Touchless SAP Vendor Invoice Posting at an OEM Manufacturer — 100K Invoices Indexed, 95% Exception Coverage

MindMap deployed a vendor-invoice automation platform composed of DocuMage (Dm) as the document-intelligence layer for the multi-format invoice ingestion, Workflow Automator (Wa) for the SAP-posting integration, Compliance Engine (Ce) for the validation-and-routing rules, and Anomaly Detector (Ad) for the duplicate-and-fraud-pattern detection.

Phase one was the invoice-format cataloguing. We catalogued the top fifty supplier invoice formats by volume, mapping each format's field-positioning, detail-level and variation patterns. The catalogue covered approximately 80% of inbound invoice volume; the residual long-tail formats fed into the DocuMage extraction layer's general-format-handling pattern.

Phase two was the multi-format ingestion-and-extraction workflow. DocuMage handles the inbound invoice across PDF, Excel, fax (post-OCR) and scanned-PDF channels. The extraction produces the structured invoice-data fields (vendor, date, amount, line-items, tax-position) with per-field confidence scores. Where the confidence on a field is high, the field flows through the validation-and-posting path; where the confidence is low, the field routes to the human-review path.

Phase three was the validation-and-posting layer. Compliance Engine validates each invoice against the SAP master-data (vendor-master, cost-centre-master, G/L-account-master, tax-code-master) and against the procurement-policy rules (PO-match for PO-based invoices, accrual-handling for non-PO invoices, contracted-pricing-validation for contracted-supplier invoices). Validated invoices route to Workflow Automator for SAP-posting with the appropriate G/L-account, tax-code and cost-centre assignment.

Phase four was the exception-handling-and-fraud-detection layer. Exception invoices route to a unified human-review UI with the invoice-image, the extracted fields, the validation-failure-reason and the recommended-resolution-action presented in a single screen. Anomaly Detector runs continuously against the posted-invoices stream to identify duplicate-invoice candidates and fraud-pattern matches, with flagged candidates routed to the fraud-review workflow.

The platform runs on the manufacturer's private cloud environment integrated with the existing SAP estate. The multi-channel ingestion uses dedicated email-mailbox ingestion, fax-gateway ingestion (with automatic fax-to-PDF conversion), supplier-portal upload and scanned-document ingestion from the central scanning hub.

DocuMage's invoice-extraction pipeline is layered. The first pass uses a base OCR model for the text-extraction-and-field-localisation; the second pass uses a fine-tuned vision-transformer ICR model trained on the manufacturer's vendor-invoice corpus for the handwritten-field handling; the third pass uses an LLM in a constrained schema-validated prompt for the cross-field reconciliation (the line-item-totals-versus-invoice-total validation, the tax-position-versus-tax-code validation).

Compliance Engine's validation-and-routing logic encodes the manufacturer's procurement-and-AP policy library. The PO-match logic compares the invoice line-items against the PO line-items with tolerance-based matching for the typical quantity-and-pricing variations; the accrual-handling logic supports the non-PO invoices with appropriate cost-centre-and-G/L-account derivation; the contracted-pricing logic validates against the contract-management system's pricing terms.

Workflow Automator's SAP-posting uses SAP's standard inbound BAPI and IDoc interfaces. The posting integration handles the per-invoice transaction with appropriate rollback-and-recovery on posting failures; the line-item-level posting supports the detailed cost-attribution that the manufacturer's cost-accounting framework requires.

Anomaly Detector's duplicate-and-fraud-pattern detection uses a combination of exact-match rules, fuzzy-match rules and statistical-anomaly detection. The detection covers the duplicate-invoice patterns (same vendor, same amount, same invoice-date within tolerance), the fraud-pattern matches (vendor-payment-instruction-changes followed by atypical-amount invoices, sequence-anomaly patterns) and the cross-invoice spend-pattern anomalies that suggest fraud-and-leakage.

Reporting dashboards provide the per-day invoice-volume statistics, the per-vendor processing-metrics, the exception-and-fraud trend analysis and the supplier-payment-cycle adherence reporting. The audit trail captures every invoice-lifecycle event with the full context preserved.