Document

Understanding
Process
Studio Template
Revision History ..................................................................................................................5

Release Notes: ....................................................................................................................6

24.4-preview: ..................................................................................................................6

23.10: ..............................................................................................................................6

22.10: ..............................................................................................................................6

22.6-preview: ..................................................................................................................7

22.4: ................................................................................................................................7

22.2-preview: ..................................................................................................................7

21.10: ..............................................................................................................................8

Overview.............................................................................................................................9

Key Features ...................................................................................................................9

Generic Document Processing Flow ................................................................................ 10

Solution Architecture ...................................................................................................... 12

Document Understanding, Part of End-to-End Business Processes ........................... 12

The One-Job-Per-File Approach................................................................................... 13

Document Understanding and Queues ....................................................................... 13

Orchestrator v20.10.8 or Newer ............................................................................. 13

Orchestrator Versions Prior to 20.10.8 ................................................................... 14

Starting Document Understanding Jobs ...................................................................... 14

Studio Project Overview .................................................................................................. 15

Settings for Attended Processes.................................................................................. 15

Settings for Unattended Processes ............................................................................. 16

Example Implementation ............................................................................................ 17

Project Files.................................................................................................................. 17

📄 Data\Config.xlsx ................................................................................................. 17

📄 Main-ActionCenter.xaml.................................................................................... 17

📄 Main-Attended.xaml .......................................................................................... 18
 📄 Framework\00_ReadConfigFile.xaml ................................................................ 18

📄 Framework\10_InitializeProcess.xaml............................................................... 18

📄 Framework\15_GetTransactionItem.xaml ........................................................ 18

📄 Framework\20_Digitize.xaml............................................................................. 19

📄 Framework\30_Classify.xaml ............................................................................ 19

📄 Framework\35_ClassificationBussinessRuleValidation.xaml ............................ 19

📄 Framework\40_TrainClassifiers.xaml ................................................................ 19

📄 Framework\50_Extract.xaml ............................................................................. 19

📄 Framework\55_ExtractionBussinessRuleValidation.xaml ................................. 20

📄 Framework\60_TrainExtractors.xaml ................................................................ 20

📄 Framework\70_Export.xaml .............................................................................. 21

📄 Framework\80_EndProcess.xaml ...................................................................... 21

📄 Framework\ReusableWorkflows\InvoicePostProcessing.xaml ......................... 21

📄 Framework\ReusableWorkflows\AutocorrectOcrMistakes.xaml ..................... 22

📄 Framework\ERR_AbortProcess.xaml................................................................. 22

📄 Framework\ReusableWorkflows\GetWritePermission.xaml ............................ 23

📄 Framework\ReusableWorkflows\GiveUpWritePermission.xaml ...................... 23

📄 Framework\ReusableWorkflows\LockFile.xaml ................................................ 23

📄 Framework\ReusableWorkflows\UnlockFile.xaml ............................................ 23

📄 Framework\ReusableWorkflows\SetTransactionProgress.xaml....................... 23

📄 Framework\ReusableWorkflows\SetTransactionStatus.xaml .......................... 23

Quick Start Guide ............................................................................................................ 24

🔧Orchestrator Configuration .................................................................................... 24

🔧 Attended Automation ........................................................................................... 24

🔧 Unattended Automation ....................................................................................... 25

🔧 Dispatcher Mechanisms for Unattended Implementations ................................. 25

Known Issues and Limitations ......................................................................................... 26
Revision History
Date Changes
June-2021 1.0
Oct-2021 2021.10
Feb-2022 22.2-preview
May-2022 22.4
June-2022 22.6-preview
Oct-2022 22.10
November-2023 23.10
April-2024 24.4-preview
Release Notes:
24.4-preview:
- fixed bug with AutoCorrectOcrMistakes Test

- Updated activities to latest preview versions.

