A-10; allocate $20 / $10 / $10 by affected incidents | Finance | Resolved in v1 |
| EX-04 | Beacon negative contribution | −$27 contribution | Complete included-cost view; investigate retry workload | Product owner | Open action; review 2026-11-17 |
The duplicate correction removes repeated telemetry for one event; it does not remove a genuinely charged retry. The supplier credit is allocated using the affected incidents, not used as an arbitrary plug to make totals match.
A zero residual is necessary for this sample but not sufficient evidence of completeness. Missing costs can offset duplicate costs. Check source coverage, stable IDs, applicable rates, delayed events and customer mappings separately. If a model price is unknown, preserve the unpriced usage instead of treating its cost as zero.
How do reconciled provider costs become customer costs?
Apply each correction to its supported customer mapping, then add other delivery costs once. The $100 recovered batch belongs $60 to Acme and $40 to Beacon. The duplicate belongs to Acme. Credit memo CM-A-10 allocates the $40 supplier credit as $20, $10 and $10 based on affected incidents.
| Customer | Initial AI cost | Late usage | Duplicate | Credit | Reconciled AI cost |
|---|---|---|---|---|---|
| Acme | $600 | $60 | −$20 | −$20 | $620 |
| Beacon | $400 | $40 | $0 | −$10 | $430 |
| Cedar | $200 | $0 | $0 | −$10 | $190 |
The customer AI costs total $620 + $430 + $190 = $1,240, matching the provider bridge. Unassigned cost is $0 only because the sample assumes all mappings were resolved. In a real pack, keep an explicit unassigned bucket until there is evidence for allocation.
Next add $120 of tools, $90 of shared infrastructure and $50 of human delivery review. The shared pool uses measured serving-time shares of 50%, 30% and 20%, assigning $45, $27 and $18. Direct tool and review costs use customer-linked records. These pools exclude the model costs already counted; adding the provider invoices again would double-count them.
How do you match subscriptions, refunds and credit revenue to the month?
Use a stated period basis and distinguish revenue allocation from cash collection. This example uses operational allocation to show customer economics; Finance must determine the accounting treatment separately.
| Customer | Reference | Basis | Period allocation |
|---|---|---|---|
| Acme | REV-A-01 | $18,000 annual service amount / 12 months | 1500 |
| Beacon | REV-B-01 | $550 monthly amount less $50 service refund | 500 |
| Cedar | REV-C-01 | 2,000 credits consumed at $0.20 from a $600 / 3,000-credit pack | 400 |
Acme’s annual amount is spread evenly over twelve months in this sample. Beacon’s $50 customer refund reduces its $550 monthly amount to $500. This customer refund affects revenue; Provider A’s $40 supplier credit affects cost. They belong on different sides of the calculation.
Cedar paid $600 for 3,000 credits and consumed 2,000 at $0.20 each. The consumption view allocates $400 to the period; $200 remains unconsumed grant value. Do not also count the full $600 cash receipt as revenue for those same actions. See the AI credit-pricing walkthrough for packs, top-ups and refunds.
Missing revenue is not confirmed zero revenue. An account with no mapped revenue should show an incomplete margin, while a confirmed free account can show a dollar cost and no defined margin percentage.
What does the final customer-margin review show?
The portfolio contributes $900 after $1,500 of included costs, but Beacon loses $27. A reconciled company total can still conceal an account whose usage is underpriced.
| Customer | Revenue | AI cost | Tools | Shared | Review | Total cost | Contribution | Margin |
|---|---|---|---|---|---|---|---|---|
| Acme | $1,500 | $620 | $60 | $45 | $0 | $725 | $775 | 51.67% |
| Beacon | $500 | $430 | $40 | $27 | $30 | $527 | −$27 | -5.40% |
| Cedar | $400 | $190 | $20 | $18 | $20 | $248 | $152 | 38.00% |
| Portfolio | $2,400 | $1,240 | $120 | $90 | $50 | $1,500 | $900 | 37.50% |
Calculate the portfolio margin from totals: $900 ÷ $2,400 = 37.5%. Do not average customer margin percentages. The costs here exclude taxes, payment fees, general support and company overhead, so the result is not full gross margin or net profit. For that distinction, see AI gross margin per customer.
Beacon’s next action is to inspect its expensive executions and retry workload, then evaluate delivery changes or pricing. This is an investigation, not proof that retries caused the loss. Use the customer-margin review worksheet to assign owners and follow-up dates.
What should the reviewer approve, return or carry forward?
