Workflows · GL and close · Tier 1 · first automations
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.
By hand: Monthly, three to six hours · Any ERP shop: energy, manufacturing, utilities, distribution, professional services.
The card, as it runs by hand
- Input
- The AP open-item extract as of month end (company, vendor, document type, document date, GL date, amount, offset account) and the trial balance report, which arrives as a PDF print.
- What I do to it
- 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.
- Output
- 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.
- Who checks
- The accounting supervisor, who replies with the approval. Looks for zero differences, timing items explained, anything marked for investigation.
- How often
- Monthly, business day three.
- How long
- Three to six hours by hand.
- How long now
- Measured in the course after the first governed run, per card. Not estimated here.
As a governed workflow
- Stays in code
- The pivot, the lookup to the control account, the comparison, and the difference per account. A query. It does not change and it is not AI.
- The AI does
- Read the trial balance PDF into rows. Classify each difference as timing or error. Draft the one-line explanation and the approval email in the reviewer's format.
- A person still
- The supervisor reads the workbook and replies with the approval, exactly as today. Nothing posts. Nothing changes in the ledger.
- The log
- Every run records the as-of date, the files it read, the differences it found, and who approved.
AI does the preparation. A named person approves it. Nothing posts to the books without that approval, and every run is recorded.
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.