How accurate is OrderPilot, really?

OrderPilot averages 99.2 % field-level accuracy across our customer base - but that number hides useful detail. Here’s what actually happens.

Day one accuracy

On the first PO from a brand-new supplier with no prior template, expect:

• Header fields (supplier, PO number, date, total): ~97 %
• Standard line-item fields (qty, unit price, line total): ~94 %
• Edge-case fields (UoM, delivery address, payment terms): ~88 %

These numbers come from our confidence-calibrated extraction, not marketing.

After 30 days (with corrections)

By the time you’ve processed 20–50 POs from a supplier, the model has learned that supplier’s template, terminology, and corrections. For established suppliers we see:

• Header fields: 99.8 %
• Line items: 99.2 %
• Edge fields: 97.5 %

What drives accuracy

Three things, in order:

1. Supplier template consistency - if a supplier sends the same format every time, we nail it
2. Your correction feedback - every correction you make in OrderPilot improves extractions from that supplier
3. SKU and vendor master completeness - without good reference data, we have nothing to match against

What doesn’t drive accuracy

Common things customers worry about that mostly don’t matter:

• Scan quality - our OCR handles low-res scans down to ~150 DPI
• Handwritten notes - these are always flagged for manual review (by design)
• Multi-page POs - handled natively, no config needed
• Non-standard formats - we learn per-supplier, so unusual formats are fine as long as they’re consistent

The 100 % claim on the homepage

The “100% accuracy” headline refers to downstream document-level accuracy after validation queue review: no PO enters your ERP with an uncorrected error. It’s a claim about process, not raw extraction.

If you want the raw extraction numbers for diligence, ask for our statistical accuracy report during onboarding.