Rejection Dutch-speaking chamber

A tender 62 per cent cheaper need not be abnormal: the Council of State accepts AI and automation as an explanation for the price gap in the Flemish framework agreement for digitising incoming mail

Ruling nr. 266840 · 28 May 2026 · XIVe kamer

NV I., ranked third in lot 1 of the Flemish framework agreement for digitising incoming mail (up to EUR 10.5 million), sought suspension because the Facilities Agency had not asked NV Y. — whose tender it said was 62.62 per cent cheaper — to justify its prices, but the Council of State held that the contracting authority had carried out a genuine price examination and could reasonably explain the gap by an in-house AI model, automation and economies of scale, and dismissed the application, contrary to the auditor’s opinion.

What happened?

The Facilities Agency (Facilitair Bedrijf) of the Flemish government used a competitive procedure with negotiation to tender framework agreements for digitising (administrative) documents, divided into three lots. The dispute concerned only lot 1, ‘Digitisation of incoming mail’: the contractor receives a customer’s mail daily, opens it, scans it, records the metadata and delivers the documents digitally. The framework runs for three years, renewable once for one year, with a maximum value of EUR 10,500,000 excl. VAT. Orders up to EUR 250,000 are placed in cascade according to the final ranking; larger or complex orders through a mini-competition. The award criteria were quality of service (45 points), key account team (10 points) and price (45 points). Five candidates were selected on 26 September 2024; four submitted an initial tender. After negotiations on 7 and 11 February 2025 one bidder withdrew. The remaining three — NV I., NV Y. and the public-law company B. — submitted a first revised tender by 11 March 2025 and, after a demonstration of their platform, a second revised tender by 14 April 2025. On 22 April 2025 the Facilities Agency asked all three, under Article 35 of the Royal Decree on award procedures, to explain the price build-up of several heavy items. A further round on the GDPR followed: B.’s tender was found substantially irregular because its processing of personal data did not match the specifications, and on 18 September 2025 B. was given until 3 October 2025 to regularise it. NV I. itself had stated after the meeting of 18 July 2025 that a certain indexing process would not apply and supplied an amended list of sub-processors. Negotiations were closed on 8 October 2025; the final tenders were opened on 29 October 2025. The award report of 22 December 2025 ranked NV Y. first with 98.89 points, B. second with 50 points and NV I. third with 37.78 points. On 6 April 2026 the Flemish Minister-President, responsible for facilities management, awarded lot 1 to the three bidders in that order. NV I. applied to the Council of State on 30 April 2026, raising three pleas. The first plea concerned price. According to NV I., Y.’s tender was 62.62 per cent cheaper than its own and 83.86 per cent cheaper than B.’s, yet the Facilities Agency had treated the prices as normal on the basis of ‘abstract’ reasons — intellectual services, artificial intelligence, economies of scale — without ever requesting a price justification, not even on the BAFOs. The Council disagreed. The award report and the confidential documents showed that the authority had carried out a general price examination at every round: comparing total prices with each other, with the average and with the estimate, examining unit prices, requesting price explanations and drawing up an internal memo on the price differences in the BAFOs. A significant price gap may be an indication, but does not necessarily point to an apparently abnormal price. The Facilities Agency explained that Y. uses an in-house AI model to scan and sort mail and read the metadata, without licence costs and with far fewer staff, whereas NV I. and B. still work largely manually; Y. had shown this during the demonstration and had already been awarded lots 2 and 3. That the specifications mentioned only OCR and ICR, and that AI was ‘not necessary’, did not prevent bidders from using it. The Council added that the margin of discretion in a negotiated procedure with a substantial intellectual-services component is wider than for a works contract, and that NV I. could not use its own experience with AI as the yardstick. Since no apparently abnormal price had been found, no inquiry on the final tenders was required either. The second plea — B. should not have been allowed to regularise its tender after negotiations had begun — failed for lack of interest. The file showed that the Facilities Agency had already raised GDPR questions on NV I.’s own tender on 7 February 2025, that NV I. had then revised its tender of its own accord, including by dropping AI applications, and that it supplied a complete list of sub-processors, including hosting, only after 18 July 2025. It had thus had the same opportunity as B. and suffered no prejudice. The third plea attacked the quality score: Y. received the maximum despite weaknesses noted in the reasoning, and the scores allegedly relied on undisclosed sub-criteria. The Council pointed out that the specifications themselves provided that the best quality tender receives the maximum score and the others are recalculated proportionally: Y. scored a provisional 40.5 out of 45 (recalculated to 45), B. 36 (40) and NV I. 27 (30) — exactly the required minimum. The award report set out strengths and weaknesses at length and compared the tenders; NV I.’s tender was called ‘predominantly theoretical and general’ and ‘less innovative’. Selectively quoting ‘empty sentences’ and a home-made table ending in ‘Etc.’ was not enough. Nor was there any sign of a weighting per assessment aspect. The application was dismissed. NV I. pays EUR 200 court fee, EUR 26 contribution and EUR 770 procedural indemnity; the intervening parties each bear half of the EUR 300 court fee for their intervention.

