Document Automation v.41 release - Review what's new and changed, and the fixes and limitations in Document Automation for the v.41 release.

Automation 360

Review what's new and changed, and the fixes and limitations in Document Automation for the v.41 release.

What's new


Document evaluation for improved accuracy Document evaluation helps you to understand how accurately a learning instance extracts data from documents. Use document evaluation when you want to validate extraction quality. Before deploying a learning instance, you can compare versions or confirm whether changes to prompts, rules, OCR providers, or other learning instance settings improve results. You can define the expected extraction result for a processed document as ground truth data for a particular document. Document Automation then automatically compares extraction results across learning instance versions with the ground truth data and calculates accuracy for the document. By comparing extraction results against defined ground truth data, you obtain a quantified accuracy score at multiple levels of granularity, enabling data-driven decisions when refining and promoting a version of the learning instance. Note: This feature is available only on Cloud deployments. Document evaluation for learning instances

What's changed


Ability to add field references in validation rules You can use field references in field and document rules to compare extracted values across form and table fields, copy values between fields, and include field values in error or warning messages. Use the field picker in supported condition and action fields to select the required form or table field references. The following are some scenarios where this feature can be useful: Cross-field validation: Compare an extracted form or table field with another field value. For example, verify that the billing address matches the shipping address. Dynamic correction: Use a referenced field's value in set value action. For example, populate one field using the value from another field. Dynamic messaging: Create dynamic error or warning messages that include actual extracted values instead of static text. Document Automation validation rules: conditions and actions
Enhanced visibility and management of Model connections in learning instances To improve the management of Model connections associated with learning instances, the following enhancements are introduced: View Model connection association: When a Model connection is associated with one or more learning instances and when you try to delete the Model connection, a notification is displayed indicating that it is currently in use. You can also view the learning instances that are associated with the Model connection. Filter learning instances by Gen AI provider: A new Gen AI provider column is available on the Learning Instances page, making it easier to identify and filter learning instances that use a specific Model connection. This helps administrators quickly locate and disassociate model connections when required. Automatic cleanup of Model connection associations: When a user is deleted, any private learning instances created by that user are also deleted. As a result, the associated Model connections are automatically disassociated, allowing administrators to delete Model connections that are no longer in use. Deleting a Model connection used in learning instances
Default regions for generative AI providers The Control Room now automatically sets the system default regions for External connections to match the region where the Control Room is deployed. The following regions are now supported as system defaults: United States, Europe, India, Australia, Japan, and Canada. For providers that do not support these regions, such as Anthropic Claude and Google Document AI and Google Cloud Vision (OCR), the system default continues to be the United States or Europe. Enable generative AI and other external connections to Document Automation

Fixes


You can now select a custom model that you created before updating to this release when you configure a learning instance. Previously, you were unable to select such custom models in such a scenario. Service Cloud Case ID: 02332382, 02331481
You will be able to process specific Spanish documents when using a generative AI provider option. Previously, extraction failed in such a scenario. Service Cloud Case ID: 02298914
When you permanently delete a request from the Requests page, any associated tasks with the request are also removed to maintain consistency across the Control Room. Previously, deleting a request did not delete its associated tasks, creating inconsistency across the Control Room. Service Cloud Case ID: 02300482
You can now get the correct field value when you draw a box around the data you want to extract in the validator. Previously, correct values were not extracted in certain scenarios. Service Cloud Case ID: 02197373
When a generative AI provider is enabled and all search queries are disabled for a learning instance, extraction results now remain unchanged for fields not using generative AI. Previously, enabling generative AI caused values in unaffected fields to change unexpectedly.
The ValidatorUI element now saves and displays the invalid reason value for a process on the Requests page. Previously, the invalid reason was not displayed in certain scenarios. Service Cloud Case ID: 02299199
If there is an existing learning instance that is using test mode and you import a learning instance with the same name, you can now use the Merge option to merge these learning instances without losing existing information. Previously, only the imported learning instance information was retained in such a scenario.
Execution mode and Queue timeout settings are now preserved after editing a process. Previously, these settings were not retained after editing and saving the process in the legacy Process editor.
When you process a document using a learning instance that is using a generative AI provider and the Adaptive Search Queries feature, the validator displays the correct Gen AI Provider. When you change the generative AI provider and check the validator without processing any documents, the Gen AI Provider will no longer display the updated generative AI provider and will correctly display the used generative AI provider. Previously, the updated generative AI provider was displayed in such a scenario.

