Bank Statement OCR Quality Checklist

Establish image quality before reviewing extracted text

Inventory all pages and verify orientation, sharpness, and readable scale. Inspect edges for cropping and note fingers, glare, shadows, folds, security patterns, or compression; a cropped margin may erase a value column even when descriptions look intact.

OCR acceptance record

Store the acceptance evidence beside the extraction so its limits remain visible to later reviewers.

Page coverage

List expected and received pages, orientation corrections, and every cropped or unreadable region.

Row sample

Trace the opening, middle, and closing transaction of each page to its image.

Arithmetic result

Preserve calculated and printed movement together with any difference that lacks an explanation.

Exception count

Count fields still under review and state whether they change totals or descriptive text only.

Test row and column reconstruction

On each page, trace at least its first, middle, and final transaction back to the image. Verify that dates remain attached to descriptions and that debit, credit, and balance figures have not shifted; wrapped descriptions often create false rows.

Look for numeric values misplaced in descriptions and words misplaced in amount fields. Extraction can misread thousands separators, decimal conventions, parentheses, trailing debit or credit markers, and currency symbols, changing the value's meaning.

Use sampling designed to expose structural failures: trace every page's first, middle, and final transaction, all wrapped rows, negative or parenthesized values, and page transitions. Compare source transaction-line count with extracted rows afterward. A correct opening page cannot validate continuation pages with different layout, type size, or debit and credit arrangement.

Use a sampling plan that targets layout risk. Check the first, middle, and final row on every page, every wrapped description, every negative or parenthesized amount, page transitions, and rows near totals. Compare the number of visible transaction lines with extracted rows. Then inspect every field flagged by low confidence or an arithmetic break. Randomly checking a few clean rows from the first page can miss a continuation page where columns shift or a scanner shadow turns a debit into a different value.

Use arithmetic as a detection tool

Rebuild movement from the opening balance through debits and credits, including interest, fees, and adjustments, then test several running balances. When a difference repeats from one row onward, that row is a strong candidate for the first sign or amount error.

A calculation can expose inconsistency but cannot supply a missing digit. Verify the value against the page image; if the source remains illegible, record uncertainty rather than choosing the number needed to make the balance work.

A balance variance points to possible error but does not authorize a replacement value. If a 90.00 difference begins after one row and persists through every later balance, inspect for a missing 90.00, a 9.00 misread as 99.00, or reversed direction, and verify candidates against the image. If the source stays illegible, leave the amount unresolved and identify the affected totals.

Arithmetic can identify the location of an error but not authorize a guessed correction. When every later balance is 80.00 away from the printed figure, inspect the first affected row for a missing 80.00 amount, an 8.00 value read as 88.00, or a reversed direction. Return to the source image and record the verified reading. When the digit remains unreadable, retain the extracted value, mark the field unresolved, and state which totals are affected. A plausible number that restores the balance is still an inference until the document supports it.

Record uncertainty at field level

Confidence belongs at field level because a readable date and description can coexist with an uncertain amount. Attach the note to the affected field instead of declaring the entire document accurate or inaccurate.

If the source file fails before extraction begins, use the bank statement PDF upload troubleshooting steps before changing or re-exporting the document.

After the rows pass OCR checks, continue with the bank statement data analysis workflow and keep uncertainty flags beside any interpreted field.

Establish image quality before reviewing extracted text

OCR acceptance record

Page coverage

Row sample

Arithmetic result

Exception count

Test row and column reconstruction

Use arithmetic as a detection tool

Record uncertainty at field level

Run the next quality check

How to Analyze Bank Statement Data

Bank Statement Spreadsheet Quality Checks

Bank Statement Reconciliation Guide

Apply this guide to a statement