Invoice Reconciliation

The situation

The AP team was manually matching roughly 200 supplier invoices per month against purchase orders and goods receipts in SAP. Each invoice arrived as a PDF — across 14 different supplier formats — and had to be checked line by line against the corresponding PO and GR records. The process consumed 32–40 hours of coordinator time per month.

The deeper problem was what wasn't being caught. The estimated overbilling miss rate was 30–40% — invoices where the supplier had billed above the agreed price, for quantities not received, or for items already invoiced. Because exceptions weren't being tracked systematically, the true cost of the miss rate was unknown. When it was finally quantified, it changed the conversation.

The AP team had also built informal workarounds: a personal dispute tracking spreadsheet maintained by the senior coordinator, a shared folder of "problem vendor" notes, a manual double-check ritual on invoices from specific suppliers. These shadow systems contained the institutional knowledge the tool would need.

The approach

Invoice reconciliation is a deterministic problem — a match holds or it doesn't. The right tool is not a language model; it's matching logic. Mimir built a Python-based three-way match tool: each incoming invoice is checked against the corresponding purchase order and goods receipt exported from SAP. Clean matches route to a confirmation queue. Exceptions are classified by reason code — price variance, quantity mismatch, missing PO, duplicate invoice, unrecognized format — and queued for AP investigation.

The tolerance rules were formalized from what the AP Manager had been applying informally: a 1% price tolerance with a $25 floor, exact quantity match required. Nothing writes to SAP, and nothing releases a payment. The AP coordinator confirms each matched invoice before SAP entry; the AP Manager reviews the exception queue and approves the payment batch. Human judgment stays where the consequences are.

The tool runs on the existing Windows network. SAP exports PO and GR data on a scheduled basis to a shared folder. The coordinator downloads invoice PDFs from the AP inbox and drops them in an intake folder. The tool processes both, produces a results file, and routes work to the appropriate queue.

The result

In the first month of live use, 78% of invoices matched straight-through. The AP team processed the same volume in 14 hours instead of 40. The tool surfaced $14,200 in overbillings — compared to an estimated $3,000 recovered under the prior manual process. Four additional exceptions were caught that the manual review would have missed.

The exception classification by reason code changed how the team works with problem vendors. Patterns became visible: one supplier had a consistent price variance issue that had been treated as individual mistakes. The AP Manager used the data to open a formal conversation with that vendor — something that hadn't been possible without structured evidence.

The vendor notes column — added in the second test cycle based on the senior coordinator's suggestion — was one of the highest-impact additions. It captured the institutional knowledge that had been in her personal spreadsheet, making it available to both coordinators on every invoice from a known problem vendor.