Understanding Bank Statement Fraud Detection
Spot edited bank statements: a 0 to 100 authenticity score for every statement with the reasons behind it, from edited PDFs and figures that contradict each other to statements that don't match the rest of the applicant's packet.
Last updated
DocuClipper gives every bank and credit card statement an authenticity score from 0 to 100 and lists the reasons behind it. A score of 90 to 100 means the statement appears authentic; below 50 means there are strong signs it was edited after the bank produced it. The score is a guide for your review, not a verdict.
The reasons fall into three groups:
- How the file was made. Whether the PDF was edited after the bank produced it, re-saved with editing software, or carries metadata that disagrees with itself, and whether an amount looks pasted in.
- Whether the statement agrees with itself. Whether the transactions reconcile to the printed opening and closing balances, and whether the printed totals add up to the printed balances.
- Whether it fits the rest of the packet. When you put all of an applicant's statements in one project, each statement is compared with the others: statements of the same account should come out of the same bank system, and each statement's closing balance should be the next one's opening balance.
You'll find it on the Fraud Detection tab of any bank or credit card project, next to Transactions and Categorize. Invoices, receipts and pay stubs have a Fraud Detection tab of their own: see Invoice and Pay Stub Fraud Detection.
Fraud Detection is included on the Business and Enterprise plans, and on every plan during the free trial. On a Starter plan you can use it for 14 days from the first time you open a statement in the tab; after that, the tab explains what it checks and has an Upgrade to Business button that opens the plans page. A plan without Fraud Detection sees the same explanation and button even in a project with no statements yet.
A statement is scored the first time you open it in the Fraud Detection tab (or ask for its fraud report through the AI assistant connection). Scoring runs in the background: the tab says Scoring this document and shows the score as soon as it is ready, usually within a minute. A statement is scored again after its transactions are edited, after another document is added to the project, and when the checks are updated, so the packet comparison always covers everything in it. While that happens the tab keeps showing the previous score with Updating this score below it, and replaces it when the new one is ready.
What the authenticity score means
| Score range | Interpretation |
|---|---|
| 90 to 100 | Document appears authentic |
| 70 to 89 | Minor concerns detected |
| 50 to 69 | Moderate risk detected |
| 0 to 49 | High risk of fraud detected |
The gauge takes the color of the band: green, amber, orange or red.
Each reason carries a confidence. Several confident reasons on the same statement lower the score the most; a single low-confidence reason, such as a layout DocuClipper has not seen before, only nudges it. Genuine statements land in "minor concerns" now and then, for example after being merged or compressed with an online PDF tool, so always read the reasons.
Check a whole packet together
Upload all of an applicant's statements to one project: every month of every account, plus pay stubs if you have them. DocuClipper then compares each document with the rest of the packet and adds what it finds to that document's reasons:
- Different software than the account's other statements. Four statements of an account made by the bank's system and a fifth made by a PDF editor stand out, even when the fifth looks perfect on its own.
- Balance doesn't carry over. Each statement should open at the previous statement's closing balance. When they disagree, one of the two was changed. A month missing in between is not counted as a break.
- Doesn't reconcile while the account's other months do. A reconciliation failure is usually an extraction issue, but when every other statement of the same account reconciles, it deserves a closer look and weighs more.
- Pay stubs against deposits. A pay stub whose net pay never reached the bank account around the pay date is flagged on the stub. See Invoice and Pay Stub Fraud Detection.
Reasons you may see
The Signal Type column names the reason; Details says what was found, with the figures involved.
Statement's Figures Disagree. The statement's printed opening balance, deposit and withdrawal totals and ending balance don't add up, while the transactions match the printed totals. One of the printed balances doesn't fit the rest of the statement, which is what changing a single figure leaves behind. On a scanned statement it can also be a misread digit, so it weighs less there. Review tip: compare the summary box with the daily balances and the last transaction.
Different Software Than Other Statements. This statement's PDF was made by different software than every other statement of the same account in the project. Legitimate causes: printing one month to PDF from the browser (weighs less), or the bank changing its statement system (then every later month changes too). Review tip: ask for that month's statement downloaded straight from online banking.
Balance Doesn't Carry Over. The closing balance of one statement is not the opening balance of the next one of the same account, or two copies of the same month show different balances. The statement that is also inconsistent on its own carries most of the weight. Review tip: line up the two statements' balances.
Doesn't Reconcile, Other Months Do. The transactions don't tie out to the printed balances, while the account's other statements in the project do.
