AI in procurement

Three applications that prove their worth

1. Searching by meaning, not by keyword

The biggest loss in public procurement comes from tenders you never saw. Titles are written in formal, bureaucratic language, so searching for exact keywords means you miss notices described differently. A language model can match a query to the meaning of a notice—"disinformation analysis" hits on "monitoring and quantitative analysis of media content," even though they share no words other than conjunctions.

2. Documentation analysis

Specifications with attachments often span hundreds of pages, yet the "go or no-go" decision is based on a dozen sentences scattered throughout: participation requirements, bid bonds, deadlines, contractual penalties, and required documents. Extracting these fragments is a mechanical task, and the model does it in a minute instead of half a day.

3. Checklist of documents to submit

Bids are lost due to a missing statement just as often as they are due to price. A compilation of everything that needs to be submitted, extracted directly from the documentation, serves as a checklist that no one has to write by hand.

Where AI should not be trusted

• Check deadlines and amounts at the source. The model can misinterpret a date or confuse a bid bond with a performance security. Always verify in the original document—the contractor bears the consequences of an error, not the tool.
• Chance assessment is not an oracle. The model does not know your actual capacity or the relationships your competitors have.
• Write the content of your bid yourself. A generated methodology description sounds smooth but says nothing—the evaluation committee reads hundreds of such texts and recognizes them immediately.

What this changes in practice

It changes the proportions. With 55 681 active proceedings, a manual review means you read a dozen or so sets of documentation per month and make your selection from those. Automatically filtering out what you do not qualify for—because you lack references, because the bid bond exceeds your capabilities, or because the deadline is unrealistic—leaves you to read only those proceedings where you actually have a chance.

We remind you of the data that justifies this: the median is 3 bids per tender, and 23% of proceedings are settled with only one bid (sample of 67 952 decisions from the last year). The market is not overcrowded—it is opaque. That is a difference that can be bridged with a tool.