13th German Procurement Day (DVNW): “Perfect AI Bids” Put to the Reality Test

13th German Procurement Day (DVNW): “Perfect AI Bids” Put to the Reality Test
Dr. Christian Teuber (Partner, Baker Tilly) and Julia Lechtenböhmer (Legal Counsel, NRW.BANK) at the DVNW's annual procurement day (all photos: HCPlambeck).
  • 11/16/2026
  • Reading time 9 Minutes

AI makes bids more persuasive. It does not automatically make the service on offer any better. One question therefore took center stage at the 13th German Procurement Day (Deutscher Vergabetag) hosted by the Deutsches Vergabenetzwerk (DVNW): how can contracting authorities tell whether a bidder can actually deliver what its bid promises?

A walk through the exhibition at the German Procurement Day showed how far things have moved. Numerous vendors showcased AI solutions for public procurement. AI now helps bidders analyze tender documents, develop bid strategies and produce tailored concepts. Contracting authorities, too, receive support with specifications, procedural workflows and bid evaluation.

Julia Lechtenböhmer, in-house counsel (Syndikusrechtsanwältin) at NRW.BANK, a public-law institution, and Dr. Christian Teuber, attorney, certified specialist in public procurement law and partner at Baker Tilly, examined the implications for procurement practice in Berlin on November 12, 2026. Their workshop was titled “Can the Bidder Really Deliver? Bid Evaluation in the Age of AI: When Perfect Bids Must Pass the Reality Test“.

When every bid sounds best-in-class

The shift goes beyond faster drafting. AI helps bidders align statements, rationales and pricing precisely with the award criteria. Even less experienced companies can now submit a highly persuasive bid. In consulting and other services, this applies in particular to concepts for team organization, project management, quality assurance and knowledge transfer.

For contracting authorities, it becomes harder to tell which differences between bids point to real differences in later performance. A polished staff substitution plan says little about whether onboarded staff are actually available. A professionally drafted consultant profile does not prove the experience it describes. AI can widen the gap between the quality of the presentation and the quality of the service.

Award criteria must capture meaningful differences

This is where Section 127 of the German Competition Act (GWB) comes in. Award criteria must be linked to the subject matter of the contract, ensure effective competition and allow effective verification of whether they are met. These requirements already apply when the criteria are designed.

For the evaluation of concepts, this raises a demanding question: is the yardstick still fit to identify the most economically advantageous tender when AI equips bidders of varying capability with similarly persuasive concepts?

Similar texts do not automatically render a criterion unlawful. It becomes problematic, however, once the evaluation no longer establishes a reliable link between the written presentation and the service actually owed. An organizational concept should therefore set out specific responsibilities, planned resources and verifiable processes. Advantages that earn points must later be contractually owed. A detailed scoring rationale cannot cure an unsuitable evaluation standard.

Detecting and banning AI only goes so far

There is no dedicated procurement law framework for AI-generated bids. The EU AI Act imposes, among other things, AI literacy obligations and, depending on the system and its use, further requirements. It does not, however, provide its own procedure for assessing AI-assisted bids. The general rules of procurement law apply, together with data protection and confidentiality rules in particular.

Moreover, the finished text does not reliably reveal how a bid was produced. Linguistic tell-tale signs and the output of AI detectors provide no reliable proof of how a text was created. A blanket ban on AI therefore raises not only questions of objective justification but also serious problems of proof.

Targeted rules, by contrast, may be necessary, for example for confidential tender documents. A confidentiality agreement can prohibit entering such documents into AI systems or make this subject to specific safeguards. Section 5(3) of the Ordinance on the Award of Public Contracts (VgV) allows contracting authorities to impose such requirements to protect confidential information.

Contracting authorities must also answer for their own use of AI

The same questions of control and accountability arise when contracting authorities use AI. Before processing bids, personal data or confidential work product, they need to agree on the purpose, system, data access and safeguards with the relevant department, data protection and information security.

AI-assisted evaluation deserves particular scrutiny. A plausibly worded rationale does not show which factors actually influenced the model. The contracting authority must review any suggestions on the merits against the bids and the published criteria, and adopt a sound evaluation as its own. Responsibility cannot be handed off to a black box.

Bid evaluation must test whether delivery is actually feasible

The easier it becomes to draft persuasive performance commitments, the more important a substantive review of the bid becomes. Contracting authorities should examine more closely whether staffing, resources and workflows can support the promised service. The use of AI alone, however, is no ground for exclusion. Nor is a tactically optimized concept automatically a false performance commitment.

