Procurement as a Lever

The public sector is one of the largest buyers of software in Austria and Germany. Every tender determines which systems will run in agencies, registers, and specialized applications for the next ten years, and it tells the market what the state considers a trustworthy system. To close our series on the self-driving state, this post is therefore about the lever that is most often overlooked in the debate on artificial intelligence in public administration: public procurement.

Why the Commission counts on procurement

In our book "Der selbstfahrende Staat" (Springer Gabler 2024, in German) we traced the European perspective on artificial intelligence in public administration. One finding is central to this post: from the European Commission's point of view, the public awarding of contracts is a key factor in the adoption of AI by the public sector. It is meant to stimulate demand and supply for trustworthy and secure AI technologies in Europe. With its "Adopt AI Program," the Commission wants to support the awarding of contracts for AI systems and redesign the procurement processes themselves. The idea behind it: the public sector should use its collective purchasing power as a catalyst and thereby also play a pioneering role for companies in the EU in developing sustainable and human-centered applications.

Specifically, administrations, explicitly including cities and municipalities, are to be supported in procuring trustworthy AI by developing minimum qualifications for algorithms that can be used in contract terms. These include interfaces for disclosing autonomous decision-making, which citizens can use to inspect decisions, and registers of the algorithms in use. Germany has created a dedicated body, the Competence Center for Innovative Procurement, to develop public tendering processes toward AI-supported procurement; in Austria, the IÖB service center for innovation-promoting public procurement plays a comparable role.

The logic is simple: whoever bundles demand sets standards. A state that requires explainability, open interfaces, and exit rights in its tenders gets a market that delivers exactly that. A state that evaluates on price and feature lists gets closed systems with long maintenance contracts.

Every tender is a statement about which systems the state considers trustworthy.

Four requirements that belong in every tender specification

The posts in this series have described the building blocks of the self-driving state one by one. In procurement they come together, not as a vision but as text in the tender specification and as a weighted award criterion.

1. Explainability. Every automated decision must be traceable to rule, data, and parameters. In the book we describe complete logging and a full history of all decisions for this purpose, so that oversight bodies from the legislature, judiciary, and executive can understand at any time what was decided and why. For procurement this means: decision logs, an explanation interface, and access for oversight bodies are mandatory requirements, not options. Why this is a principle of the rule of law and not merely a technical property is set out in our post Explainability as a principle of the rule of law.
2. Register access and open interfaces. The self-driving state offers its services on its own platforms and additionally provides them as interfaces so that commercial platforms and providers can integrate them. A specialized application that draws data from registers instead of requesting it again presupposes exactly such interfaces. In the tender specification this means documented standard interfaces, no proprietary data formats, and connection to the existing registers, as described in Registers instead of certificates.
3. Separation of rules and application. The digital legal twin is the formal representation of a law: decision rules, parameters, and data structures in machine-readable form, published together with the prose text, under a free open-source license, and audited by the judiciary before publication. It is then built into specialized applications, online platforms, and commercial software such as ERP systems by the executive, other state bodies, and companies, where it serves as the control unit for software decisions. For procurement this yields a clear division of roles: the rules belong to the state, the vendor delivers the application around them. A change in the law then becomes an update of the twin, not a change request to the manufacturer. And because the logic is defined by the law, no dependency on an individual vendor arises.
4. Sovereignty and exit. Data held in Europe, a documented data model, access to the source code or an escrow arrangement, and a contractually regulated return of all data and configurations when changing vendors. What a state must not do without here is set out in the previous post Digital sovereignty. In the tender, exit is a criterion weighted at award, not a clause in the contract that is first read by whoever succeeds the person responsible.

None of these four requirements is new. What is new is that together they decide whether a system can follow the stages from the digital to the automated to the self-driving state at all.

The tendering procedure itself becomes self-driving

Procurement is not only a lever; it is itself a procedure that passes through the stages of the self-driving state. In the book we describe the public tender as one of the business situations that change from the analog through the digital to the self-driving administration.

