Digital Compliance: Antitrust Screening

Cartels have long since ceased to be forged exclusively at highway service stations. Back in 2017, we highlighted the effects of digitalization on antitrust compliance in an article ("Antitrust law 4.0"). Since then, there have been several cases of "digital collusion" that have been taken up by antitrust authorities worldwide and in some cases also sanctioned (see, for example, the article Dohrn/Reinhold, "Collusion by Code": Challenges for antitrust compliance, in Wirtschaft und Wettbewerb, Issue 10, p. 540 et seq.). This last article in our "Digital Compliance" series is dedicated to digital antitrust compliance in the form of so-called "antitrust screening tools". The article describes how they work and provides an outlook on (future) areas of application in cartel prevention and the repressive processing and prosecution of cartels.

The basics of algorithm-based antitrust compliance

Antitrust screenings serve as a method for the automatic detection of antitrust violations by examining data for competitive irregularities. They are therefore part of "reactive screening" (read our Part I of the series of articles on this distinction). As clear as the objective of cartel screening is, its implementation is just as varied.

A first distinction is made with the question "What is being examined?". One method is so-called structure screening. Algorithms analyze the structure of certain markets and uncover anomalies ("anomaly detection"). These are then examined for factors that are known to promote or maintain agreements that violate antitrust law. The advantage of this type of screening is that aggregated data is already sufficient for the analysis. However, this in turn reduces the reliability of the output. The behavioral screening method, on the other hand, relies on individual data from market participants and therefore assesses the behavior of individual companies on the market. However, the advantages in the accuracy of the result are offset by the need for complex statistical tests.

A distinction can also be made when it comes to the question of "How are investigations carried out?". Classic algorithms are currently still the simplest and most common method. They are characterized by the fact that the programmer sets up fixed rules in advance and thus figuratively implements a network of filters. In the application, the selected data is run through these filters, making it possible to assess whether or not signs of cartel agreements can be found.

The use of artificial intelligence is more complex. The programmer trains a program that independently recognizes, applies and continuously improves itself (so-called "machine learning"). In order for AI to recognize, apply and further develop certain patterns, it must be "fed" with sufficient information. Data that reliably shows when a certain market behavior is indicative of anti-competitive behavior and when it is not is therefore the learning material for such programs and generates their ground truth.

How cartel screening programs work

A prominent example of the use of digital cartel screening tools in practice is the cartel screening program developed by Deutsche Bahn AG to detect bidding cartels. Algorithms are used to screen behavior on the bidding market under investigation.

The procedure for such screening programs can usually be divided into several phases (for details, see Gillhuber/Kauermann/Hauner (eds.), Künstliche Intelligenz und Data Science in Theorie und Praxis, Springer Spektrum, 2023, p. 267 ff.). In a first phase, the necessary data is processed, selected and automatically fed into the next phase according to a specific scheme (known as the "data pipeline"). As only previously categorized data types are required for the assessment (such as data on the bidder or the amount of the bid), the analysis process is made more effective. However, this step is also used for automation and is therefore particularly attractive for companies for compliance purposes. This allows a constant review to be carried out.

The actual analysis takes place in the second phase. It is possible to look at individual bidding procedures in isolation as well as to take a cross-award approach. For this purpose, the data fed in is processed into statistics or key figures according to previously defined rules and examined for cartel-related anomalies. This analysis is based on the fundamental assumption that cartel-related market defects create certain patterns that can be identified and represented by programs. An essential prerequisite is therefore to define these patterns abstractly beforehand. This can be done, for example, by defining benchmarks for certain award procedures that reflect the limits of "healthy" bidding competition. If the data deviates from these benchmarks, the program sounds the alarm. The more patterns are detected by the program, the more likely it is that the program has discovered a cartel.

For the third phase, the individual analyses are consolidated and presented in a front end. The aim is to visualize the results and the assessment steps in such a way that the program user can independently identify antitrust anomalies and thus verify the assessment of the program. Antitrust screenings are not intended to replace the work of compliance departments or antitrust lawyers. Rather, they are intended to supplement and simplify the work of all parties involved.

Data analysis becomes part of everyday life for antitrust authorities

Antitrust authorities are now increasingly relying not only on traditional leniency applications and whistleblowers, but are also working on their own programs to identify (bidding) cartels. Most antitrust authorities have now started to collect large data sets and systematically analyze them. This includes, for example, a project by the British antitrust authority CMA. A corresponding tool from the Swiss antitrust authority COMCO has already achieved concrete successes (e. g. in uncovering the road surfacing cartel in Ticino). The South Korean antitrust authority has initiated several antitrust proceedings based on its Bid-Rigging Indicator Analysis System tool (BRIAS (p. 62)) and concluded each case with a decision imposing a fine. For its screening tool, the Greek antitrust authority uses (daily updated) price data from the major supermarket chains for more than 2,000 different products. The German Federal Cartel Office now also uses screening to systematically examine markets for anomalies.

Specialized knowledge and a suitable IT infrastructure are required for the development and use of cartel screening programs. Accordingly, more and more antitrust authorities are hiring data scientists and setting up special departments dedicated to the use and development of new analytical tools. For example, the UK competition authority has implemented a new Data, Technology and Analytics (DaTA) unit. The Dutch authority has appointed an AI-experienced "Chief Data Officer" and placed a team of 15 to 20 data engineers and data scientists at his side ("Taskforce on Data and Algorithms"). The EU Commission (p. 48) now also has a "Data Analysis and Technology" unit and has created the position of Chief Technology Officer.

Tools can offer companies high added value

To date, cartel screening has only been used very sporadically in corporate practice. In the absence of regulatory requirements for cartel screening, the introduction of such systems is currently still primarily a commercial decision. It certainly does not help that mature software solutions - as far as can be seen - are still in short supply on the market and companies are therefore still primarily reliant on their own resources.

However, companies will have to upgrade their compliance work in this area sooner or later in order to keep pace with the digital cartel enforcement of the antitrust authorities. Companies can benefit from cartel screening tools, e.g. because they can identify cartels more quickly as potential cartel victims.Antitrustscreening tools can also be of great added value for a company's own repressive compliance work, as they increase the chance of early internal detection of committed or imminent infringements. If an infringement has already occurred, this also increases the chance of being the first to report an antitrust infringement to the relevant antitrust authorities and to benefit from the leniency programs. But even if the "rat race" for the first leniency position is not won, implemented screening tools can be taken into account as part of an effective compliance management system to reduce fines (see e.g., Section 81d (1) no. 4 ARC for Germany; see also our article on the requirements for an "appropriate and effective" compliance system). Last but not least, screening tools can support companies in allocating their resources by prioritizing internal antitrust audits to particularly serious (suspected) infringements.

*(see, for example, the article Dohrn/Reinhold, "Collusion by Code": Challenges for antitrust compliance, in Wirtschaft und Wettbewerb, Issue 10, p. 540 et seq.)

Back to list