Contracting authorities are generally entitled to rely on information provided by bidders. Where specific doubts arise, for example because staffing levels, timeline and promised controls do not add up, they must clarify whether the bid can be delivered. A further blanket confirmation that all requirements will be met does not, as a rule, dispel such doubts. What is needed are answers that address the specific contract, the staff deployed and the proposed working method.

Clarification must establish who will deliver which service with which resources. In an open procedure, it must not become a means of improving the bid after the fact: Section 15(5) VgV prohibits negotiations on changes to bids and prices. Where actual discrepancies or misleading information material to the award decision come to light, the relevant grounds for exclusion must be examined. Culpably misleading information may, for instance, trigger Section 124(1) no. 9(c) GWB. What matters is the established deficiency, not the suspected use of AI.

Verification presentations make commitments testable

Beyond clarification in individual cases, the tender conditions can provide for a verification presentation by bidders from the outset. Its purpose is to test selected information and performance commitments through specific verification steps. In consulting engagements, for example, the named individuals can be asked ad hoc questions about their stated experience and their own role on a project. Which decisions did they make themselves? How did they resolve a difficult situation? Depending on the contract, short case studies or proof-of-concept tests may also be appropriate.

Verification must follow a standard that is transparent in advance. A verification matrix defines the items to be tested, the essential minimum requirements, the conditions and the permitted aids. Above all, it determines when the test is passed and when it is not. Under this approach, no additional ranking points are awarded. Limiting the test initially to the provisionally first-ranked bid is an option if objectively justified and specified in advance.

A single uncertain answer must not be equated with a lack of experience. The decisive question is whether the announced essential requirements are met overall under the defined standard. Elimination requires proof that a mandatory, validly specified test item has not been met. Test tasks and permitted aids must match the service later owed. This turns the presentation into a transparent reality test of the claims made in the bid.

AI delivery concepts as a qualitative award criterion

Contracting authorities can also make the use of AI in contract performance an express part of the competition. Under Section 58 VgV, a contract-specific AI delivery concept can be designed as a qualitative award criterion. This makes it possible to assess how the planned use of AI contributes to the quality, reliability and efficiency of the specific service.

The bidder explains, for example, at which stages and with which systems it intends to use AI, how sources and outputs will be reviewed by subject-matter experts, and which tasks remain with the responsible staff. The concept should set out the promised performance improvements, any input required from the contracting authority, and how errors or system outages will be handled. What is evaluated is the contract-specific benefit of a robust approach. Merely naming an AI system does not justify additional points.

AI surcharges and efficiency gains in pricing

Remuneration also merits a closer look. Contracting authorities expect AI-driven efficiency gains to be reflected in bid prices, for example through a binding AI discount. Bidders, on the other hand, seek to charge AI surcharges to recover additional costs, particularly for licenses and usage-based fees. Fewer staff hours may therefore come with higher technology costs.

The price schedule can make this interplay transparent. What savings does the contracting authority receive? Which AI fees are added, and how are they calculated? Discounts and surcharges must relate to the same service offered and be reflected in a comparable total price. A headline AI bonus must not obscure the fact that additional fees eat up the claimed price advantage.

Binding performance commitments in the contract and verifying them through AI audits

What scored points in the competition must be delivered during contract performance. The relevant performance and AI delivery concepts, with their specific commitments, must therefore be incorporated into the contract on a binding basis. These include the named staffing, workflows, promised quality controls and agreed deliverables. With AI-optimized bids in particular, it must be clear against which yardstick actual performance will be measured.

A contractual AI audit clause can secure this link in practice. It allows the contracting authority to verify, on set dates, how the commitments are actually being implemented. Are the agreed systems being used? Are the promised expert reviews taking place? Is the AI delivery concept achieving the promised services and quality levels? This requires appropriate evidence, information and inspection rights, and cooperation obligations. Data protection and trade secrets must be safeguarded.

For selected key performance commitments, contractual penalties for culpable breaches may also be considered. They require clearly defined breaches, reasonable amounts and an appropriate overall cap. A contract-specific probationary period with defined grounds for termination can also help secure compliance with core bid commitments.

Whatever wins the award must be verifiable in advance and enforceable under the contract. AI makes that link more important. Contracting authorities should therefore develop evaluation criteria, verification measures and contract terms together. Bidders, in turn, must check before submission that their optimized commitments are sound in terms of staffing, expertise and commercial viability.

Baker Tilly’s public procurement team advises contracting authorities and companies on aligning AI use, procedure design and contractual safeguards. Get in touch. 

Share this article:

Author of this article

Dr. Christian Teuber

Partner

Attorney-at-Law (Rechtsanwalt), Specialist Lawyer for Public Procurement Law

What can we do for you?

Talk to us. Simply without obligation

Get in touch