Limitations


Data extraction might fail with the following error while using the Cloud Extraction Service: `INVALID_ARGUMENT: Invalid UUID string`. Workaround: Use an input form instead of process-level input variables to pass the required inputs to the learning instance process.
When documents are processed using Cloud Extraction Service, the default validator displays default queries instead of the user-defined Adaptive Search Queries even though user-defined queries are used for actual extraction. The test-mode validator displays user-defined queries correctly.
The following limitations are applicable when you download accuracy report data: When you download accuracy report data for a version of the document that does not contain extraction results, the download might fail. When you download accuracy report data for multiple documents and if extraction results are not available for one or more documents, the output might contain incomplete data. When you download accuracy report data for multiple documents, the output report might include an incorrect accuracy report header.
When documents are processed using Cloud Extraction Service, the validator displays the default queries instead of Adaptive Search Queries even though the Adaptive Search Queries are used for data extraction.
If a learning instance was using Pandas, oracledb, or Pillow Python scripts for data extraction in IQ Bot and you moved the learning instance to Document Automation using the IQ Bot - DA Bridge package, the Python scripts are not considered for data extraction when you process documents in Document Automation.
When you use actions in the IQ Bot - DA Bridge package, the icons for the actions are not displayed in the Task Bot editor.


Limitations from previous releases
The following limitations are applicable to Model connections: Model connections using the Automation Anywhere vendor is not supported. Workaround: Use the built-in generative AI providers when creating learning instances. In Model connections, the Fine-tuned and Grounded model types are not yet supported. Workaround: Use the Standard model type.
When you try to check in a process without having the required permissions, the audit log entry for this event will include Source as Control Room instead of Document Workspace and Item name as Unknown instead of <learning-instance-name>.
When you use a regular expression for a table field without defining an alias, the regular expression might not be considered for extracting the data. Workaround: Define an alias for the table field before using a regular expression.
When you process documents containing multiple tables in Standard Forms, the order of the tables extracted is reversed.
Document extraction will fail with the error “500: Internal Failure , Failed to initialize ABBYY’ when you change the default path (C:/ProgramData/AutomationAnywhere/GlobalCache) of Global Cache Location for your devices.
When you process documents with file names greater than or equal to 75 characters in test mode, you might not see improvements in the reprocessing time for such documents.
If you have disabled the OCR provider in the administrator settings and if you are using a language other than English for your Control Room, you will see an error to enable the OCR provider settings in English in the following scenarios: When you create a learning instance that is using an OCR provider When you change the OCR provider for an existing learning instance
When you use the Document Classifier actions (Classify, Classify documents, and Train Classifier) and Extract Data action in the Document Extraction package together in a bot, the bot will fail to execute. Workaround: Ensure that you create separate bots when using any of the actions from the Document Classifier package and the Extract Data action of the Document Extraction package. If you need to execute these bots in a sequence, include these bots in an Automation Co-Pilot process.
When a user processes a document on a custom process learning instance, the validate document count is not updated post extraction. Also, if the user submits the document, the validate document count is updated with a negative value.
A user with the Automation Co-Pilot administrator permissions is unable to view the Document Automation tasks that are assigned or requested and are in pending or complete status.
When you are using the IQ Bot Pre-processor package actions and if the Output folder path contains Japanese characters, you will see an error when processing documents. Workaround: Create an output folder in a folder path that does not contain any Japanese characters and provide the path in the output folder path field.