How many statements should I test?
Five is a useful screening minimum, not statistical proof. For a purchasing decision, include multiple examples of every high-volume bank and layout you rely on.
Free, vendor-neutral worksheet
Compare converters on the same redacted bank statements instead of trusting a headline accuracy claim. Use the private browser calculator or download the scorecard to measure extraction errors and human correction time.
This is a testing methodology and blank worksheet, not a cross-vendor ranking. StmtPilot does not publish competitor accuracy numbers without running the same disclosed corpus under the same conditions.
Runs entirely in your browser
Enter aggregate counts from one converter test. The calculator does not upload or store your data.
Row recall
100.00%
Duplicate rate
0.00%
Critical-field error rate
0.00%
Description error rate
0.00%
File failure rate
0.00%
Total correction time
0.0 min
Strict pass
Strict pass requires no failed files, missing or duplicate rows, and no date, amount, or balance errors.
Use at least five redacted statements from the banks, account types, and PDF conditions you actually process. Include a difficult file; do not test only the cleanest sample.
Manually count the transaction rows and record the correct dates, descriptions, amounts, signs, and balances before looking at any converter output.
Upload the same files, use default settings, record failures, and export the closest comparable spreadsheet format from every product.
Count missing and duplicate rows, then count incorrect fields. Time the review and cleanup work required to reach an acceptable final file.
Report digital and scanned PDFs separately, and split results by bank or layout. A blended average can hide a product that fails your most common statements.
| Metric | Formula | Why it matters |
|---|---|---|
| Row recall | (expected rows − missing rows) ÷ expected rows | How completely the converter found the real transactions. |
| Duplicate rate | duplicate rows ÷ expected rows | How much false extra data must be removed. |
| Critical-field error rate | (incorrect dates + incorrect amounts + incorrect balances) ÷ expected rows | Errors most likely to break reconciliation or downstream imports. |
| Description error rate | incorrect description fields ÷ expected rows | Rows whose payee or memo text needs material correction. |
| Failure rate | failed files ÷ files tested | Statements that do not produce a usable export at all. |
| Human correction time | review minutes + export cleanup minutes | The labor cost that a headline extraction score leaves out. |
Mark a statement as passing only when there are no missing or duplicate transactions, no incorrect dates or amounts, the balance check passes when the statement provides enough information, and the export can enter the next workflow without material restructuring. Keep description quality as a separately reported measure because acceptable memo text depends on the downstream use.
First eliminate tools that fail your security or required-format constraints. Next compare statement pass rate and critical-field errors. Use correction time as the tie-breaker, then model the monthly cost at your real page volume. A product that costs less per page can still cost more after senior bookkeeping review.
Five is a useful screening minimum, not statistical proof. For a purchasing decision, include multiple examples of every high-volume bank and layout you rely on.
Test both if both occur in your workflow, but report them as separate segments. OCR and digital text extraction have different failure modes.
Only if the vendor defines the unit being measured, discloses the corpus and exclusions, and the test resembles your files. Field accuracy, row accuracy, and statement pass rate are not interchangeable.
No. StmtPilot publishes narrow, evidence-gated bank support and provides a review-first workflow. Use this worksheet to test it against alternatives on your own redacted documents.