23.10:
New features:

• AutocorrectOcrMistakes (See dedicated chapter):

o Added AutocorrectOcrMistakes under the reusable components folder.
o Added AutocorrectOcrMistakes sheet in the Config file.
• Orchestrator Configuration Quick Start (See dedicated chapter) – workflow for
quickly setting up Storage Buckets, Queues, Assets and/or Task Catalogs.
Changes:

• Updated all dependencies to the latest stable versions.

22.10:
New features:

• Testing:
o Added 19 tests under the test folder that test basic DU and framework
functionality.
o Added BatchTestingUtilities under the test folder. These help the tests to
run faster, but also during the development to save data to cache, and
not run the whole process again.
o Added a Cache folder based on the sample file we have.
• Main Logic:
o Added a flow decision that would make the workflow end successfully if
no more transaction items are present.
o Automatic detection of decimal points in Invoice Post Processing
o Added Receipt Post Processing
• Other:
o Validated new Out of the Box Insight dashboards.

Changes:

• Updated all dependencies to the latest stable version.

• Cleaned up the Config File and added the Receipt Post Processing part.
• Hidden the option of C# for now.
Bugfixes:

• “For each” activity in “Parallel for each” issue should be solved.

22.6-preview:
New features:

• Added C# support.
• Introduced Asset naming convention for better visibility of Document
Understanding Assets in Orchestrator
• Added new overriding Assets for Orchestrator Configuration.

Changes:

• Updated all dependencies to the latest stable version.

Bugfixes:

• Fixed error in reading of the Config file in Main-Attended.xaml.

• Fixed the Log Message from the Classify workflow in C#. It now displays the
correct classification results.

22.4:
New features:

• Added .NET 5.0 support.

Changes:

• Removed Windows – Legacy support.

• Updated all dependencies to the latest stable versions.
• Rescoped the extractionResults variable in all Main workflows to prevent misuse

Bugfixes:

• Corrected small typos.

22.2-preview:
Changes:

• Updated dependencies for Document Understanding activities to 22.2-preview

release.
 • Framework\ReusableWorkflows\SetTransactionProgress.xaml - Retry Scope
within was updated with ContinueOnError = True
• Framework\40_TrainClassifiers.xaml – Invoke UnlockFile activity within was
updated with ContinueOnError = True

Bugfixes:

• Fixed the page range logging on document exceptions in Main-Attended.xaml

21.10:
New features:

• New workflows for advanced Business Rule validation for Classification and
Extraction
• New workflow for post processing Invoices based on the OOTB (out of the box)
model.
• Validation Station/Action now only displays the relevant pages instead of the
whole file. Removed the PDF splitting functionality.
• Support for ML Extractor Trainer to use the auto-retrain feature in AI Center
• New assets / settings in the config file for skipping classification/extraction
training.
• New assets / settings in the config file to force documents through manual
validation (useful in development/UAT especially)
• New exception type: DocumentRejectedByUserException. This exception is
thrown when a user explicitly rejects a document during validation. When setting
the transaction status, a DocumentRejectedByUserException is treated the same
way as a BusinessRuleException

Changes:

• Exception handling workflows now receive the Exception as an argument instead

of an exception message.
• Improved logic for loading Orchestrator assets
• Improved exception messages
• Improved comments and annotations
• Improved logging
Overview
As the number of Document Understanding engagements keeps increasing in
complexity, the need for a common implementation/RPA approach becomes evident.
Document Understanding processes have a particular logical flow and requirements that
are better suited for a dedicated approach of their own, instead of relying on the existing
RE-Framework.

