How AI Scores a Tender Against Your Product Catalogue

Most tender alert tools work on keywords. You type in a few terms, the system emails you every notice that contains them, and you spend your mornings deleting the ones that do not apply. The problem is not that keyword alerts find too little. It is that they find too much of the wrong thing, and miss the right thing when the buyer happens to use different words.

A better approach scores each tender against your actual product catalogue. This article explains how that works, step by step, and why catalogue-level matching produces a short list you can trust.

Start with what you sell, not what you search

The foundation of accurate matching is a precise picture of what your company actually does. A keyword is a guess. A catalogue is a fact.

Trinta begins by reading a company's products and its website. From that, it builds a multilingual keyword profile: the materials, the product families, the standards you hold, the sectors you serve, and the words a buyer in each of your markets would use to describe what you make. A medical-gas manufacturer in Portugal does not just sell "gas." It sells oxygen plants, vacuum systems, medical air units, pipeline components, and the certifications that go with them, described in Portuguese, French, Spanish, English, and Arabic depending on where the buyer sits.

That profile is the reference point. Every tender is then measured against it.

The limits of keyword matching

Keyword matching fails for two structural reasons.

First, vocabulary drift. "Supply and installation of an oxygen generation plant for a regional hospital" and "fornecimento de central de produção de oxigénio medicinal" describe the same opportunity in two languages. A keyword set tuned to one will miss the other entirely. Even within a single language, "PSA oxygen generator," "oxygen concentrator plant," and "on-site oxygen production system" are three phrasings of one product.

Second, false positives. A search for "oxygen" returns water treatment tenders, environmental monitoring contracts, and welding gas supply notices that have nothing to do with hospital infrastructure. The buyer used your keyword. The opportunity is not yours.

Catalogue-level scoring addresses both. It reads for meaning, not for string matches, and it weighs how closely the meaning lines up with what you actually sell.

Semantic matching: reading for meaning

The core technique is the vector embedding. An embedding turns a piece of text into a list of numbers that captures its meaning. Two texts that mean the same thing produce similar numbers, even when they share no words and sit in different languages.

When a tender is published, Trinta generates an embedding from its title, description, classification codes, and buyer. It compares that against the embedding of your keyword profile. The closer the two, the higher the semantic match. This is how "photovoltaic system procurement for public facilities" can be recognised as relevant to a solar installer whose catalogue never used the word "photovoltaic."

Semantic similarity is the strongest single signal. On its own, though, it is not enough.

Fit scoring: the factors beyond similarity

A tender can read as semantically close and still be a poor fit. A two-person consultancy should not chase a contract that needs a hundred-person delivery team. A company licensed only in Kenya should not be shown a notice that can only be served from inside Brazil.

So semantic similarity is combined with structured factors that describe fit:

Industry overlap. Do the tender's classification codes (CPV in Europe, UNSPSC at UN agencies, national equivalents elsewhere) line up with the sectors in your catalogue? This is metadata the buyer themselves attached, which makes it a reliable cross-check on the semantic read.

Geographic fit. Is the company active in, or able to deliver to, the tender's country? An opportunity you cannot legally serve is not an opportunity.

Company size fit. Does the contract value sit within the range your company can realistically deliver? Value alignment keeps the short list grounded in what you can win and execute.

Historical signals. Has the company engaged with similar tenders before, or won contracts of this kind? Past behaviour is a quiet but useful predictor of future relevance.

Each factor adjusts the score up or down. A tender that is semantically close, correctly classified, in your country, at your scale, and similar to work you have done before rises to the top. A tender that is close on words but wrong on geography or scale falls away.

From thousands of notices to a daily short list

The point of scoring is selection. Trinta has analysed more than 1.15 million tenders from official sources in over 145 countries. No team can read that volume. The scoring engine reduces it to the handful that genuinely match your catalogue, ranked by fit, and delivers them as a daily digest.

Instead of opening twenty portals and deleting forty irrelevant alerts, you open one email and read the ten opportunities most worth your time. The work you save on searching goes straight into the work that wins: understanding the requirement and writing a strong bid.

Why catalogue precision compounds

A keyword alert is static. It returns the same kind of noise on day one and day one hundred. Catalogue-level scoring improves. When you mark an opportunity as relevant, the system learns to surface more like it. When you dismiss one, it learns to filter it out. Over weeks, the short list narrows toward exactly the work your company is built to win.

There is a real-world proof point. Ultra Controlo, a 40-year medical-gas manufacturer in Portugal, put its catalogue into Trinta and surfaced a EUR 700,000 tender in its first week. It won that contract two weeks later. In month one, the platform surfaced more than EUR 30M in qualified opportunities matched to the company's products. None of that came from typing keywords into a search box. It came from the system knowing, in detail, what the company makes and where it can win.

The takeaway

Keyword alerts answer the question "which notices contain my words?" Catalogue scoring answers a far more useful question: "which public tenders is my company actually built to win?" The first floods your inbox. The second hands you a short list. For any company that sells real products into public markets, that difference is the difference between monitoring procurement and competing in it.

Share this article

Related articles