Check an invoice in seconds. Invoice verification: totals and VAT recomputed, duplicates spotted, invoice fraud score from 0 to 100.
Create a free account to run the analysis.
Create a free accountFor document volumes or an API, talk to an expert.
The invoice checker used by insurers, banks and large companies
4.8 / 5
A fake invoice can be an edited PDF, an invoice made up from scratch, or a genuine invoice with new bank details. The fake invoice detector runs four levels of checks: in the file, in the image and outside the document.
Our proprietary AI model, trained on fakes, analyses the invoice image. It flags an invoice generated from scratch as well as an amount, a date or bank details retouched on part of the page, on scans and photos too.
AI detected82/100An edited PDF often keeps its past. On one of our own invoices edited by an AI assistant, the screen shows 50,900.00 while the stored text says 50,400.00. On another, the original number and date were still sleeping in the file, recovered without the original.
The ABA routing number is validated, the vendor's name and EIN are compared with the W-9 on file, and the company is looked up in its state business register. New remit to details or an unknown company never go through in silence.
Every line is extracted and recomputed: quantity times unit price, subtotal, sales tax, freight and total due. The invoice is also compared with those already received, on content rather than invoice number, to stop duplicate payments before the payment run.
Without a tool, these three checks stop clumsy fakes in ten minutes. A tip: ask vendors for native PDFs, or invoices through your AP portal, rather than scans or photos.
Multiply each quantity by its unit price, add the lines to get the subtotal, then check the sales tax: rates vary by state and locality, and many B2B purchases are exempt. Then run a three way match: the invoice must agree with the purchase order and the receiving report.
The name and EIN must match the W-9 on file, and the company must exist in its state register. The invoice number should follow the vendor's usual sequence. Compare the bank account with your vendor master file: any change is confirmed by calling a number you already had.
Use our free PDF metadata viewer and look at Creator, Producer and the timestamps. A native invoice comes out of an ERP or billing software such as QuickBooks, NetSuite or SAP. Word, Canva or a PDF editor are red flags, and an edit after the issue date deserves a question.
Koncile's fraud detection API gives you a score (0-100) and detailed explanations of manipulation risks for every invoice, right inside your AP tools.
Lines, sales tax and totals recomputed.
High, medium or low level, with a one sentence explanation.
High risk from 80.
A one line summary ready for your ERP before the payment run.
curl -X POST https://api.koncile.ai/v1/upload_file \ -H "Authorization: Bearer $API_KEY" \ -F "files=@invoice.pdf"requests.post("https://api.koncile.ai/v1/upload_file/", headers={"Authorization": f"Bearer {API_KEY}"}, files=[("files", open("invoice.pdf", "rb"))]){ "status": "DONE", "General_fields": { "Vendor_name": {"value": "Harbor Lane Supply LLC", "confidence_score": 0.99}, "Total_due": {"value": "50900.00", "confidence_score": 0.98} }, "Line_fields": { "Line_total": [{"value": "48500.00"}, {"value": "2400.00"}] }, "Fraud_detection": { "risk_score": 74, "summary": "Displayed total differs from stored text: 50900.00 vs 50400.00", "signals": [ {"name": "Character substitution", "severity": "high"}, {"name": "File rewritten after issue", "severity": "medium"} ] }}Wherever an invoice triggers a payment or a reimbursement, a fake one is costly. Here is where the checker fits in.
Every incoming invoice is extracted (lines, tax, bank details) then checked before the payment run, by API or through a dedicated email address. Ecotable runs tens of thousands of invoices a year this way.
Invoice matched with the purchase order, the receiving report and the contract, line by line. In our billing audits, we generally recover 1 to 3% of the amounts invoiced, sometimes 4%.
Receipts and expense invoices checked in batches: inconsistent dates, wrong tax, receipt numbers with no continuity, diverging point of sale software versions, AI generated documents.
Repair shop, contractor or replacement invoices attached to a claim: an inflated amount, a recycled estimate, a company nobody can find. Anomalies come out before payout, with a score the adjuster can explain.
Invoices for care, equipment or services submitted for reimbursement: an edited amount, an unknown provider, the same invoice sent twice. The check runs on every upload.
An invoice sold or pledged for financing must exist only once and come from a real customer. The checker reviews the file, the debtor and duplicates before funds are released.