Key Features
• Seamless getting started process with new Document Understanding projects.
• Suitable for all use cases: from quick demos to scalable implementation projects
• Production-ready. Implements built-in logging, exception handling, and retry
mechanisms.
• Common architecture for both Attended and Unattended (plus Action Center)
implementations. Simple to switch between solutions.
• 2021.10: Ability to post-process Invoices to minimize the number of hits to
Action Center
• Designed to make development, testing, deployment, debugging, and scaling
easy.
• Follows the best practices pertaining to RPA, Document Understanding,
Orchestration Processes, and Long-running workflows.
Generic Document Processing Flow

Keep in mind that the diagram above shows that the most detailed logical flow split into the smallest possible modules. In practice, it is to
be expected that some parts could be merged or might be completely excluded, as they are not required in a particular implementation.
🚩 Pre-Digitization Processing

Any processing that needs to take place before digitizing the document. E.g., applying
grayscale or a skew correction.

🚩 Digitize Document

A digital version of the document is obtained. In case the document is scanned, OCR
(Optical Character Recognition) is required.

🚩 Classify Document

The document is classified.

🚩 Classification Successful?

The logic required to determine whether the classification was successful or not. Based
on this decision, classification validation might be needed.
Note: Deciding whether human validation is needed or not is a business decision.

🚩 Human Classification / Validation

If the automatic classification was unsuccessful (low confidence, business rules are not
met, etc.), send the document to a human operator for manual classification/validation
or rejection.

🚩 Classifier Feedback (Learning)

Integration of the learning mechanisms, if any, for classifier(s). It is recommended to

only train with human-validated data, but not mandatory.

🚩 Pre-Extraction Processing

Any processing that needs to take place before the actual data extraction begins.

🚩 Data Extraction

Extract the data using the appropriate extractor(s).

🚩 Human Validation?

The logic required to determine whether the human validation of the extracted data
(validating the extracted data against pre-defined business rules, checking confidence
levels, OCR confidence, missing data, etc.) is needed or not.

Note: Deciding whether human validation is needed or not is a business decision.

🚩 Human Data Validation / Extraction

Human-in-the-loop step where the human operator can correct automatically extracted
data or manually extract the missing data.

🚩 Extractor Feedback (Learning)

Integration of the learning mechanisms, if any, of the used extractor(s). It is

recommended to only train with human-validated data, but it is not mandatory.

🚩 Post-Extraction Processing

Any processing that should be done to the extracted data before exporting it. E.g.,
formatting the data in a specific manner.

🚩 Data Export

Exporting the data to a persistent location so that it can be used by other processes or
by business users. Common examples: exporting to the Data Service, updating a
Database, or writing the results to an Excel, CSV, or JSON file.

🚩 Post-Export Processing

Any processing that should be done after the data is exported. E.g., merging data
extracted from multiple documents into a single result or sending a notification email.

Solution Architecture

Document Understanding, Part of End-to-End Business Processes

Document Understanding processes are not standalone. They are part of bigger business
processes to be automated.

Benefits:

• Prevents external issues from impacting Document Understanding processes

and from causing unneeded consumption of license pages through re-execution.
• Better overview of workload and robot utilization.
• Easier to scale.

The architecture for an end-to-end Business Process involving Document Understanding

consists of:
1. Upstream automation which handles all the business logic that should be executed
prior. It consists of one or several RPA processes and acts as a Dispatcher for
Document Understanding jobs.
2. Document Understanding process - acts as a Performer for "Upstream" and as a
Dispatcher for "Downstream" jobs.
3. Downstream automation which handles all the business logic that makes use of the
extracted data. It consists of one or several RPA processes and acts as a Performer
for Document Understanding jobs.

The One-Job-Per-File Approach

Document Understanding processes will not run as batch jobs. Instead, an individual job
starts for each file to be processed. This approach is used for both Attended and
Unattended implementations.

Document Understanding and Queues

Orchestrator v20.10.8 or Newer
Orchestrator version 20.10.8 introduced queue support for Persistence activities
enabling queues to be used.

Note: To avoid Document Understanding license consumption through re-executions,

the Auto Retry functionality of the queue should be disabled.
Orchestrator Versions Prior to 20.10.8
Queues should not be used for Document Understanding processes, as they are not
supported. Items waiting for Human Validation for more than 24h will be marked as
Abandoned.