Approve the stated scope and evidence, not merely a zero variance. The illustrative review resolves the data discrepancies while retaining Beacon’s commercial follow-up. An open business action is different from an unresolved cost-completeness issue.
| Check | Result |
|---|---|
| Period and currency | Matched for all included sources |
| Usage-to-provider bridge | $1,200 + $100 − $20 − $40 = $1,240 |
| Provider invoices | $1,040 + $200 = $1,240; residual $0 |
| Customer AI allocation | $620 + $430 + $190 = $1,240; unassigned $0 |
| Other delivery costs | $120 tools + $90 shared infrastructure + $50 human review = $260 |
| Total included cost | $1,240 + $260 = $1,500 |
| Operational revenue | $1,500 + $500 + $400 = $2,400 |
| Contribution | $2,400 − $1,500 = $900; 37.5% |
| Evidence completeness | Assumed complete for this fictional example; total agreement alone does not prove completeness |
| Review decision | Illustrative approval of stated operational scope; retain Beacon follow-up |
| Prepared / reviewed | FinOps preparer / Finance reviewer, fictional roles, 2026-11-10 |
| Revision policy | Preserve v1 and issue v2 for later corrections; no real sign-off is represented |
If an invoice has not arrived, usage remains unpriced, or revenue is unresolved, follow your finance policy for a provisional close. Show the estimate or exclusion, its exposure, owner and due date. Do not silently mark the pack final. Later corrections should produce a new version that references the original review and explains the change.
The downloadable pack includes the usage bridge, customer input schedule, revenue basis, margin output, discrepancy register and review record. It is a static example for inspection, not a live accounting system or an actual product-generated report.
How can you run this review with Ganivra?
Start by proving one provider comparison, then add customer and revenue context. The free AI Invoice Reconciliation tool accepts supplied OpenAI usage costs and manually entered invoice figures for one USD account and period. It does not ingest raw token exports or invoice PDFs, and it does not perform the customer revenue schedule in this article.
For an ongoing review, the AI spend-close workflow describes provider comparisons, allocations, exceptions and review history. Connect the relevant evidence and use the customer-margin workflow to investigate account-level results. Check the plan requirements and integration guide for your scope.
Frequently asked questions about monthly AI reconciliation
How do I reconcile AI usage with provider invoices each month?
Match the provider account, services, currency and billing period. Compare recorded usage with invoice usage charges, investigate missing or duplicate events and rate differences, then apply documented credits and adjustments. Keep an evidence-backed bridge from initial usage to the invoice net subtotal and preserve unresolved differences.
What should an AI monthly close pack contain?
Include the reporting scope, usage snapshot, provider invoices, reconciliation bridge, customer allocation, revenue schedule, discrepancy register and final review. Preserve source references, rate and allocation versions, owners, dates and unresolved items so another reviewer can reproduce the result.
Why does my AI invoice differ from my usage dashboard?
Possible causes include different account scope, billing cutoffs, late events, duplicate telemetry, negotiated prices, credits and rounding. An invoice mismatch is a reason to investigate, not proof of an overcharge. Matching totals also do not prove that individual records are complete and correctly attributed.
How do I connect reconciled LLM costs to customer revenue?
Map costs to stable paying-customer IDs, allocate shared delivery costs using documented drivers, and join revenue for the same period on a stated basis. Contribution is that revenue minus included costs. Keep missing revenue, unpriced usage and unassigned costs visible rather than treating them as zero.
Should I add provider invoices to the costs already recorded from usage?
No, not when the invoice and usage records describe the same cost. Use the invoice to reconcile or adjust that cost, not to add it a second time. Separate genuinely additional charges and preserve their source and allocation rule.
Can I close the month with unresolved AI cost discrepancies?
Follow your finance review policy. Record each unresolved item with its amount or unknown exposure, evidence, owner and due date. If a provisional review is permitted, label it clearly and document approved estimates or exclusions. Do not silently force a difference to zero or present incomplete margin as final.
Is reconciled customer contribution the same as accounting gross margin?
No. This walkthrough uses an operational revenue allocation and explicitly included delivery costs. Accounting gross margin requires the appropriate recognized revenue and complete applicable cost of revenue. Reconciliation of one cost category does not make the whole report an accounting statement or net-profit calculation.
Can Ganivra help prepare an AI spend close?
Ganivra describes an AI spend-close workflow with provider comparisons, allocations, exceptions and review history. Its free invoice reconciliation tool compares supplied OpenAI usage costs and manually entered invoice figures for one USD account and period. Customer revenue matching and the full monthly review require additional records and workflow steps.
Inspect a close pack before building your own
Follow each adjustment from provider usage through the final customer review.
Download the sample close pack →Method: original fictional example by Ganivra Editorial. Product and framework references reviewed November 10, 2026. The sample does not assert accounting compliance or guaranteed financial outcomes.