Why does this matter?

In this judgment the Council of State expressly accepts that the use of artificial intelligence and automation can explain a large price gap without the contracting authority having to ask the cheapest bidder for a formal price justification. That is an important signal now that more and more service providers cut their prices with in-house AI tools. The judgment sharpens the distinction between two stages: the general price examination under Article 35 is mandatory and must be carried out carefully, but the inquiry under Article 36 follows only if that examination reveals an apparently abnormal price — a large difference alone is not enough. Anyone challenging that examination must show that the authority exceeded its margin of discretion, and according to the Council that margin is wider in negotiated contracts with a large intellectual-services component. Notably, the Council departed from the auditor’s opinion. The second plea shows how the interest requirement works: a complaint about a competitor’s regularisation falls away if you were given the same opportunity. And the third plea confirms that a global assessment with a recalculation formula can lead to a maximum score despite weaknesses.

The lesson

For bidders: a spectacular price gap with the winner is not a plea in itself. If you allege a deficient price examination, show concretely why the authority’s explanation — here AI, automation, economies of scale — does not hold for the items that make the difference, and request access to the file to be able to do so. Your own way of working is not the yardstick for what a competitor can do. Only challenge a competitor’s regularisation after checking that you were not given the same opportunity to adjust. And if you dispute your quality score, address the reasoning as a whole rather than lifting out a few sentences. For contracting authorities: this judgment shows what keeps a price examination standing — comparing total and unit prices with each other, the average and the estimate, a targeted request for price explanations on the heavy items, an internal memo on the final tenders and an award report that names the causes of the differences per category of items. A demonstration of the proposed solution during negotiations can later prove invaluable in supporting that explanation.

Ask yourself

If you challenge a much cheaper competitor, can you show item by item why its price cannot be explained by technology, automation or scale? Have you requested access to the administrative file to see what price examination the authority actually carried out, including the confidential documents the Council can consult? Were you yourself given the opportunity during negotiations to correct ambiguities in your tender (for instance on the GDPR) before you challenge a competitor’s regularisation? Do you understand the recalculation formula in the specifications and what it means for the score of the top-ranked tender, even if it has weaknesses? As a contracting authority: does your award report explain per category of items where the large price differences come from, and did you repeat that examination at every tender round? As a contracting authority: do you keep an internal memo on the price analysis of the final tenders, even if you decide not to hold an inquiry?

About this database

The Council of State (Raad van State / Conseil d'État) is Belgium's supreme administrative court. In disputes over public procurement — from contract awards to tenderer exclusions — the Council of State is the final arbiter. The rulings in this database are summarised by TenderWolf in plain language, with practical lessons for tenderers and contracting authorities. View all rulings →