Greece’s Independent Authority for Public Revenue (AADE) has used a new risk-analysis system based on the Jaccard similarity index to identify a network of shell companies and suspected front persons in Thessaloniki.
The system applies an advanced algorithm to detect connections between businesses that may be created and shut down in succession to avoid paying taxes and social security contributions. Rather than examining each company separately, it maps relationships between different tax identification numbers (AFMs), looking for recurring patterns that could point to coordinated activity.
According to AADE sources, the tool helped authorities uncover a large network involving 380 businesses and 205 individuals who had accumulated €43 million in debts to the state.
How the Jaccard index identifies suspicious links
The Jaccard index measures how much two sets of data overlap. In this case, the system treats each business as a collection of information and compares it with others. A higher similarity score, closer to 1, indicates a greater degree of overlap and a potentially stronger connection between companies.
The algorithm examines several categories of information:
- People involved in businesses: It checks whether the same partners, shareholders, managers or legal representatives appear across different corporate structures.
- Shared infrastructure and contact details: It identifies common business addresses, telephone numbers, email addresses, accountants and bank accounts.
- Commercial transactions: It compares customers and suppliers using data from myDATA, Greece’s electronic bookkeeping and tax reporting platform, as well as VAT declarations. If a newly established company serves the same customers as a business that has just closed while leaving debts unpaid, the similarity score can rise sharply.
- Connections across the network: It maps the relationships between companies and individuals to create a broader picture of the suspected network.
- The index works by comparing the number of shared companies with the total number of distinct companies associated with two individuals.
For example, if one person appears in 15 companies and another appears in 15 different companies, with 10 businesses shared between them, the total number of distinct companies is 20. Dividing the 10 shared companies by the 20 distinct ones produces a similarity score of 50%.
Pairs of individuals with high similarity scores and shared business connections are treated as links. These links can then be combined into chains and groups of related companies, which are examined for indications of illegal activity.
What the Thessaloniki investigation revealed
In the Thessaloniki case, AADE’s Operational Center for the Detection and Control of Economic Crimes (DEOS) identified hundreds of links between tax identification numbers. Each link represented a pair of tax IDs connected through at least two common companies.
When the links were combined, they formed chains and groups of business relationships. The results varied depending on the similarity threshold applied:
- Above 5%: 719 links connecting 369 tax IDs across 29 groups. The largest group included 262 tax IDs.
- Above 10%: 504 links connecting 330 tax IDs across 60 groups. The largest group included 34 tax IDs.
- Above 20%: 358 links connecting 270 tax IDs across 77 groups. The largest group included 13 tax IDs.
The way these connections are formed is important. If individual A is linked to B, and B is linked to C, the system creates a chain even if A and C have no direct connection.
As the similarity threshold rises, only stronger links remain under examination. Larger chains are consequently divided into smaller groups, allowing investigators to focus on the specific companies and relationships connecting the individuals involved.
AADE stresses that the groups identified by the algorithm are not automatically considered proven criminal networks. Instead, they represent patterns and indications that require further investigation. The system makes it significantly easier for authorities to identify these connections and direct their checks accordingly.
How data analysis is changing tax inspections
The integration of the Jaccard index into Greece’s tax enforcement system has changed how authorities identify and prioritize suspicious cases.
Algorithms can automatically flag high-risk tax identification numbers and select cases for targeted inspections, directing auditors toward the core of potentially fraudulent networks.
The system also combines information from multiple databases, including the General Commercial Registry (GEMI), banks, land registry records and electronic payments. This allows authorities to build a unified digital risk profile for businesses and individuals.
AADE has also released a video illustrating how the investigation in Thessaloniki developed. The authority says the methodology is applied in similar cases, with the red-outlined boxes in the final stage of the visualization representing cases classified as high risk.
Source: tovima.com






































