Tender Monitoring with AI: How to Filter CPV Codes and Assess Fit

Where CPV Helps and Where It Is No Longer Enough

In tender monitoring, the CPV code is a good starting filter, but it does not show the full picture of a contract. According to Minerva, the CPV code alone provides only a rough match, while the actual requirements are usually hidden in the SWZ and its annexes.

Minerva also describes a practical filtering setup: region, type of works or CPV codes, and the contract value threshold. Atlas Przetargów adds location, offer submission deadline and an "active only" filter, which makes it possible to quickly narrow the list of procedures before further assessment.

How AI Assesses the Fit of an Offer

In AI tools, fit assessment does not rely solely on a single code. According to TenderYeti, the system reads the company description in plain language and classifies new licences or tenders on a scale from 0.00 to 1.00, based on keywords and context.

Minerva describes a similar mechanism in more detail: the system compares the meaning of the entire documentation with the company profile, not only the tender title and CPV code. According to Minerva, such contextual search can catch procedures with an incorrect, overly general or unusual code that a classic filter might not show.

How to Build a Sensible Monitoring Setup

A layered setup works best. First, you set broad formal filters, then you review the documentation, and only at the end do you prioritise offers that fit the business. Apify shows this scheme directly: you can filter TED by CPV code, country and publication date, and then retrieve data such as value, deadline and links to notices.

1. Start with main and additional CPV codes, not with a single number.
2. Add country, region and the contract value range.
3. Assess the full description of the contract, not only its name and code.
4. Set up a ranking of chances if the tool shows a match score.
5. Verify whether the system cites the source of the documentation and lets you reach the original notice.

In practice, this approach reduces information noise. Bidly states that it combines keywords, target country, CPV codes and the contract value range, while TenderYeti emphasises that the user does not have to remember Boolean syntax or the CPV catalogue if they describe their business in simple language.

What to Watch Out for During Implementation

The biggest risk is a filter that is too narrow. If the contracting authority assigns an overly general code, monitoring based solely on CPV may miss a valuable procedure. Minerva also points out that a real assessment requires reading the SWZ, not just the tender title and code.

The second risk is an excess of results without prioritisation. That is why it is better to choose tools that combine formal filters with semantic assessment and show why a given procedure was considered a match.

Lub System helps B2B companies implement AI, automation and IT solutions end-to-end - from strategy to deployment. See our services or get in touch to discuss your case.