What Prometiam serves today, what we are building next and what we are exploring. Each item states its limits, and a rule applied to a register is not called AI.
In use today. Each item states where it applies.
Scores public contracting bodies from their award history with two models: how likely the buyer’s next award is to draw a single bid compared with similar tenders, and how likely a supplier it awards to is to enter insolvency within 24 months. Each score comes with its probability and the awards behind it. Served by GET /procurement/buyer/{id}, GET /procurement/buyers and GET /procurement/relationship.
Limits: Beta. Covers only Spain and France. A score describes the buyer’s history, never a supplier or a single tender. The counterparty score is a weak signal: read it with the evidence awards, not alone. A buyer needs at least 20 awards to be scored. France discloses a bid count on only part of its awards. No credit score and no supplier score.
Read the use case · API documentation
What we are building. No dates: an item moves up when it is live.
Send the names and addresses from a CRM export or an invoice file as they are written, and get back the register record each one most likely is, with a confidence and the reason for the match: the name without its legal form, the same city, the same identifier.
Limits: A suggested match is not a verified identity: check a low-confidence match before you rely on it.
One paragraph in plain words about a company, written only from the fields the API returns (status, age, size class, recent events, insolvency notices, the sanctions screen), each sentence tied to the field it comes from, in the reader’s language.
Limits: It adds nothing the record does not hold: no opinion, no credit view and no guess where a field is empty.
A weekly summary of what changed across the companies you monitor, the most serious first (an insolvency notice before a new address), by e-mail or webhook, with a link to each alert.
Limits: It summarises the alerts monitoring already sends and watches nothing they do not.
Ready-made workflows for n8n, Make, Zapier and AI assistants that put the API to work: a supplier onboarding check, KYB when a customer signs up, a weekly review of a portfolio.
We are evaluating the public-buyer risk score for more countries.
Limits: Until then, the score covers only the countries listed under Available now.
Research. An item is offered only once it passes its test.
The probability that a company enters insolvency proceedings within the next 12 months, read from register signals such as late accounts, changes of officers or address and reductions of capital.
Limits: Offered only for countries where it beats a simple baseline on companies it never saw in training (an out-of-time test). We serve no credit score and no supplier score today.
For each possible match in a sanctions screen, a short account of what matched and what differs (the name variant, the date of birth, the nationality, the list), so an analyst decides faster.
Limits: It never clears or confirms a match on its own: the decision stays with the analyst. Sanctions screening is beta.
Monitoring alerts and risk flags are rules applied to official records, not models, so they are not listed here.