DocuWare IDP now passes results that need a human check to the DocuWare client. A new IDP Validations tab lists every document that is waiting for a check, so the review happens in the same client that people already use for their documents.
IDP runs its pipeline on incoming documents: splitting, classification, and extraction. Each result carries a confidence value. Where a result needs to be confirmed by a person, the document is listed for validation instead of being processed automatically. Processing continues once the result is confirmed, and the document is stored.
The tab is visible to users with the Configure IDP right. The people who set up IDP therefore also release the documents that need a check.
Components: DocuWare Cloud, Web Client, IDP
Applications
An employee from Peters Engineering scans a stack of mixed contracts as one file. IDP splits it with a confidence value below the threshold, and the IDP administrator confirms the document boundaries before the contracts are stored.
IDP classifies an incoming document with low confidence. The employee sets the correct document type, so the document is stored in the intended file cabinet instead of a general one.
An IDP configuration requires validation for AI results. The configurator compares an ambiguous invoice total with the document shown beside it and corrects the value before the document is stored.
Here’s how
You find the list under DocuWare client > IDP Validations, next to Document trays, Search, and Tasks. Each row shows the columns Document, Configuration, Validation Type, and Date/Time.
Validation is set separately for each processing step.
For splitting and classification, the setting Validate if confidence is lower than in DocuWare Configurations > DocuWare IDP applies to the whole organization, with a separate threshold for splitting and for classification. A result at or above the value is processed automatically, and only results below it are sent to the list.
For extraction, the setting User validation required for AI results is part of each IDP configuration and becomes available once an extraction model has been trained. It has no threshold: either every extraction result is validated, or none.
The configuration name is shown in the list, so several configurations can send documents to the same list. A document stays in the list until its result is confirmed. Processing then resumes at the validation step and stores the document, so a document that is never validated is never stored. A corrected result is applied, and the document continues with the corrected values.
More information