OCR Extraction Error Audit Sample Worksheet
You cannot tune document capture without measuring it on your own paperwork. This worksheet records a sample audit by document type and computes field accuracy and the share of fully clean documents, ready to export to Excel.
| Document type | Source or supplier | Documents sampled | Fields checked | Fields correct | Fields wrong or missing | Docs with at least one error | Field accuracy | Clean document rate | Most error-prone field | |
|---|---|---|---|---|---|---|---|---|---|---|
8 | 98.7% | 85% | ||||||||
29 | 93.6% | 53.3% | ||||||||
17 | 93.2% | 56% | ||||||||
42 | 89.5% | 35% | ||||||||
9 | 96.3% | 76.7% |
Compare the extracted value with the source document, not with another system. Sample documents at random within each type and include your poorest sources. Rows are examples: replace them with your audit data.
Audit automation
Optional: ask about automating the sampling and scoring from your document store.
Optional. The tool and every export work without this. No spam.
How to use the worksheet
Pull a random sample of processed documents from each type and supplier group. For each row, count how many documents you sampled, how many fields you checked, how many were correct, and how many documents had at least one error. Name the most error-prone field so that follow-up targets the right problem. The sheet computes field accuracy and clean-document rate per row and in total.
- Check against the original image, not the ERP record.
- Sample at least a few dozen documents per type where you can.
- Add one row per supplier or source if they differ in quality.
What the numbers mean
Field accuracy is correct fields over checked fields. Clean-document rate is documents without any error over documents sampled. The two differ because errors cluster: a bad scan produces several wrong fields in one document. Both numbers are useful. Field accuracy tells you how the model performs; clean-document rate tells you how much review or rework to expect.
Using the results
Sort rows by clean-document rate to find the weak sources. A scanned certificate from one vendor often accounts for most errors, and a template, a request for digital documents or a tighter confidence threshold for that source can fix it. Repeat the audit after changes, and keep the sheet to show auditors that extraction quality is monitored.
Common mistakes
Sampling only documents that were flagged for review hides errors that went through. Mixing document types in one row blurs the picture. And changing the list of fields between audits makes results incomparable, so fix the field list per document type first.
Frequently Asked Questions
How many documents should I sample?
Enough to see a stable result for each document type and the main sources, often a few dozen per group at minimum. Larger samples are needed to detect small changes or rare fields.
What counts as a wrong field?
Any checked field whose extracted value differs from the source in a way that matters, including missing values. Decide on rules for formatting differences, such as dates and currency symbols, before you start.
Why track the most error-prone field?
It shows where effort pays off. If one field such as lot number or tax amount causes most errors, targeted rules or training on that field are cheaper than retraining everything.
How often should I run the audit?
Monthly or quarterly is common, and after any change to the model, templates, suppliers or scan process. Regular sampling also gives you evidence for quality audits.
If you want sampling and scoring automated against your ERP and document store, Netray can build it on-prem with human approval on any write-back.
Related Tools
OCR Field Accuracy to Document Accuracy Calculator
See why 98 percent field accuracy does not mean 98 percent clean documents, and what field accuracy a target requires.
AI Agents & AutomationExtraction Precision, Recall and F1 Calculator
Score a document extraction or classification model with precision, recall and F1 from a labeled sample.
AI Agents & AutomationOCR Human Review Queue Staffing Calculator
Size the number of reviewers needed for a document capture exception queue, including peak days and the impact of the confidence threshold.
Costing & FinanceInvoice Processing Cost Per Invoice Calculator
Work out your fully loaded cost per invoice including labor, software, OCR page fees and exception handling outside AP.
More Quality & Inspection tools
See all- Cpk Confidence Interval Calculator
- Pp Ppk Process Performance Calculator
- DPMO and Sigma Level Calculator
- X-bar and R Control Chart Limits Worksheet
- Control Chart Constants Table (A2, D3, D4, d2, c4)
- P Chart Control Limits Calculator
- Defect Pareto Analysis Worksheet
- DMAIC Project Charter Template
- SIPOC Diagram Template