Norway publishes a remarkable amount of open data. The Brønnøysund Register Centre alone holds structured records on every registered company in the country, available under an open licence. The raw material for understanding a whole industry is, in principle, just sitting there. In practice, almost no one can use it to make a decision.
The gap between data and an answer
A buyer at an organization does not want a database dump. They want to know which local manufacturer can actually do the job, and they want to trust the answer. Bridging that gap means joining messy sources, filling the fields the registry leaves blank, and presenting it as a profile a human can read in seconds. The hard part is not the volume. It is making the result both useful and trustworthy.
Evidence-graded, not just generated
It is easy now to have a language model write a confident paragraph about any company. It is dangerous, too, because a confident paragraph with no source is indistinguishable from a fabricated one. So the rule we work to is simple: every claim on a profile must be traceable to where it came from. Generated text is allowed to summarise; it is not allowed to assert anything the underlying data cannot back up.
Open data is everywhere; usable, trustworthy answers are rare. The value is in the grading, not the generating.
That discipline shapes the whole system. Sources are attributed. Enrichment runs are budgeted and reproducible. Where the data is thin, the profile says so rather than inventing confidence. The output is graded by how well it is evidenced, so a reader knows the difference between a fact from the registry and an inference.
Where this is going
This is the work behind the discovery service we build for Locat3D, starting with suppliers across Nordland and Troms and growing from there. The pattern generalises far beyond manufacturing: wherever there is open public data and a decision that depends on it, the same approach (join, enrich responsibly, grade by evidence) turns a registry into something people can act on. That is what applied AI looks like when it is built to be trusted.
Oleksandr Kozachuk, Kaizenkodo