Intelligent document processing (IDP) enhancements

DocAI Studio introduces Classification accuracy benchmarking. Users can upload test documents to the correct class. DocAI Studio then shows how accurately they are classified, including an overview of all classes (in a "confusion matrix"). The solutions overview page also lists the most recently measured classification score for each solution. The accuracy is a single number (f1-score) for the entire solution.

Introduces a new Save to Local Drive/Share activity type that writes processed documents and extracted OCR data to a local drive or network share. The feature is as follows, depending on the studio:

• In TotalAgility Advanced Studio and TotalAgility Studio, this automatic activity is available in the Capture group of the activity palette. This activity can be added to a process like any other activity node. Refer to Save to Local Drive/Share activity.
• In DocAI Studio, a new Local Drive / Share export channel is added in the Document Export step, alongside the existing None, Download, Azure Blob Storage, and REST API channels. Refer to Use Local Drive / Share for export.

In a Tenant environment, this activity runs on Integration Server as the cloud server has no access to on-premises file systems, and Integration Server is the only valid execution target.

A lightweight, in-context editor in DocAI Studio enables users to author self-contained validation and formatting rules directly on a document field or document type, without creating reusable business rules in TotalAgility Advanced Studio. The editor gives Standard-tier users their only rule-authoring capability, while reserving reusable and advanced rules as an Advanced Studio functionality.

The same inline rules functionality is also available within an Extraction group in the TotalAgility Advanced Studio.

The new AI Table extraction introduced in an earlier release takes detected tables, sends them to an LLM, which then identifies the correct one and maps the columns to the customer’s field schema. Table detection is a prerequisite and is performed by a proprietary algorithm that is conservative (designed to find only “real” tables, with no false positives). As there are use cases where items like lists need to be extracted as tables, this release introduces an option for the table detection algorithm; customers can choose between the proprietary model and Microsoft’s Azure AI Document Intelligence Layout models via a simple setting in DocAI Studio.