Starting Document Understanding Jobs

A dedicated Dispatcher mechanism is required to start Document Understanding jobs in
Unattended scenarios. There are two recommended approaches:

1. A Queue Trigger, tasked to start a new job for every new Queue item, using the
settings stated in the image below

Note: Due to the current Job Count Strategy for triggers, some queue items experience
delays before a processing job is triggered. This behavior is on the roadmap to be
addressed.
2. Having the Dispatcher process use a Start Job activity

Note: Currently, only the Start Job And Get Reference activity has support for passing
job arguments. This activity is part of the UiPath.Persistence.Activities package.

Studio Project Overview

The Document Understanding Process is available as a UiPath Studio project template
and the projects created using it automatically include all files.

Settings for Attended Processes

Main workflow for Attended Processes

Right-click on Main-Attended.xaml and set it as the Main workflow for the project.

Disable Background Running & Persistence Support

Open the Project Settings and make sure that Starts in Background and Supports
Persistence are set to No.
Settings for Unattended Processes
Main workflow for Unattended Processes

Right-click on Main-ActionCenter.xaml and set it as the Main workflow for the project.

Enable Persistence Support

Open the Project Settings and make sure that Supports Persistence is set to Yes.
Example Implementation
To facilitate understanding the Document Understanding Process template, it comes
with a pre-implemented example. It showcases the processing of 4 types of documents.

Workflow-specific details can be found in the next chapter.

Project Files

📄 Data\Config.xlsx
Configuration file for project settings. Minimum configuration required:

• Configure your Document Understanding Api Key under an Orchestrator Asset

named DocumentUnderstandingApiKey.
• If using Action Center, configure the StorageBucketName under the Settings
sheet.
• Configure the DocumentUnderstandingQueueName if using queues.

📄 Main-ActionCenter.xaml
Workflow to be set and used as Main for attended Document Understanding processes
that use Action Center for Human-in-the-Loop.
Arguments:

• in_TargetFile (default Nothing): specifies what file should be processed.

• in_UseQueue (default True): specifies whether Orchestrator queues are used. If
set to True, the value of the in_TargetFile argument is ignored and the file to be
processed is fetched from the Transaction Item.

📄 Main-Attended.xaml
Workflow to be set and used as Main for attended Document Understanding processes.

Arguments:

• in_TargetFile (default Nothing): specifies what file should be processed.

• in_UseQueue (default False): specifies whether Orchestrator queues are used. If
set to True, the value of the in_TargetFile argument is ignored and the file to be
processed is fetched from the Transaction Item.

📄 Framework\00_ReadConfigFile.xaml
Reads the contents of the Config file into a Config dictionary at runtime.

Note: Does not load Orchestrator assets!

No custom code was added here for the purpose of creating the example
implementation.

📄 Framework\10_InitializeProcess.xaml
Workflow that loads the Taxonomy and Orchestrator assets. Any process-specific
initialization code belongs here.

No custom code was added here for the purpose of creating the example
implementation.

📄 Framework\15_GetTransactionItem.xaml
Gets the next Transaction Item when using Orchestrator queues. The target file to be
processed is expected to be found under the TargetFile key of the Transaction Item’s
SpecificContent.

Also loads all the Transaction Item's SpecificContent into the Config dictionary for ease
of use.

No custom code was added here for the purpose of creating the example
implementation.
📄 Framework\20_Digitize.xaml
Workflow for Pre-Digitization and Digitization logic.

The example implementation uses the UiPath Document OCR engine.

📄 Framework\30_Classify.xaml
Workflow for Classification.

The example implementation uses the Intelligent Keyword Classifier for all document
types. Please know that using the LearningData string is also possible.

📄 Framework\35_ClassificationBussinessRuleValidation.xaml

2021.10: Workflow for checking Classification Success. Custom logic is required to

determine classification success based on process-specific business rules. A flag can be
set to force all documents through manual validation.

By default, all results are sent to human validation.

