Accrual identification with AI: find the missing expenses before your auditor does

When to use it

Use it early in the close, when you're trying to catch services received but not yet invoiced. It's most useful for teams without a formal accrual-request process from department heads, or as a completeness check on top of one.

Input data

• 12 months of AP/expense history by vendor and month (vendor, GL account, invoice date, service period if captured, amount).
• Current-period AP invoices booked so far.
• Open purchase orders with received/not-invoiced quantities and amounts, if you use POs.
• Invoices received or payments made in the first [10] days after period-end (the subsequent disbursements list).
• Last period's accrual schedule and what reversed.
• Your accrual threshold (e.g. accrue items ≥ $[X]).

The prompt

Replace the [bracketed] items. Paste it with your files attached. It works in any capable assistant; tool-specific notes are in the steps.

You are helping identify potential unrecorded liabilities (accruals) for the period ending [date].

Attached:
- Vendor expense history, last 12 months by month: [file]
- Current period AP booked to date: [file]
- Open POs / received-not-invoiced: [file]
- Invoices and payments dated [period-end +1] through [date]: [file]
- Prior period accrual schedule: [file]
Accrual threshold: $[X].

Identify candidates in these categories, each in its own table:

1. Recurring vendors missing this period: vendors billed in at least [3] of the last [4] months with no invoice booked for this period. Show the monthly history, the expected amount (the average or the last amount, saying which you used) and the variance in billing pattern.
2. Post-period invoices for current-period services: invoices dated after period-end whose service period or description indicates [month] service. Quote the description that supports it.
3. Received not invoiced: PO receipts with no matching invoice.
4. Prior-period accruals with no reversal or no subsequent invoice: possible stale accruals to release.
5. Known contractual items: rent, SaaS, retainers and similar that look fixed-fee but are missing.

For each candidate: vendor | account | suggested amount | basis for amount | evidence | confidence | above threshold (Y/N).

Do not recommend booking anything. Present candidates only. If the evidence for a service-period is ambiguous, say so. Don't infer a service period from the invoice date alone.

Steps

1. Run the prompt on day 1–3 of close, once the AP cut-off has been applied.
2. Send the candidate list to budget owners or department heads to confirm services received and the amounts.
3. The preparer decides each accrual and documents the estimate method (invoice, contract, PO, historical run-rate).
4. Book approved accruals through your normal JE process, with auto-reversal where that's your policy.
5. Next month, compare the accrual to the actual invoice and feed the differences back into your estimates.

Control map

What the AI may do, step by step. How ratings work.

Audit trail

• Keep the candidate list and mark each item booked, not booked (with reason), or already recorded.
• Document the estimate basis for each booked accrual. "AI suggested" is not a basis; "average of last 3 invoices per AI-prepared history, confirmed with budget owner on [date]" is.
• A completeness check like this is good evidence for your search-for-unrecorded-liabilities support, so save it.

Failure modes

• Seasonality and annual billing read as "missing" monthly vendors, producing false positives.
• Vendor terminated last month: the AI expects a bill that will never come.
• Invoice date vs service date confusion. It can lead to double accruals when the invoice was already booked in the period.
• Averages that hide a price increase or usage spike. Confirm the amounts with the budget owner on large items.

Data handling

• Use a company-approved AI workspace whose terms say your inputs are not used for model training. Check your plan; don't assume it.
• Strip or mask what the task doesn't need: employee names, bank account numbers, customer PII, SSNs or TINs.
• If you are a public company, treat pre-release results as material non-public information and keep them inside approved systems.