Skip to main content
DoneThat

AI Adoption GuideFinanceClose

Accrual auto-suggestion

LLM proposes period-end accruals from open POs, contracts, and prior patterns.

Finance processPlanBudgetInvoiceCollectPayCloseReportAudit

By Don, DoneThat’s AI coach · updated

Cite the PO, clause, or pattern, or leave the line empty

A period-end accrual draft is usable only when it carries an amount, the accounts, and a source that supports both. Valid sources are an open purchase order with unreceived or unbilled quantity, a contract clause that creates an obligation in this period, or a prior-period accrual that reversed cleanly and still matches current activity. If none of those exist, the correct draft is empty. A blank line is the quality outcome, not a failed run.

The model proposes. The controller posts. Treat every suggested debit and credit as a working paper, the same way you treat a first-pass bonus or prepaid accrual from staff. Quality means you can accept, edit, or reject from the cite. Filling a number so the close "looks complete" is the error this page exists to prevent.

In a Workday, SAP, FloQast, or BlackLine close, treat POs, contracts, and prior journals as the source class. The suggestion step should read those records and return a proposal with a cite. It does not replace posting control, materiality, or the decision that nothing was incurred.

Pull open POs, contracts, and prior-period accruals before you propose

Load three source sets, in this order, cut off at period end. Do not propose until they are current.

Open purchase orders. You need remaining quantity, unit price, receiving status, and invoicing status. Goods received and not invoiced need the receipt plus the PO line. Services against a time-and-materials PO need evidence of performance in the period, not only an open commitment. A blanket PO with no receipt and no performance note is not an accrual source.

Contracts. Pull clauses that create a period cost: minimums, retainers, usage true-ups, and professional fees billed in arrears. Put the clause identifier and the measurement basis on the draft. If the clause is a lease payment schedule, do not accrue it here. Send that obligation to the lease accounting agent and keep this queue for non-lease spend.

Prior-period accruals. Bring the last close's accrual journal, whether it auto-reversed, and what actually billed. A repeating pattern is a source only when the same vendor, account, and activity still exist. A reversed accrual that never billed is a warning against copying the amount. It is not a template.

Align the pull with receiving and AP. If three-way match already cleared the receipt and the invoice, do not accrue the same PO. The three-way match agent confirms goods and bill already met. Accrual auto-suggestion covers the gap where the PO or contract is live and the invoice is not.

Walk one unbilled receipt from source to draft

One example. Facilities has an open PO for HVAC filters at month-end. The warehouse received two pallets on the 28th. AP has no vendor invoice. The PO line has quantity, unit price, and the supplies account.

The draft, in substance: debit supplies (or inventory, if that is policy) for quantity received times PO unit price, credit accrued liabilities, cite the PO line and the receipt, and note AP unmatched. The cite is what makes the draft usable. The same open PO with no receipt is a commitment. It is not an accrual.

Policy still belongs to you. Some teams accrue GR/IR at standard and true up on invoice. Some accrue only after goods receipt. The suggestion must follow the policy you loaded, not a generic rule to accrue every open PO. If policy is receipt-based and there is no receipt, the line stays empty even though the PO is open.

After you post, the entry should reverse on schedule or clear when the invoice arrives. If the invoice amount differs, that difference is a price or quantity variance. It is not a reason to have invented a different number at cutoff.

Do not invent a run-rate to fill a gap

Three failure modes belong in the review notes so a second reviewer can see what you refused.

Accruing without a source. An AP vendor master record, a budget line, or a habit of "we always book something for legal" is not a source. No open PO, no clause, no prior-period pattern that still matches: leave the amount blank. A blank you later replace with a late invoice is easier to defend than a posted number with no document.

Treating the draft as posted. A suggestion queue is not the subledger. Until you approve and the journal hits the GL, or the accrual book that feeds it, nothing is recorded. Do not tell FP&A the close is done because every row in the queue has a number. Unposted drafts are not accruals.

Inventing a run-rate. Averaging recent invoices and booking that average as this period's cost, without a PO, clause, or repeating accrual that reversed and rebilled, is invention. Run-rate is a forecast move. Accrual accounting needs an obligation incurred this period. If you face a known event with no document yet (a claim or a dispute), write a judgment entry yourself, with a memo. Do not let auto-suggestion mint that number.

If a draft cites "historical average" or "typical monthly spend" and cannot open a document ID, reject it. Demand the PO, the clause, or the prior journal. If those cannot be produced, the proposed amount is blank.

Review the draft, then post it yourself

Work the queue the way you work any close checklist.

Read the cite before the number. If the cite does not open a PO line, a clause, or a prior journal, reject. If it opens but remaining quantity or the clause's measurement does not support the amount, edit. If it is clean, post through the normal journal path: preparer, reviewer, system of record.

Keep auto-post off for this queue. Skipping the approval click is how the draft gets treated as the entry. Segregation of duties should match manual accruals: a person posts.

After post, reconcile accrued liabilities. The account reconciliation agent should see the new balance, the supporting PO and receipt list, and the reversing convention. If recon and the suggestion queue disagree on what hit the GL, stop and fix the journal before you certify.

Check the journal's shape, not only the amount. A debit to a P&L account that vendor never uses, a credit in the wrong company code, or a round figure with no quantity extension belongs in journal entry anomaly detection before certify.

Materiality stays with you. A cited supplies accrual below threshold does not need to clutter the GL. Keep the cite in the file and skip the post. A cited material retainage should not sit in the queue because the calendar is tight. The tool proposes. You prioritize.

Keep suggestions from double-counting work already done in close

Auto-suggestion fails when it books what match, recon, or lease accounting already handled. If match cleared the invoice, do not accrue. If last period's accrual is still on the books because reverse failed, do not layer a second entry on the same PO. If the contract is a lease, the lease engine owns the liability.

Put a short exclusion list on the run: POs already matched this period, POs already accrued this period, lease contracts, intercompany, and items under threshold if you use one. Exclusions protect the empty state. The model should not "find" an amount by wandering into accounts you already closed.

The certify test is narrow. Every posted accrual this period cites an open PO, a contract clause, or a prior-period pattern that still applies. Every suggested row without a cite is blank and unposted. You can open the cite from the journal. The books do not contain a run-rate invented to avoid a blank.

Load sources. Propose with cites. Leave blanks. You post.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first