More control and trust in automated document processing
Intelligent Document Processing (IDP) enables the automated processing of documents within DocuWare, particularly in the areas of splitting, classification, and extraction. This automation delivers significant efficiency gains, but at the same time introduces the challenge of ensuring the quality of AI-generated results before they are used in downstream processes or stored in file cabinets.
With the introduction of IDP validation in classic DocuWare, customers can now validate key steps of the document processing workflow directly within the Monitoring UI, so AI results can be reviewed before processing continues.
Why IDP validation is important
In automated document processes, even small deviations in splitting, classification, or extraction can impact downstream processes. IDP validation addresses this by introducing a controlled step within the automated process where results can be reviewed and confirmed before the documents are processed further.
At the same time, each validation action serves as feedback to continuously improve the underlying AI models. This creates a system that evolves with use and delivers increasingly reliable results over time.
Validation in the IDP workflow
Validation can be applied across all key processing steps. It is available for splitting and classification across all configurations and can be configured per extraction setup. When validation is enabled, the IDP workflow pauses automatically at the validation step until user confirmation is provided. Once validation is completed, the IDP workflow continues with the confirmed data, while the feedback is used to further improve the system.
During validation, the question arises as to how returned feedback will be used for model training in the future, and whether automatic retraining will take place or if manual steps will still be required. A corresponding feedback function directly within the iframe is currently being implemented.
Integration
The feature is fully integrated into the DocuWare Monitoring UI, which serves as the central interface for monitoring IDP processes. To make this additional step visible, a new status has been introduced to indicate when a processing job is waiting for validation.
In this context, a “job” refers to an automated processing instance within the IDP system, where a document or a set of documents is processed through defined steps such as splitting, classification, and extraction. This ensures that it is always clear when a process has been paused and requires validation.
Implementation
Implementation is carried out within the Monitoring UI to ensure consistent integration into existing IDP workflows. This allows the entire interaction to stay in the familiar environment and fit seamlessly into established processes.
Application in real-world scenarios
IDP Validation is particularly relevant in scenarios where the quality of automated document processing is critical and results must be reviewed before further processing.
A typical process begins with the automated processing of an incoming document through splitting and classification within the IDP system.
If validation is enabled for specific steps, the process pauses accordingly — beginning with splitting and continuing with classification and/or extraction — to allow user verification.
The result is presented to the user for review and validated within the Monitoring UI. The feedback is then sent directly back to the system before the process continues with extraction. The document is subsequently processed further or stored based on the validated results.
This approach creates a controlled transition between automation and human validation, ensuring that only verified results enter downstream processes.
Building the foundation for adaptive document intelligence
With IDP Validation in DocuWare Monitoring, a controlled step is introduced into automated document processing that improves the quality and reliability of results. At the same time, it establishes a continuous improvement loop where user feedback directly contributes to the optimization of AI models.
In this way, the feature combines automation with targeted control, ensuring that document processing results become more accurate and reliable over time. Validation is currently configured as a binary decision per step — meaning it is enabled or disabled for splitting and classification, as well as per extraction configuration. However, it will be enhanced in the future through the integration of confidence scores, allowing validation to be triggered only when it is actually needed.