Skip to content
← Back to research

Evaluation study / Evaluation & assurance

Published by HephaKnotSupplier document processingStudy date not reported
Office workstations illustrating the document-processing setting.
Office workstations illustrating the document-processing setting.

Overview

An Asian food-and-beverage group spent too much time processing supplier documents by hand. It used Knotest to assess an AI provider before committing to the system.

Method

The team tested poor scans, overlapping stamps, and inconsistent formats. It agreed on performance targets and rollout criteria with the provider.

Findings

A one-month A/B study compared manual and AI processing of invoices, delivery orders, and credit notes. Processing time fell by 65.3%, with improvements in accuracy and error rate.

Scope & limitations

The evaluation compared manual and AI processing of invoices, delivery orders, and credit notes over one month. The available summary does not report document counts, per-type baseline values, or uncertainty estimates.

Data & evidence

This page preserves the published project summary. Raw data, individual test results, and reproducible code are not supplied with this record. Contact HephaKnot with questions about the evidence or method.

Original project account

Processing supplier invoices, delivery orders and credit notes had become a major bottleneck for an Asia-based multinational F&B group. Poor scans, stamps covering text, and inconsistent formats made the work slow and error-prone.

“Most of the time, our team was just typing, checking, re-checking,” said operations manager Mr. Lee. “When documents looked different or scans were bad, everything slowed down. Errors would slip through.”

Before committing to an AI solution, the group turned to Knotest, developed by HephaKnot, to validate the decision with real data.

First, they ran a model performance test targeting tough scenarios: poor scan quality, stamp obstructions and inconsistent formats. The Intelligent Document Processing System passed. Next came a one-month A/B Business Impact Assessment, with Group A using manual processing and Group B the AI system.

The results were clear:

  • Average Processing Time per Document: (−65.3%)
  • Documents Processed per Hour: (+188.5%)
  • Document Processing Accuracy: (+10.6%)
  • Error Rate: (−69.6%)
  • Daily Manual Workload (Team): (−32.6%)

“Once you see the data, the choice is obvious,” Mr. Lee said. “This isn’t just faster — it’s more reliable. Our team can focus on what actually matters now.”