How AADE’s algorithm unmasked front men in Thessaloniki

Using the Jaccard algorithm, AADE maps hidden links between businesses, identifies front men and shell companies, and unravels the networks behind tax evasion schemes. A network of 380 businesses and 205 people with debts of €43 million is under scrutiny.

How AADE’s algorithm unmasked front men in Thessaloniki

This article is an AI translation of an original piece published in Greek. Read original

The Independent Authority for Public Revenue (AADE) uses the Jaccard similarity index in an advanced model for detecting business links to identify and dismantle networks of fictitious companies (“shell companies”) and front men. Rather than examining each business in isolation, the algorithm maps hidden links between different tax IDs that are opened and closed in succession to avoid paying taxes and social security contributions.

AADE’s risk analysis system treats each business as a “data set” and compares its characteristics with those of other businesses. The higher the Jaccard score (closer to 1), the more likely it is that the companies belong to the same network.

When comparing individuals, the algorithm checks how many partners, shareholders, managers or legal representatives appear repeatedly across different corporate structures. It also cross-checks infrastructure and contact details to identify shared business addresses, phone numbers, email addresses, accountants or bank accounts.

How it works

In its analysis of commercial transactions, the system measures the overlap in customers and suppliers using myDATA data and VAT returns. If a new company serves exactly the same customers as a company that has just shut up shop, leaving debts behind, its similarity score soars.

By mapping networks, the tool led to the dismantling of a large network involving 380 businesses and 205 people that had accumulated debts of €43 million to the state.

The Jaccard index compares only two people at a time: if one appears in 15 companies and the other in 15, with 10 in common, they are associated with 15 + 15 − 10 = 20 distinct companies, since shared companies are counted only once. The formula is: Jaccard = companies in common ÷ total number of distinct companies = 10 ÷ 20 = 0.5, or 50%. In other words, half of the 20 companies are common to both.

The Thessaloniki case

Pairs with high similarity scores and people in common are joined as “links”. These links form “chains”, revealing networks of business relationships and indications of co-operation. We identify the network; the investigation determines whether illegal activity is taking place within it.

In the Thessaloniki case, using Jaccard, AADE’s DEOS identified hundreds of “links” between tax IDs—that is, pairs with at least two companies in common. Joining these links revealed specific chains and groups, depending on the similarity score.

JACCARD INDEX THRESHOLDLinksTotal tax IDsGroupsLargest group
Above 5%71936929262 tax IDs
Above 10%5043306034 tax IDs
Above 20%3582707713 tax IDs

In practice, A is linked to B and B to C, forming a chain, even if A is not directly linked to C. As the threshold rises, only relationships with greater similarity are retained, and large chains break into smaller groups. This helps direct the investigation towards specific links and the companies they have in common, although the groups alone do not prove that networks are operating.

Tax audits

Adding the Jaccard index to the tax administration’s toolkit has transformed how audits are conducted. AADE’s services can now quickly identify suspicious patterns and tax IDs.

At the same time, targeted checks are carried out without human intervention, as algorithms automatically select high-risk cases, directing auditors straight to the heart of the networks.

Finally, databases are linked by combining the index with data from the General Commercial Registry (GEMI), banks, land registries and electronic payments to create a unified digital risk profile.

v
Privacy