MMOOZZGG.com

AI against corruption

AI procurement fraud detection: how AI finds hidden links between bidders

7 min read4 sources

AI procurement fraud detection checks whether different companies are really one person or one group, and whether a supplier is tied to the buyer. AI joins company registers and contracts and shows links that no single tender record shows. In Colombia, the Océano platform found 8,749 companies sharing one phone number this way.

On this page · 6 sections

In short

AI procurement fraud detection brings company registers, owners, contacts and contracts into one table and looks for matches: a shared director, founder, phone, address or email. AI does this for every procedure, not for a sample. In Colombia, the Océano platform found 54,155 cases where one representative stands behind several companies. A match means risk, not a violation.

Which links to look for

Related parties in procurement come in two kinds. In the first, bidders in the same tender are linked to each other, and the competition is only staged. In the second, the supplier is linked to the buyer or to whoever makes the purchase decision.

A tender record does not show such links: it holds the names and tax IDs of bidders, prices and dates. A link appears only when the tax IDs lead to the register of legal entities, and the register leads to people, addresses and phones. So a related-party check always starts by joining several sources.

A link shows through shared elements:

1. The same director, founder or ultimate owner at different companies.

2. The same major shareholder: according to the Open Contracting Partnership, this is a sign that the companies belong to one group.

3. The same phone, address, email or contact person.

4. The bidder’s contacts match the buyer’s contacts.

5. A family tie, or past work of the buyer’s official at the supplier company.

The last kind of link is closer to a conflict of interest, and it needs other data: information on officials and their relatives. This case is covered on the page on conflicts of interest.

Registers that AI joins

One database will not show such links. They become visible only where several sources meet:

1. The procurement register: buyers, bidders, winners, amounts and dates.

2. Bidders’ applications: contacts, signatories, attached documents.

3. The register of legal entities: directors, founders, addresses, history of changes.

4. Data on ultimate owners, if disclosed.

5. Debarment lists of suppliers and sanctions lists.

6. Data on the buyer’s officials, if the audit body has lawful access to it.

The main work of AI here is matching, not searching. The same person is written with typos and in different grammatical forms, an address is given with and without an office number, a phone in different formats. The model brings records to one form and decides where it is the same person or address and where names merely coincide.

OCP separately flags a bidder that is not in the supplier register: such a company may be fictitious or may give false information. According to OCP, a bidder on a debarment or sanctions list also carries a higher risk.

Information about people is personal data. Only someone with a lawful basis processes it, only for the check and with an access log: who opened the record, when and why. This protects both those checked and the inspectors themselves.

Colombia: 54,155 shared representatives

Colombia’s Comptroller General checked links between bidders across the whole public procurement data set. The Océano platform analyzed 8,678,464 public contracts from 2014–2019.

The result of the matching: 54,155 cases where one representative stands behind several companies, and 8,749 companies sharing one phone number. It also found 45 contracts with people who had died.

You cannot get such numbers by hand. An inspector sees one tender and one bidder, while a shared phone among companies from different regions and years shows only when the whole data set is matched at once.

What can be taken from this experience: first match the whole set of contracts by simple fields, phone, address and representative, and only then go through single cases. This way the inspector gets a short list of linked groups instead of thousands of tenders in no order.

Each match on this list remains a signal for a check. One representative at several companies can be lawful, for example a lawyer or a management company that runs several firms. More about the platform is on the page on the Colombian Océano system.

Company networks around one buyer

One link rarely means anything. Risk shows when a network forms around one buyer: several suppliers with shared people and contacts that together win a large share of its contracts.

OCP proposes a separate flag for a supplier’s share of one buyer’s contracts: the supplier and the buyer are flagged if the share is above a threshold, for example 40%. If there are several such suppliers and they are linked to each other, the share of the whole group may be much higher.

It is convenient to view as a graph such a network: companies, people, phones, addresses and contracts in the nodes, with the links between them labeled. A made-up example: Supplier A and Supplier B win contracts from Buyer C; they have different directors but the same phone, and the founder of Supplier B used to run Supplier A. A table does not show this; a graph shows it at once.

What AI flags in such a network: linked bidders in the same tender, a group with a high share of the buyer’s contracts, a link between a supplier and a buyer’s employee and chains through intermediaries two or three steps long. Each flag leads to the register records it is built on.

The graph approach has been tested on large data sets. In the Panama Papers investigation, more than 370 journalists from 76 countries went through 11.5 million files on 214,488 offshore entities, and ICIJ loaded the links between people and companies into the Neo4j graph database and opened them through Linkurious.

For an audit body, the graph becomes a case model: new contracts and register records are added to it, and the inspector sees how the network changes.

Owners and officials: the EU experience

The European Commission looks for links in projects funded by the EU budget through the Arachne+ system. It serves to detect fraud, conflicts of interest and irregularities.

The system has 78 risk indicators; the old version had 107. Arachne+ uses data on company owners, sanctions lists and politically exposed persons, exactly the sources where related parties become visible.

For a Russian audit body, two lessons follow from the EU experience. First: connect data on owners and officials from the start, or the system will see only shared phones. Second: each indicator must explain why it fired, so the inspector can check the conclusion against documents.

The working principle matters more than the number of indicators: the system makes no automatic decisions. It shows the inspector links and risks, and a person draws the conclusions. More about the system is on the page about Arachne in the European Union.

What counts as a risk and what does not

Related parties are often lawful in themselves. OCP states plainly that a shared ultimate owner among bidders in one procedure often breaks no law, but it may point to collusion or to companies that only pretend to compete.

So the system scores not the link but its context. The risk is higher if linked companies bid against each other in the same tender, if the link is hidden through intermediaries or recent register changes, or if a linked group wins a large share of the buyer’s contracts.

The risk is lower if the link is open and disclosed in the application, if the companies do not meet in tenders, or if a shared address is a business center with hundreds of tenants. The system marks such cases with a low score or filters them out by a rule.

What a person does after a flag: looks at which links the system found and from which sources, requests documents on owners and signatories and compares the dates of register changes with the tender dates. A flag means risk, not proof: the conclusion about a violation is made by the inspector, the competition authority or a court.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. El Espectador, 25 October 2019 (elespectador.com)
  2. European Commission, Anti-Fraud Knowledge Centre (checked 16 October 2026), 16 October 2026 (antifraud-knowledge-centre.ec.europa.eu)
  3. ICIJ, 3 April 2016 and 25 April 2016, 3 April 2016 (icij.org)
  4. Open Contracting Partnership, Red flags in public procurement, December 2024 (open-contracting.org)

Found a mistake? Tell us

Publications

For government bodies

We build such systems for government bodies and state companies, from data analysis to a pilot on one scheme. The data stays on the organization's own servers.

Contact us