How matching works
From your website to a ranked list: profile, geographic gate, keyword, CPV and meaning-based retrieval, then ranking.
Matching answers one question: of everything published today, what could your company credibly win? Here is what happens, in plain language.
We read your website
We fetch your homepage and a handful of relevant subpages and derive a profile: a short summary of what your company does, your sector, the keywords a buyer would use for it, and matching CPV codes. You review and correct that profile before anything is searched.
A profile created manually skips this step: the summary, keywords and CPV codes you typed are the profile, and every later step runs on them alone.
Geographic gate
Countries and regions from the profile are applied first. A tender outside your area never enters the ranking, no matter how well it fits otherwise.
Three ways of finding candidates
Three retrieval legs run in parallel over the open, biddable notices:
- Keywords: full-text search over title and description.
- CPV: how closely the tender's classification matches yours, from an exact code down to the same division.
- Meaning: a multilingual similarity search that finds tenders describing your work in words you did not supply.
A fourth signal comes from award history: what your company (or companies like it) has won before, and whether the buyer has awarded to you in the past.
Fusion and ranking
The results of the legs are merged so that a tender found by several legs rises to the top, and the best candidates are then scored by a language model against your profile. Anything it scores as irrelevant is dropped rather than ranked low.
The list you see
What remains is stored as your match list, with a score, a reason and a snippet per entry. The list is living: a re-run adds new hits and keeps the ones you already worked on.
Why results are never a black box
Every match can explain itself: which region rule let it through, which CPV tier hit, which keywords matched. See Why did this match.
What matching does not do
- It does not read documents. Ranking is based on the notice, not on the specification inside the attached ZIP. Use the AI connector to dig into documents.
- It does not judge whether you have capacity, references or certificates. It finds candidates; the go/no-go decision is yours.