Reviewing extracted data

An extracted fact in enisa.ai is a proposal that comes from reading your documents — with its document, its page and the exact source sentence — and it is never used until you confirm it. By the end you will know how to accept, correct or reject each proposal and why that minute of review decides the quality of your application.

What an extracted fact is

When a document is processed, we look for the key facts an application needs: share capital, incorporation date, last year's revenue… Each fact found is stored as a proposal, with its full provenance: which document it is in, on which page, and the exact sentence it comes from.

The rule that governs everything

A proposed fact is not used until you confirm it. The AI never drafts from a pending proposal — only from accepted facts. That is the difference between a tool that gets the numbers right and one that confidently invents them.

Reviewing the proposal queue

1. Open your company's data review: you will see the already-accepted facts with their sources and the proposals awaiting review.
2. For each proposal, read the quoted sentence and compare it with the proposed value.
3. Choose one of three:
   • Accept — the value is correct as is.
   • Correct — type the right value. Your correction does not erase the original proposal: it is recorded as superseded, so the provenance is preserved.
   • Reject — the fact does not apply (wrong document, outdated figure).

When two documents contradict each other

If the deed says share capital of €3,000 and the deck says €3,500, we do not silently pick for you: you will see both sentences, side by side, and you decide which is the truth. Internal contradictions are among the first things an ENISA analyst spots — better to resolve them here.

Why correcting matters

Accepted facts are what the AI cites when it drafts and what feeds the consistency checks. A wrongly accepted fact propagates: it appears in several answers, citation included, looking verified. A minute reviewing the queue is worth an hour correcting the draft.

Also, if a fact is missing during generation, the question is left at Faltan datos and whatever you answer there enters as a fact supplied by you.

Common issues

What if a proposed fact comes from an old document?

Reject or correct it. If the whole document is outdated, consider deleting it so it stops generating proposals.

What if I accepted a wrong fact?

Correct it: the correction supersedes the previous value and is what will be used from then on, without losing the trail of where each version came from.

What if a fact that is in the document was not extracted?

Check the document is ready (not "Analizando" or "Sin texto legible"). If the fact still does not appear, the AI will ask you for it during the run via "Faltan datos", and your answer is recorded.