Fraud Detection 2026 – Trends in Theses  
Part 4: What does content-based analysis involve?

The fourth Sunday of Advent is approaching—so it is time for the final instalment of our series on modern fraud detection. This part focuses on content-based analysis, because as the saying goes, the devil is in the detail: consistency, sequential numbering, and logical coherence are the natural enemies of fraud. Content-based analysis is the final step in reliably uncovering manipulation.

What does it include?

1. Date formats should be consistent across different invoices—deviations such as “5.11.25” when a practice would normally write “05.11.2025” may indicate forgery.

2. Invoice numbers must follow a continuous sequence in accordance with German tax law—decreasing or duplicate numbers are suspicious. However, accurately reproducing the correct invoice number, or preventing duplicates, is virtually impossible for fraudsters. ICO.Fraud, on the other hand, detects such inconsistencies.

3. Invoice or purchase dates that fall on Sundays or public holidays? That is unusual—and such inconsistencies can reveal manipulation.

4. It is also important to verify VAT calculations and check whether they match the stated amounts. Especially when figures have been altered after the fact, this is often not the case.  
What else?

5. Verifying check digits for IBANs, EANs, VAT IDs, and social security numbers is another strong element of content-based validation rules.

6. Does the issuing company actually exist? Checking companies in the commercial register is essential, as not only are invoices from real companies forged, but entirely fictitious companies are also invented.

7. And while we are at it: the authenticity of listed websites and the consistency of QR codes should, of course, also be verified…

8. …as should Data Matrix codes and device IDs (IMEI), for example on smartphone invoices, which offer further technical means of verification.  
What does this mean overall?

9. Ultimately, the consistency of relative spacing, fonts, and text alignment within the document must also be analysed, because…

10. …as described in Part 3, layout and formatting errors often reveal fraud only at second glance.

Forgery is wrong — but those who forge documents make mistakes. That means: if we find the errors, we find the fraudsters—and in doing so, we protect the collective community. At the same time, software used to combat insurance and loan application fraud — combining OCR, computer vision, and AI — never makes final decisions on its own. After all, these cases may involve criminal offences. Instead, the software acts as an indicator system, flagging suspicious cases. The final review — and therefore the decision—remains, in line with the EU AI Act, with human investigators in fraud departments.

Image: prompted by ICO-LUX, provided by Perplexity.ai