Our approach to detect document fraud
They try to detect image manipulation, pixel tampering, compression artifacts.
These techniques are useful, but they miss a large category of fraud. Modern document fraud often happens at the semantic level.
Fraudsters don't always alter the visual structure of a document. Instead, they modify the information inside the document. Examples include modifying an invoice total, changing payment details, changing a number in a pay stub, replacing bank account numbers.
These modifications may look visually legitimate. But they introduce logical inconsistencies. They are invisible to visual analysis methods.
Koncile approaches document fraud detection differently.
Each fraud detection model is designed with domain experts who understand the structure and logic of real documents. Instead of relying only on visual signals, Koncile adds a layer of domain-specific, logical and legal controls.
Each detected inconsistency contributes to a fraud risk score between 0 and 1. When the score exceeds a predefined threshold, the document is flagged as potentially fraudulent.
We detect logic problems
Document fraud often reveals itself through inconsistencies in the data. Koncile models detect several categories of anomalies. These are mathematical or structural inconsistencies within the document. Examples include invoice totals that do not match line items, incorrect tax calculations, inconsistent quantities and unit prices, currency mismatches, totals that cannot be recomputed.
These checks immediately reveal many forms of invoice fraud and accounting manipulation.
We spot semantic issues
Some documents contain information that is technically valid but implausible in context. Examples include a supplier address inconsistent with its registration data, a company number that does not match the business name, a tax rate inconsistent with the jurisdiction, a document referencing entities that cannot logically coexist.
These anomalies require knowledge-based reasoning and external data validation.
We uncover regulatory and domain-specific inconsistencies
Some documents may appear internally consistent but still violate regulatory rules or domain-specific constraints. Koncile models incorporate legal and operational knowledge to detect situations where a document does not comply with expected business or regulatory logic. Examples include an invoice missing mandatory legal information (VAT number, invoice date, supplier identity), an invoice that does not comply with required invoicing regulations, a payslip where payroll amounts are inconsistent with labor law calculations, a payslip where mandatory contributions or deductions are missing or incorrectly calculated.
These cases require understanding not only the document itself, but also the legal and operational framework in which the document exists.
Document fraud detection requires more than pixel analysis. Most document fraud detection tools rely only on visual analysis.
.png)
