Too many irrelevant tenders
Irrelevant tenders in Tender Navigator come from two places: a filter whose keywords and codes pick up notices that are not about your work, and a prompt that is too vague for the AI to mark them Unmatched. Tighten the criteria first, then sharpen the prompt using the reasoning on the results.
Last updated
Decide which problem you have
- The filter reads too much. Open the filter: a high Evaluations performed count with mostly Unmatched results means the keywords and codes pick up notices that are not about your work. Every tender read costs an evaluation, so fix this first.
- The verdicts are too generous. Many Matched results that you would never bid on means the AI prompt does not state your criteria and exclusions clearly enough.
Make the filter read less
- Replace broad keywords with specific ones. "cloud" matches weather and printing notices; "cloud migration" and "Azure" do not. Remember that keywords are ORed, so removing a broad one narrows the filter more than adding a narrow one.
- Use precise classification codes. A CPV family such as 72000000 (IT services) reads far more than 72212000 (programming services). Pick the codes buyers actually attach to the work you do.
- Limit the countries to markets you can deliver in.
- Add a budget range where you know your typical contract size, keeping tenders without a published budget included unless you are sure.
- Set a deadline offset so that tenders closing in a few days, which you could not bid on anyway, are skipped.
- Split one broad filter into two narrow ones with different prompts, within your plan's active-filter allowance.
See Keywords, countries and codes and Creating and editing a filter.
Make the verdicts stricter
- Open a few wrongly Matched results and read the reasoning in the Tender Info panel. It shows which of your criteria the model relied on.
- Add the missing exclusions to the prompt: sectors you do not serve, certifications you lack, minimum contract sizes, on-site or language requirements you cannot meet.
- Say how to treat uncertainty ("if the notice does not clearly ask for X, mark it Unmatched").
- Save the prompt. The change applies from the next daily run; existing results keep their verdicts.
See Writing the AI prompt and Why a tender was matched or unmatched.
Make the noise easier to handle
- On the Results page, set Match to Matched to hide Unmatched results, and use the Favorites heart to keep the ones worth a look.
- Set the filter's digest to Weekly if daily emails are too many; evaluation still runs daily.
Related articles
- Writing the AI prompt for a filterHow to write the business criteria that Tender Navigator's AI evaluation judges every tender against: what to include, what to leave out, worked structure, and how to sharpen a prompt when verdicts look wrong.
- Why a tender was matched or unmatchedHow to find out why Tender Navigator's AI marked a tender as Matched or Unmatched: the reasoning on the result, the filter snapshot it was judged against, and how to change future verdicts.
- Keywords, countries and codes: how search criteria workHow Tender Navigator combines keywords, countries and classification codes in a search: every list widens the search (OR), one constraint is required, and multi-word phrases stay intact.
- Filter usage, stuck tenders and evaluation limitsWhat the usage counters on a Tender Navigator filter mean: Evaluations performed, Stuck tenders, the per-filter Evaluations limit, and how they relate to the plan's evaluations per billing cycle.