Looking for the Quick Start Guide? Head this way.
Check out the AI & Intelligence forum to get the best content about ServiceNow's AI products!
As of the Utah release
Table of contents:
Getting Started
Document Intelligence and other solutions
AI and Training
Limitations
Implementation & Troubleshooting
Getting Started
What is Document Intelligence and how do I get started?
The Document Intelligence Quick Start Guide is available in the community.
How do I get Document Intelligence?
Document Intelligence can be installed via a Store Application; it is compatible with San Diego and later releases.
It requires a special license. It is available for Automation Engine PRO, CSM PRO and FSO PRO, APO (as of Vancouver) customers but reach out to your Account team to inquire how to get a license. It is tied to a quota of pages with the option to acquire additional pages via Document Intelligence via add-on bundles with the license.
How do I get Document Intelligence Admin?
Document Intelligence Admin is a separate plugin that can be installed via a Store Application; it is compatible with Tokyo and later releases.
What use cases does it cover?
The focus of the product is the extraction of data from documents to drive high-value workflows.
For example, a document is loaded, the information is extracted, and it drives actions, such as approvals, flows and integrations.
Check with the appropriate BU (Automation Engine, CSM, FSO, APO) for specific use cases.
An overview was also presented during the AI Academy: Get started with Document Intelligence
Document Intelligence and other solutions
What is the difference between Document Intelligence for Automation Engine and CSM/FSO/APO?
The Document Intelligence solution is pre-packaged and integrated with CSM, FSO and APO; this limits the configuration needed. For other use cases, the Automation/Creator tools (Flow, etc.) can be used with Document Intelligence to process documents and extract values in the Now Platform.
Is Document Intelligence the same as Robotic Process Automation (RPA)?
The AI in Document Intelligence and the human-in-the-loop design make it possible to address use cases for highly variable documents. RPA with Document Intelligence is a powerful combination for customers wanting to support complex cross-enterprise workflows spanning multiple teams and 3rd party systems.
AI and Training
What “models” is Document Intelligence trained on? What algorithms does it use?
It's a multi-faceted approach developed by ServiceNow that cannot be oversimplified as a computer vision and classification technique. Our approach aims to develop highly trusted models for various use cases using several techniques to deliver those results.
Additionally, Document Intelligence is fully integrated into the Now Platform to drive end-to-end workflows and limit the number of point solutions running their own algorithms.
When we say that Document Intelligence is “trained,” does that mean I can process the same document multiple times to train the AI model?
Not really; Document Intelligence uses human-in-the-loop, meaning that the training happens as agents process documents. That process should happen with documents that are reflective of your workflow. Forcing the training with a single or too few documents can skew (or overfit) the model.
Can I tell Document Intelligence where my field is on the document (instead of typing and selecting the recommendation)?
Yes, the Draw tool allows to select values from the document directly.
Where is the data trained? Is the Machine Learning process happening on my instance?
Training isn’t done directly in the ServiceNow instance; it is done on a secure server with specific hardware to perform Machine Learning in ServiceNow’s Data Center. The process is encrypted and doesn't cache or store data.
When we delete a trained task in Document Intelligence, does the training data related to that task get deleted?
No.
Limitations
What languages are supported? What file types are supported?
See the docs for the full list of limits.
Are there any limits to be aware of?
The following are not supported:
- ‘Secured’, encrypted and password-protected PDFs
- Handwritten documents
- Free-form text
- Synchronous processes
See the docs for the full list of limits.
Can the extracted data be validated by the submitter, for example on a portal?
Not at this time. The review is done by an agent (with a reviewer role) on the Document Intelligence Workspace.
Can I extract handwritten documents?
Although handwriting is not officially supported, the extraction can be accurate in cases where the text is legible. We strongly recommend additional validation for handwritten documents to assess the results.
Can I extract a whole paragraph or free-form text?
No, this is not currently supported.
Is it supported in Self-Hosted Environments?
Document Intelligence is now available for self-hosted customers, see this KB.
Is it supported in ServiceNow Regulated Environments?
Yes; refer to that KB article on Now Support.
Is it supported on my Personal Developer Instance (PDI)?
No.
If I upload a document with multi pages but only need the data on some of the pages, can Document Intelligence split the document or allow me to select which pages to process?
Yes, the page range filter demonstrated during the AI Academy "What's new in Washington DC for Document Intelligence", allows to limit the pages from which data extraction is executed.
