Whats the best way to identify incorrect documents? (Other than a low confidence level on fields?)
We have a SQL database of uploaded ID documents of differing type which we've trained individually and is working really well. (export to file then analyse)
Some staff have uploaded wrong ID'for example they have uploaded an receipt for the ID rather than the ID itself? Currently Ive created a new field call 'receipt' trained a bunch of receipt documents to find that field then say that field isnt available in other collections.
Therefore if I find a high confidence Receipt I know it likely they have uploaded a receipt.
I think what I want is an overall confidence of the document from any collection, therefore a low score would indicate that its an unexpected document (rather than best guessing and finding data and continuing)
Hi @TimAustinMills - nice use case for AI Builder!
You can use the field Collection confidence score. It has a value between 0 and 1. Where 1 indicates 100% confidence that the AI model knows how to extract data for that document. If this field is low (less than 0.75 for example) it probably means that the custom AI model hasn't been trained to handle that document.
You can use a condition to handle documents accordingly.
I hope this helps!
WarrenBelz
791
Most Valuable Professional
MS.Ragavendar
410
mmbr1606
275
Super User 2025 Season 1