Digitization in the new Public Procurement Code: New technological frontiers and artificial intelligence
Thanks to Denise Moretti for collaborating on this article
The digitization of tendering procedures is one of the most important issues addressed by the new Public Procurement Code, which introduces systematic reorganization and significant changes. Indeed, an entire part of the code (Book I, Part II) is devoted to the subject.
As stated in the introductory report, digitization not only simplifies tendering procedures and makes planning, procurement, and contract management activities more effective, but also constitutes an effective corruption prevention measure, as it allows transparency, traceability, and control of all activities in a sector that is widely exposed to corruption risk.
On the other hand, digitization of the public sector is one of the main objectives of the NRRP, mission 1 .With reference to tendering procedures, the mission envisages “defining the modalities for digitizing the procedures for all public contracts and concessions and defining the requirements for interoperability and interconnectivity,” as well as the implementation of a national e-procurement system interoperable with the management systems of public administrations.
The stated objective is the complete digitization of the entire “life cycle of public contracts,” understood as all activities from planning to the establishment of requirements to the complete execution of the contract. This is an onerous task for administrations in terms of both time and resources. This provision has a disruptive scope in that one of its program objectives is to perform each phase in the public supply chain digitally.
According to the letter of the code, this ambitious goal should be achieved as early as January 1, 2024, when the digitization provisions take full effect. However, this is unlikely to happen. In fact, digitization of procedures under the new code is one of numerous actions envisaged by the PNNR for which the government has asked the European Commission for “review and remodeling” of targets; it says that for this a “paradigm shift in the architecture of the system for digitization purposes” is needed. Postponement of implementation of these standards is therefore expected.
That said, this contribution aims to analyze some of the most significant provisions on the subject of digitization based on their potential to change the current context, with a focus on those related to the national digital procurement ecosystem (known as e-procurement) and automation of the life cycle of public contracts. In addition, in light of the new code’s endorsement of the use of artificial intelligence (AI) systems for management of tenders, we will focus on possible implementation of predictive algorithms to determine supply requirements and identify tender base prices, particularly as regards the health sector.
The national digital procurement ecosystem
The national digital procurement ecosystem, or e-procurement, consists of digital infrastructure platforms and services that enable the life-cycle management of public contracts (primarily ANAC’s National Data Bank) and digital procurement platforms used by contracting authorities. Regulation of e-procurement is innovative. With an eye to complete digitization of procedures, it envisions the use of interoperable telematic platforms as a standard means of managing the contract life cycle, with the aim of creating a single portal for contracts. In fact, as is widely known, to date not all contracting authorities use telematic platforms for issuing and managing tenders. In any case, various platforms that do not communicate with each other are often used. The new code stipulates that digital procurement platforms must be used for all communications or publications pursuant to the code (and for anything not provided by the platforms, through the use of the digital domicile).
The National Public Contracts Database set up at ANAC has been given a central role—already covered under the previous procurement code and the Digital Administration Code—and will become the technological backbone infrastructure of the national e-procurement ecosystem. It will be divided into six sections, some of which are already operational, while others have yet to be implemented. This database has access to (and is fed by) the information contained in other databases managed by administrations (such as the registry of natural and legal persons and the business registry), public companies, and public service concessionaires that hold data necessary for the digital life cycle of contracts. It also interfaces with the digital procurement platforms used by individual contracting stations and central purchasing bodies, making available to all interested parties (the contracting stations and economic operators above all) the data and information needed to manage procedures. One example is individual contracting stations checking that economic operators fulfill the general and specific requirements for participation in tenders.
Once implemented, the e-procurement system and the related platforms and services will allow economic operators and contracting stations to draft and acquire documents in digital format, publish and transmit data to the National Public Contracts Database, access all tender documents, submit bids and the Single European Tender Document, and carry out technical, accounting, and administrative oversight of contracts during execution.
With a view to facilitating the digital conduct of procedures, Article 25 of the code also provides that a contracting station that does not have a platform may make use of those made available by other contracting stations or other awarding bodies.
The use of automated procedures in the life cycle of public contracts
A rule that deserves special attention for its innovative scope is the new Article 30 of the code concerning the “use of automated procedures in the life cycle of public contracts.” Automated decision-making can be understood as all decisions concerning issues arising during the procurement process that are made by an algorithm according to specific predetermined criteria.
Through this new article, the code seems to support implementation of such technologies—starting in the tender evaluation phase—by providing their use “where possible” in order to improve the efficiency and effectiveness of tendering procedures and expressly mentioning technological solutions such as artificial intelligence. In any case, this is a provision aimed at regulating the (near?) future, as indicated by the introductory report to the code, insofar as “at present, in the context of tendering procedures, mostly non-learning algorithms are used for the automatic comparison of certain parameters characterizing the bids that are knowable.”
However, it is possible that in the near future the availability of large quantities of data may allow the training of learning algorithms to be applied to more complex tendering procedures: hence the usefulness of including regulations that expressly list the principles intended to govern the use of such systems. In fact, Article 30 sets forth certain technical standards that should guarantee transparency and security when implemented. Specifically, it stipulates that contracting stations must make available the source code and any documents necessary to understand the logic underlying operation of an automated system and must introduce tender clauses that ensure that assistance and maintenance activities necessary to correct any errors arising from automation are conducted. Furthermore, automated decisions must be knowable and understandable, and there must always be human input capable of “checking, validating, or disproving” such decisions.
Automated procedures and artificial intelligence: How may they be used?
Article 30 of the new code opens up numerous opportunities for administrations in the area of digitization. It also gives the green light to development of automated systems that use artificial intelligence, which can potentially be applied to all tendering and pre- and post-award activities.