📄 Framework\40_TrainClassifiers.xaml
Workflow for Classifier training.

The example implementation uses the Intelligent Keyword Classifier Trainer for all
document types. It also uses the LockFile/UnlockFile reusable workflows in order to
regulate access when updating the learning file. Please know that using the
LearningData string is also possible.

📄 Framework\50_Extract.xaml
Workflow for Pre-Extraction Processing, Data Extraction.

The example implementation uses 4 extractors:

• Regex Based Extractor for fixed-form Certificates

• Intelligent Form Extractor for fixed-form W9 Forms
• Machine Learning Extractor for Receipts – using “Receipts” Framework Alias for
ML Extractor training
• Machine Learning Extractor for Invoices – using “Invoices” Framework Alias for
ML Extractor training
📄 Framework\55_ExtractionBussinessRuleValidation.xaml
2021.10: Workflow for checking Extraction Success. Custom logic is required to
determine extraction success based on process-specific business rules. A flag can be set
to force all documents through manual validation.

By default, all results are sent to human validation, except for invoices that go through
the Invoice Post Processing workflow.

📄 Framework\60_TrainExtractors.xaml
Workflow for Extractor training.

The example implementation uses 2 extractor trainers:

• Machine Learning Extractor Trainer for Receipts – using “Receipts” Framework

Alias for ML Extractor training
• Machine Learning Extractor Trainer for Invoices – using “Invoices” Framework
Alias for ML Extractor training
📄 Framework\70_Export.xaml
Workflow for Post-Extraction Processing and Data Export logic.

The example implementation uses a simple mechanic of writing the extracted data to
XLSX files. This is only one of many export possibilities besides using Data Service, a
Database, using CSV, JSON format, etc.

📄 Framework\80_EndProcess.xaml
Workflow for Post-Export Processing and Process Cleanup logic.

No custom code was added here for the purpose of creating the example
implementation.

📄 Framework\ReusableWorkflows\InvoicePostProcessing.xaml
2021.10: This workflow implements invoice post-processing using the default OOTB
model. The ruleset to check against is defined in a dedicated sheet of the Config file.

The ruleset and the logic should be updated as required by the business process.

Default post-processing rules & steps:

• Verify that all Mandatory Fields and Columns are extracted

• Verify that all table rows match the rule: Quantity * Unit Price = Line Amount
• Verify that the sum of all Line Amounts = Net Amount
• Sum up Net Amount with all the values defined as Config "SubTotalAdditions".
Verify that the sum of Net Amount + "SubTotalAdditions" = Total
• Verify the extraction confidence of all defined "ConfidenceFields" against their
individual confidence thresholds
 • Verify the extraction confidence of all the other fields against the "other-
Confidence" threshold

This should NOT be used as-is, except for demo purposes. For a real implementation,
the post-processing & validation should be tailored to the specifics of the business
process.

The default implementation uses EN-US culture information. This means '.' is the
decimal separator and "," is the thousand separator. (e.g.: 10,000.00)

📄 Framework\ReusableWorkflows\AutocorrectOcrMistakes.xaml
This workflow implements OCR auto correction by saving corrections made in Human in
the Loop. This feature is meant to be used for fields where the extracted value is
expected to be part of a limited set of options that do not have a high level of similarity
between them.

Examples of fields that are suitable for the OCR auto correction – Logo, Company Name,
Document Issuer Name.

Examples of fields that are not suitable for the OCR auto correction – Amount, Date, PO
Number, First Name, Last Name.

The workflow is not enabled by default and the following settings found in the
“AutocorrectOcrMistakes” sheet in the config file need to be customized in order to use
it:

• AutocorrectionEnabled – boolean flag that enables/disables the flow

• AutocorrectionFieldName – the name of the field to be corrected

As this feature uses changes done in Human in the Loop, adding blacklisted OCR values
or performing maintenance may be required. This can be done by modifying the learning
file saved in the DU storage bucket (check StorageBucketName and
AutocorrectStorageBucketDirectoryPath in the config file).

