Finance Workflow Examples · GL and close · Tier 1 · first governed workflows
AP sub-ledger to GL control-account reconciliation
The reconciliation every ERP shop has: open payables by company and offset, tied to the trial balance, with every difference explained as timing or error. The approval email already exists in the process.
Current effort: Monthly, three to six hours · Any ERP shop: energy, manufacturing, utilities, distribution, professional services.
On the one-page 1113 Workflow Card
The 1113 Workflow Card is a structured way to examine one recurring task before deciding where AI belongs: seven fields, one page. It is not, by itself, a production specification, execution log, evidence that a review occurred, or control certification.
- 01 · Purpose and triggerWhy the task exists, when it starts, and what business outcome it supports.
Confirm that the AP sub-ledger agrees to the GL control accounts before the books close, so the supervisor can approve the balance and any error is found before reporting. Starts on business day three each month, when the month-end AP extract and trial balance are available.
- 02 · Inputs and permissionsWhat comes in, where it comes from, its sensitivity, and whether it may be used in the approved AI environment.
The AP open-item extract as of month end (company, vendor, document type, document date, GL date, amount, offset account) from the ERP, and the trial balance report, which arrives as a PDF print. Internal financial data, not personal. Whether the extract may enter the approved AI environment depends on the organization's policy; practice or sanitized files are used until that is confirmed.
- 03 · Current stepsWhat the person does today, in order.
Pivot open items by company and GL offset. Map each offset to its GL control account. Compare to the trial balance line. For each difference: if the GL date is after the activity date, mark timing; otherwise mark for investigation and write one line saying why. Note any report issues. Draft the approval email.
- 04 · Rules, judgment, and AI’s roleWhat must remain deterministic, where judgment is required, and exactly what AI prepares.
Deterministic (computed in code, the same answer every time): The pivot, the lookup to the control account, the comparison, and the difference per account. This is a query and it does not change from month to month.
Needs judgment: Whether a difference is genuinely timing or an error, and whether an unusual item needs follow-up with AP.
AI prepares: Read the trial balance PDF into rows. Propose a timing-versus-error classification for each difference. Draft the one-line explanation per item and the approval email in the reviewer's format. It does not calculate the tie.
- 05 · Output and downstream useWhat is produced, who uses it, and whether it informs a decision or changes another system.
The reconciliation workbook (summary, TB totals, AP-versus-GL by account with difference and classification) and a draft approval email listing zero-difference accounts and explained timing items. Used by the accounting supervisor to approve the month's balance; nothing is written to the ledger.
- 06 · Checks, exceptions, and responsibilityWhat a person reviews, how errors are identified, what happens when something is wrong, and who remains accountable.
The accounting supervisor reviews the workbook and replies with the approval, exactly as today, looking for zero differences, timing items explained, and anything marked for investigation. A difference the AI classified as timing is checked against the dates. Any item marked for investigation goes back to AP through the existing process. The supervisor remains accountable for the reconciliation.
- 07 · Evidence and next decisionCurrent effort, expected benefit, quality measure, test result, and the decision to continue, revise, decompose, or stop.
Current effort: three to six hours by hand each month. Expected benefit: less time on the PDF re-keying and the explanation drafting; a hypothesis until measured. Quality measure: every account ties or has an explained difference, and no AI-classified timing item is later found to be an error. Test result: illustrative example, not yet tested in your environment. Next decision, continue, revise, decompose, or stop: to be established during evaluation.
Control in this example. Nothing changes in the ledger. Any correction follows the organization's existing journal approval.
AI handles defined preparation. A person reviews. Your controls remain in place.
Where it goes wrong
Letting the AI do the arithmetic. It will be right most months, and nobody will check the month it is not. The tie stays in the query.
Why it lands in a room
Everyone has one, the GL-date-versus-activity-date story is instantly recognizable, and the worst outcome of a bad run is an unexplained difference a reviewer catches.
This example illustrates the method. It is not a production-ready workflow, a software specification, or a substitute for your organization’s policies, controls, and approved technology environment.