The decisive sentence is in the last row: only if all criteria are transparently met and no plausibility check fails can the award decision be made automatically. In all other cases it is prepared automatically and remains with the responsible persons. For complex IT procurements with qualitative award criteria this will be the rule for a long time, and rightly so.

The benefit lies less in the decision than in the procedure. A high degree of standardization ensures that all bidding companies receive the same information and that the selection process is fair and transparent. Time and staff effort decrease, the procedure becomes accessible to a wider range of potential bidders, which increases competition and leads to more qualified bids. Fully documented data makes it easier to review every decision, compliance reports can be generated autonomously, and the risk of errors, undue influence, and corruption decreases because predefined rules and criteria apply equally to everyone. We consider this point in particular to be underestimated: a traceable award procedure protects not only the bidding companies but also the case officers who are accountable for it.

What the market needs from the state

A market for trustworthy systems does not emerge from appeals but from reliable requirements. There are three things the state can deliver.

Predictable criteria. Vendors invest in explainability, logging, and open interfaces when they know that these properties will be evaluated in the next tender, in the next municipality just as in the next ministry. The minimum qualifications for algorithms suggested by the Commission are the right approach, as are the model contractual clauses for the procurement of AI systems that the Commission makes available to public buyers. And the AI Act, Regulation (EU) 2024/1689, in force since August 2024 with its first provisions applicable since February 2, 2025, specifies what is required of high-risk systems in terms of documentation, logging, and human oversight. Whoever writes these requirements into tender specifications today procures systems that will not have to meet the regulation's further obligations retroactively.

Open rule sets. A digital legal twin published under a free license may be used, developed, and exploited by anyone, without license fees. That opens the market: from the large ERP vendor to the specialized provider of a municipal application, everyone can build in the same certified rule set and compete on quality, usability, and operations rather than on exclusive knowledge of an area of law. The state thereby specifies the part that is sovereign and leaves to competition the part that is not.

Staged demand. Not every tender has to be a major project. Pilot projects in municipalities and cities, as we proposed in Municipalities first, create references with manageable volumes that later awards at state and federal level can build on. That, too, is shaping the market: small and medium-sized vendors get a chance before the standard is set.

What we see in procurement projects

From our work in the public sector, a few observations can be drawn; they are experience from ReqPOOL projects, not statistics.

• The award is decided before the award. Most problems arise not in the procedure but in the specification beforehand: unclear requirements, criteria copied from the last tender, no idea which decisions the system is later supposed to make on its own. A tender specification that does not contain the four requirements above will not get them into the contract either.
• "Should" criteria do not work. Explainability, interfaces, and exit must be weighted award criteria. Formulated as "should," they are displaced by the cheapest bid, and the administration buys a system for ten years that will not follow the next stage.
• Clarify ownership. Rules, data model, configuration, and documentation belong to the contracting authority. This applies to custom developments as much as to standard software, where at least the configuration and the data must be exportable.
• Evaluate independently. We are vendor-independent and do not sell licenses. In procurement, this independence is not a footnote but the precondition for an evaluation of a bid that is not colored by an interest in a product.

The self-driving state as we describe it in the book makes around 80 percent of its important decisions in real time with software systems. No single major project leads there, but a long series of procurements that all point in the same direction. That is why procurement is the lever.

The next step

This post concludes our twelve-part series on the self-driving state, which began in March 2024 with the question of why public administration is the ideal candidate. The answer runs through registers, explainability, the civil service, municipalities, the limits of autonomy, resilience, and sovereignty all the way to procurement: the state holds every lever itself.

You will find the complete model on the page about the book Der selbstfahrende Staat. How we accompany administrations from specification through tendering to rollout is described on the page Public administration. If you are preparing a procurement that is meant to be more than a replacement of the old system, we would be glad to discuss how the four requirements find their way into your tender specification.