AI does the reading. People make the decisions.
Our customers make decisions about loans, claims and taxes. Those decisions affect people, so our AI is built to be checked, explained and overruled.
Human review
Fields below a confidence threshold that you set go to a review queue instead of straight into your systems. Shield flags a document for a person to look at; it does not reject applications on its own. Final decisions stay with your team.
Confidence scores
Every extracted value carries a confidence score. You choose the thresholds: which fields must always be checked, and which can pass automatically when confidence is high.
Source for every answer
Each value links to its page and position in the original document. Assist cites the document behind each answer, and says so when the answer is not in your data.
Audit trail
We record what the system extracted, which checks ran, what was flagged, who reviewed it and what they changed. You can show an auditor or regulator how any number reached your books.
Error handling
When a document cannot be read, the system says so instead of guessing. Unreadable pages, unsupported formats and failed checks are reported clearly, with the reason. Corrections your team makes are logged.
Explained flags
A fraud or risk flag always comes with what was found and where. No unexplained scores. If your team disagrees, they can mark the flag as wrong, and that is recorded too.
What we will not do
- Make credit, claim or tax decisions on our own. We provide data and signals; you decide.
- Train models on your documents.
- Hide how a result was produced.
Questions about how we use AI? Talk to us. See also Security.