Reconciliation Failed. The transactions don't tie out to the printed opening and closing balances. Legitimate causes: partial statements, excluded pages or a misread row; this is more often an extraction issue than tampering, so it weighs little on its own. Review tip: fix the statement and the score updates.
Edited After Creation. Changes were saved on top of the original file after it was created. Banks produce statements in one pass. Legitimate causes: some signing or annotation tools save the same way.
Metadata Mismatch. The file's two metadata records name different producing software, which usually means an editor rewrote one of them.
Suspicious PDF Producer. The PDF was made with software that is unusual for a bank statement: a word processor or image editor, or an online PDF editor (weaker, since those are also used to merge or compress statements). Saving from a browser or macOS Preview is the normal way to export a statement and is not flagged.
Modified After Creation. The file was last modified long after it was created.
Suspicious PDF Version. The file declares an unusual or very old PDF version for a bank export.
Inconsistent Field Height. An amount is set at a different text size from both the rest of its row and the other amounts in its column, as pasted-in amounts often are. Statements that simply set their amounts, totals or headline balances in a different size from the descriptions do not raise it. Legitimate causes: an unusual one-off figure in the statement's own design. Review tip: zoom in on the amount and compare it with the figures around it.
Layout Fingerprint Mismatch. The statement's layout doesn't match one already confirmed as a genuine statement. Weak on its own: many genuine statements come from layouts DocuClipper hasn't confirmed yet.
Photos, screenshots and CSV or Excel exports of a statement get the reconciliation check only; the PDF checks don't apply to them, and the tab says so (an Image Upload or CSV or Spreadsheet Upload note that doesn't lower the score) instead of reporting the file as authentic.
The Fraud Detection tab shows the PDF viewer beside the analysis: pick the document at the top, read its authenticity score as a gauge with the plain-language interpretation below it, then the table of signals found - each with its type, confidence and details. A Text column shows the figure a signal points at (for example a pasted-in amount) and appears only when a signal has one. You can also Download Project Fraud Report as a CSV for the whole project; its scoreStatus column says whether each score is current, being updated (rescoring), being computed for the first time (scoring) or not scored yet.
Best practices for review
- Put the whole packet in one project before you review, so each statement is compared with the others.
- Read the reasons, not just the score. Several confident reasons on one statement matter more than one weak reason.
- Consider context. Scans, print-to-PDF and merging tools cause benign findings.
- When a statement is flagged, ask for that month again, downloaded straight from online banking, or verify it with the bank.
Notes
- The presence of any single signal is not proof of fraud. Use multiple signals plus context to make decisions.
- If you connect DocuClipper to ChatGPT or Claude, you can ask the assistant whether a statement looks edited. It gets the same score, risk band and signals shown in this tab.
FAQs
What does DocuClipper's authenticity score mean?
It is a 0 to 100 summary of the reasons found on a statement, weighted by how confident each one is. 90 to 100 looks authentic, 70 to 89 has minor concerns, 50 to 69 is moderate risk, and 0 to 49 is high risk. The score is a guide, not a final decision.
What does DocuClipper check on a bank statement?
How the PDF was made (edits saved after creation, conflicting metadata, editing software, pasted-in amounts, an unfamiliar layout), whether the statement agrees with itself (reconciliation, and printed totals against printed balances), and how it fits the other documents in the project (the same account's software and balances, and pay stubs against deposits).
Why should I upload all of an applicant's statements to one project?
Because some edits only show against the rest of the packet. A statement can look perfect on its own and still be made with different software than the same account's other months, or open at a balance the previous month never closed at. DocuClipper compares every document in a project with the others and lists what it finds among that document's reasons.
Can a legitimate statement get a low score?
Yes. Scans, print-to-PDF workflows, online merge or compress tools and corporate PDF tooling can produce benign findings. Review the reasons and compare with another statement of the same account before flagging the document.
Is one fraud signal enough to reject a statement?
No. A single signal is not proof of tampering. Prioritize statements where several confident reasons stack up, and consider context before making a decision.
Does DocuClipper check invoices and pay stubs too?
Yes. Invoices, receipts and pay stubs get the PDF checks plus checks of their own (invoice totals; pay stub gross, deductions, net, taxes and year to date; and a pay stub's net pay against the deposits in the project). See Invoice and Pay Stub Fraud Detection.
Which plans include bank statement fraud detection?
Business and Enterprise, plus every plan during the free trial. Starter plans can use the Fraud Detection tab for 14 days from the first time a statement is opened in it; after that, upgrade to Business to keep using it.