A carefully edited invoice shows no arithmetic error and no misalignment to the eye. The fake invoice detector relies on invisible signals: how the file was written, what it keeps of its earlier versions, and what outside records say about the vendor.
| Invoice element | What Koncile checks | What gives a fraud away |
|---|---|---|
| Issuing software | Creator and Producer compared with the issuer's known invoices | Word, Canva, a PDF editor or a Python script where the vendor uses its ERP |
| File writing style | Object order, compression, cross reference table: the library that wrote the PDF | An invoice born in Chrome rewritten by MuPDF, qpdf or pypdf, whatever Producer it shows |
| Displayed and stored text | Each amount compared with its character map | 50,900.00 on screen, 50,400.00 in the file |
| Versions and dormant objects | Versions appended at the end of the file, old page content | The original number and date found under the new ones |
| Fonts | Embedded fonts, subsets and encoding, compared with the vendor's template | A font added next to the template's own to retype an amount |
| Image | On a scan or photo: cloned areas, white patches, text laid over, AI generation | Bank details or a total pasted over the original image |
| Arithmetic | Lines, subtotal, sales tax, freight, credits and total due | A correct total on wrong lines, a tax rate that does not match the amount |
| Invoice number and PO | Number sequence, invoice and due dates, purchase order reference | A number outside the vendor's sequence, a PO that does not exist |
| Vendor and bank details | Name and EIN against the W-9 on file, routing number, account against your vendor master file | A name that does not match the EIN, a new account, a bank in another state or country |
| Duplicates | Vendor, amounts, lines and period compared with invoices already received | The same service under another number, or sent again as a scan |
An invoice holds your vendors' bank details, your purchase prices and your volumes. Founded by a former lawyer, Koncile was built with data protection at the core of its architecture: third party SOC 2 audit, GDPR compliance, secure hosting. Invoices are encrypted, deleted after analysis and never used to train a model.
It is a tool that combines several technologies to spot the anomalies that give away an edit or a fraud: machine learning that analyses the image, and analysis of the PDF file, which is written like code. The goal is to stop the few percent of fake invoices that slip into your payment processes.
You never know for certain: you measure a risk. It becomes high when one fraud test goes over its threshold, or when several tests point the same way: arithmetic that does not add up, metadata that does not match the vendor, a file rewritten after it was issued. That body of evidence is what should raise the alarm.
It can help, mainly to spot inconsistencies in context. It lacks three things: models specialised in images edited or generated by AI, knowledge of the traces each editing tool leaves, and a base of each issuer's usual invoices, which brings the mismatches out. A language model is only one part of the machine.
It depends on your volume. The price combines a base subscription and a cost per page or per document processed. The plans are detailed on the pricing page.
A fraudster takes over or imitates a vendor's mailbox and sends a genuine looking invoice with new bank details and a note of urgency: pay today, keep it quiet. Only the account is fake. The FBI counted $3.05 billion of reported business email compromise losses in 2025. The defence: compare bank details with your vendor master file and call back at a known number.
Do not trust the number: an invoice sent again often changes by one digit, a date or a format. Compare the content, vendor, amount, lines and period, with the invoices already paid. Koncile automatically compares every new invoice with those already received, including a scan sent after the PDF was paid.
Yes. An invoice scheme sent by email or wire falls under wire fraud (18 U.S.C. § 1343): up to 20 years in prison, 30 when a financial institution is affected. Sent by mail, it falls under mail fraud (18 U.S.C. § 1341). In the United Kingdom, it falls under the Fraud Act 2006.
A false invoice issued by a real vendor from its usual software has no editing trace: only matching it with the purchase order or contract brings it down. A photo loses the PDF structure, so only the image model judges it. Finally, a high score is a signal to verify, not an accusation.
Call your bank at once and ask for a recall of the payment: the first hours count. Keep the invoice and the emails, then report it to the FBI's IC3 and to the FTC in the United States, or to Action Fraud in the United Kingdom. Check your vendor master file before the next payment run.
Each document gets its own method: a proprietary ML model for photos, known metadata and forensics for native PDFs, and contextual cross checks.
Inflated balances, deleted lines, a PDF rewritten with software other than the bank's.
See the detection
Net pay changed, contributions that do not match gross pay, a layout that does not match the payroll software.
See the detection
Photo replaced, date of birth or number changed, an MRZ that no longer matches the fields.
See the detection
Address or name replaced on an energy or phone bill, date moved forward.
See the detectionUpload your vendor invoices, credit notes or expense receipts: Koncile extracts the data and spots fakes in seconds.
Talk to an expert