Artificial intelligence systems primarily will assist operators in planning tenders and evaluating offers. There is a wealth of research on the use of these technologies in public procurement. Among those most useful for increasing efficiency of tender management are predictive algorithms that exploit AI systems, in particular machine learning (ML) technologies. Predictive algorithms work by means of ML functions that detect patterns in historical supply chain data, i.e., the critical points and crucial factors for determining the correct quantity of supplies, such as unpredictable risks, logistics, and optimization of procedures. Using ML systems, a machine learns from the data it relies upon and conducts analyses without a defined structure or specific instructions. Systems of this type have the potential to be very efficient when planning quantities or auction bases for supply tenders.
One sector that is poised to benefit greatly is the healthcare sector. These technologies may make it easier to pinpoint the quantities of drugs/medical devices to be purchased and the relevant auction base. In fact, when the demand for pharmaceutical and medical products is underestimated, healthcare facilities often must resort to purchasing products outside the tender (including through direct procurement) at a higher cost. The same problem often arises in the case of unsuccessful tenders—a phenomenon that is not infrequent in this sector—when contracting authorities aim for economies of scale with prices that are so low that suppliers refrain from participating because at the price set by the administration they would not turn a profit if awarded the resulting contract.
Algorithms for forecasting requirements and contract prices
A forecasting algorithm was the subject of the STEINBOCC Project (TENDER NEEDS: IMPACT variaBles and prediCtive data analytiCs), conducted by researchers at the Healthcare Datascience LAB at LIUC-Università Cattaneo, in collaboration with EGUALIA (generic, biosimilar, and value-added medicines industries). This research developed a forecasting algorithm, accessible via a web interface, that can be adapted for different active ingredients and used free of charge by regional contracting authorities and pharmaceutical companies to identify future needs or size purchases for future tenders.
Another predictive algorithm that has been the subject of recent research and is potentially useful in the health sector is the “piece price estimator,” designed to identify the optimal price for the auction base—the threshold below which the probability of unsuccessful tenders increases. After conducting research on the piece price estimator, the University of Oviedo proposed an ML mechanism for the identification of the auction base for tenders for the supply of medicines. Estimating the price actually paid by an administration is particularly complicated due to unpredictability of the market, as well as the potential for changes to the contract during execution that directly or indirectly affect the cost of supply.
What are the prospects?
The inclusion in the code of a section devoted to the digitization of procedures and the objective of digitizing the entire life cycle of contracts is undoubtedly a favorable development, regardless of the “technical time” required to implement these measures in reality. Similarly, expressly mentioning the possible use of automated decision-making systems, including those that exploit AI, opens the way for the design and development of new technologies specifically designed for public procurement, though these are prospects for the near future.
Indeed, while the code mentions the possible “use of automated procedures” for decision-making purposes as well, it does not identify what type of procedures, nor does it provide any specific indications or examples in this regard. It is limited to indicating the precautions to be taken from a technical point of view to ensure transparency and security when using new technologies. On the other hand, the use of true machine learning algorithms—instead of mere systems for automatic comparison of data and parameters (e.g., algorithms for the automatic selection of the best offer in tenders awarded according to the lowest price criterion)—raises numerous questions both about how these algorithms may operate in the context of the administrative discretion that governs and directs many choices made by entities and about possible responsibility deriving from decisions made via automated and AI systems, particularly when their use has a direct and significant impact on the economic operators that participate in tenders.
A first possible, albeit incomplete, answer to this question may be found in Article 30, where human intervention is deemed necessary to a decision-making process using an algorithm, for the purpose of overseeing and validating the decisions made by the automated system. This then leads to the concept that this human contribution can guarantee the permanence of discretion in the choice of the administration, if necessary for the specific activity involved, as well as the possible identification of a “person responsible” for the decisions made with the assistance of these new technologies. In any case, the issue is clearly very broad and complex, and the related problems that arise will differ depending on the type of algorithm and how it is used in each tender procedure.
 Legislative Decree 36/2023.
 The Italian “National Recovery and Resilience Plan.”
 Article 19 of the Code.
 “Proposals for the revision of the NRP and RepowerEU chapter” available at the following link: https://www.agenziacoesione.gov.it/download/le-proposte-del-governo-per-la-revisione-del-pnrr-e-il-capitolo-repowereu/ (target M1C1-75, T4 2023).
 Autorità Nazionale Anticorruzione, or the Italian National Anti-corruption Authority.
 Article 213.
 Article 62-bis.
 Pursuant to ANAC Resolution No. 261 of June 20, 2023, the BDNCP is divided into the following sections: (a) Anagrafe unica delle stazioni appaltanti (AUSA); (b) Piattaforma contratti pubblici (PCP); (c) Piattaforma per la pubblicità legale degli atti; (d) Fascicolo virtuale dell’operatore economico (FVOE); (e) Casellario informatico; (f) Anagrafe degli operatori economici.
 Article 22.
 García Rodríguez MJ et al., “Bidders Recommender for Public Procurement Auctions Using Machine Learning: Data Analysis, Algorithm, and Case Study with Tenders from Spain” (link); García Rodríguez MJ et al., “Collusion Detection in Public Procurement Auctions with Machine Learning Algorithms” (link); Torres-Berru Y and López Batista VF, “Data Mining to Identify Anomalies in Public Procurement Rating Parameters” (link) .
 On these topics, consult “The Evolution of the Drug and Device Purchasing System” (link), a document developed by the Italian Society of Hospital Pharmacy and Pharmaceutical Services of Health Authorities (SIFO) with the contribution of companies and associations in the sector; see in particular Chapter 9.