We successfully automated our three-way invoice matching process using AgilePoint Process Designer, eliminating manual reconciliation bottlenecks in accounts payable. Previously, our AP team spent 4-5 hours daily matching purchase orders, goods receipts, and vendor invoices manually, leading to payment delays and occasional duplicate payments.
Our implementation leverages OCR data extraction from incoming PDF invoices, API integration with our ERP system to fetch PO and GR data, and custom business rule validation to handle matching tolerances. The workflow automatically routes exceptions to AP specialists only when discrepancies exceed defined thresholds (±2% for quantity, ±5% for amount).
Since deployment three months ago, we’ve achieved 87% straight-through processing for standard invoices, reduced processing time from 48 hours to 4 hours average, and cut matching errors by 94%. Happy to share our approach for anyone tackling similar AP automation challenges.
The business rule validation piece interests me most. How complex are your matching rules? We have situations where partial deliveries, price adjustments, and freight charges make simple three-way matching insufficient. Did you handle these edge cases in your initial deployment or phase them in later?
How did you structure the API integration with your ERP system? We’re running SAP and concerned about performance impact from real-time API calls during invoice processing. Did you implement any caching strategy or batch processing?
We’re also on SAP (ECC 6.0). We use AgilePoint’s REST connector to call custom SAP OData services that retrieve PO and GR data. Performance was initially a concern, so we implemented a hybrid approach.
For high-volume vendors (invoices daily), we run a nightly batch job that caches open PO data in AgilePoint’s database using custom entities. Real-time processing queries this cache first, falling back to live SAP calls only for recent POs not yet cached.
For low-volume vendors, we make direct API calls since the overhead is minimal. We also set connection pooling and timeout parameters aggressively - 5 second timeout forces exception routing rather than hanging workflows. Average API response time is under 800ms for cached data, 2-3 seconds for live SAP queries.
This is exactly what we’re trying to implement! Could you elaborate on your OCR data extraction approach? We’ve been evaluating different OCR engines and struggling with accuracy on invoices with varying formats from different vendors. What extraction accuracy rates are you seeing, and how do you handle extraction failures or low-confidence fields?
We integrated ABBYY FlexiCapture via REST API for OCR extraction. Our accuracy is around 92-95% for structured invoices from regular vendors, dropping to 75-80% for one-off suppliers with non-standard formats.
For low-confidence extractions (below 85% confidence score), the workflow automatically creates a manual review task with the original PDF and extracted data side-by-side. The reviewer corrects any errors, and we feed those corrections back to train the OCR model. We also maintain vendor-specific templates for our top 50 suppliers which significantly improved accuracy.
The key was not expecting perfect OCR - build exception handling into your workflow from day one.
From an audit perspective, how do you maintain the audit trail for automated approvals versus manual interventions? Our auditors require detailed logs showing who approved what and why, especially for exceptions that were auto-approved within tolerance thresholds.