Implementation & Troubleshooting
I created the Extract Values Flow, but I noticed the values are not written in my target record after the extraction; what’s going on?
First, make sure the values are populated in the Field Values / Extracted Value table [(sys_)di_extracted_value]. Then, ensure the Extract Value Flow is running correctly and the Flow properties are set to run as System User; make sure you see the data in the Flow Execution logs. Finally, check the Table configuration:
- Review ACLs on the Table
- Application access on the Table
- Check the cross-scope privileges
- Make sure the target field type matches the type of data
For a live demo of these steps, refer to this AI Academy session.
How can I monitor ML Solutions to track the processing of a Document Intelligence task?
Type ml_solution.list in the navigation menu.
For a live demo of this step, refer to this AI Academy session.
Can I use a Key Group to extract values to a different table even though the data is not a table?
No, this is a bad idea, leading to lousy prediction performance and bad User Experience. The Key Groups are meant for extracting data displayed in a table or a list with a set of common fields or columns.
My PDFs are not being read/displayed by DocIntel, what’s going on?
There are a few reasons a PDF might not be processed properly:
- The PDF is unaccessible – check that the PDF is not encrypted, secured or password-protected
- The ACLs on the sys_attachment table prevent the creation of the processed record
- The "glide.attachment.extensions" system property doesn’t have "jpeg" listed
My Document task shows “Failed to process the task” how can I troubleshoot?
Type ml_solution.list in the navigation menu and identify the failed solution, open it and open the “Show training progress” related link to find more information. The reason why a solution fails can be:
- The instance doesn’t have access to the ML infrastructure, for example, a Personal Developer Instance (PDI) or a CloudLab instance
- An issue with the system property [glide.shared_service_scheduler.url], which should be set to https://sncmlscheduler.service-now.com/
- The “sharedservice.worker” user doesn’t exist or is not Active.
For a live demo of these steps, refer to this AI Academy session.
What's the best way to migrate and import/export task definition along with the trained models from one instance to another?
The easiest way is to open the use case on the Admin experience and select 'Add to update set'. This is available in Document Intelligence Admin v2, which requires Vancouver.
If you are using a previous version, follow the steps in this article.
I created Document tasks for my Use Case but they are never processed, what’s happening?
Make sure your Use Case record has a reference for each of the 4 ML Solution Definition fields. One common issue is Use Cases that were exported via Update Sets but the referenced ML Solution Definition records were missed.
For a live demo of this step, refer to this AI Academy session.
What are the changes in the table schema from v3 (Vancouver)?
In v3, the base tables have been migrated to a new table schema:
The 'Use cases' table is sys_di_task_definition (previously di_task_definition)
The 'Document tasks' table is sys_di_task (previously di_task)
The 'Field values' table is sys_di_extracted_value (previously di_extracted_value)
Note, that it can also mean they don't appear in some menus (like the Flow Designer "Table" selector), for this, you might need to update some system properties.
What are the changes in terminology from v2.3?
How can I monitor usage and metrics for Document Intelligence?
To access the Document Intelligence Usage dashboard, navigate to All > Document Intelligence > Document Data Extraction Administration > Monitoring. Learn more in Product Docs .
What are the stages of the processing for a Document Task [(sys_)di_task]?
More info in Get Started with Document Intelligence [How-To].
Where are the Extracted Values/Field Values stored?
The extracted values are stored in the Field Value / Extracted Value table [(sys_)di_extracted_value]. To access them, navigate to Document Intelligence > Extracted Value and filter with the Task.
More info in Get Started with Document Intelligence [How-To].
Hi @Loic1 could you please help me in solving this error this is the error i found while trouble shooting from show training progress Thanks in advance.
I am unable to select the draw tool to help identify my table data any suggestions?
Hi @Loic1,
I created a subflow, where I copied the DocIntel process flow Integration Actions. And I have given input to subflow as sys_id of a table record. And its working fine as same as Docintel OOO integration step(Process flow integration).
But, when I am trying to add file attachments string to input (trying to give selected attachments of a record as input to subflow instead of all the attached files of a record) , then the Action step is not taking those array.string of file attachments in type pill. Please find the screenshots below and guide me if anything needs to be added.
After saving of the sub flow, I am not able to see the file array.string data pill in Attachment IDs. And even test also failed as the Attachment IDs is not taking my input sys_ids of the attachments.
Please check and guide me.