📄 Framework\ERR_HandleDocumentError.xaml

Workflow that is executed when an exception occurs when processing a classified

document.

No custom code was added here for the purpose of creating the example
implementation.

📄 Framework\ERR_AbortProcess.xaml
Workflow that is executed if the process is aborted due to a terminating exception. Code
for error cleanup or for sending error notifications belongs here.
No custom code was added here for the purpose of creating the example
implementation.

📄 Framework\ReusableWorkflows\GetWritePermission.xaml
Helper workflow using a Queue with a single queue item as a semaphore mechanism to
regulate write access to classifier training files.

📄 Framework\ReusableWorkflows\GiveUpWritePermission.xaml
Helper workflow using a Queue with a single queue item as a semaphore mechanism to
regulate write access to classifier training files.

📄 Framework\ReusableWorkflows\LockFile.xaml
Helper workflow using simple lock/unlock mechanism to regulate write access to
classifier training files.

📄 Framework\ReusableWorkflows\UnlockFile.xaml
Helper workflow using simple lock/unlock mechanism to regulate write access to
classifier training files.

📄 Framework\ReusableWorkflows\SetTransactionProgress.xaml
Updates the TransactionProgress for the Transaction Item that is being processed (if
using queues).

📄 Framework\ReusableWorkflows\SetTransactionStatus.xaml
Sets and log the transaction's status. The approach is like the one used by the RE-
Framework.

📄 QuickStart\IntializeOrchestrator\IntializeOrchestrator.xaml

Workflow that can be used for quickly setting up the necessary Orchestrator resources
(Assets, Queues, Storage Buckets, Task Catalogs).
Quick Start Guide
Please see Document Understanding and Queues for the required Orchestrator version
of using queues.

🔧Orchestrator Configuration
• Configure your Document Understanding Api Key under an Orchestrator Asset
named DocumentUnderstandingApiKey.
• If using Action Center, create a Storage Bucket for your process.
• If needed, create a Queue for your process.

Fast batch creation of items in Orchestrator can be done by running the workflow
IntializeOrchestrator.xml placed in QuickStart\IntializeOrchestrator.
The process is using the IntializeOrchestratorConfig.xslx part by loading an excel
document in the assigned section. This can be found in
QuickStart\IntializeOrchestrator\Data.
The Process can create the following 4 item types in orchestrator:
1. Storage Bucket
2. Queue
3. Asset
4. Task Catalog
The parameters to be filled in are the following:
1. Name: Displayed name on Orchestrator
2. Value: Value that the specific item will have, if needed
3. Type: One of four listed above
4. Folder: Folder location where the item must be create in Orchestrator
Description: item description

🔧 Attended Automation
• Configure your Document Understanding Api Key under an Orchestrator Asset
named DocumentUnderstandingApiKey.
• If using queues, configure the DocumentUnderstandingQueueName.
• Configure the Settings for Attended Processes.

Note: We recommend becoming familiar with the solution and the project before
removing the example implementation and the taxonomy.
🔧 Unattended Automation
• Configure your Document Understanding Api Key under an Orchestrator Asset
named DocumentUnderstandingApiKey.
• If using Action Center, configure the StorageBucketName under the Settings
sheet,
• If using queues, configure the DocumentUnderstandingQueueName.
• Configure the Settings for Unattended Processes.

Note: When developing, it is quite simple to test your implementation in Attended

mode. Simply run Main – Attended.xaml in order to use the local validation activities
instead of going to Action Center (it is not needed to change process settings while
testing this way).
Note: We recommend becoming familiar with the solution and the project before
removing the example implementation and the taxonomy.

🔧 Dispatcher Mechanisms for Unattended Implementations

Please see Starting Document Understanding Jobs.
Known Issues and Limitations
• When using the default LockFile/UnlockFile implementation, there is a small
chance that some training data will occasionally be lost – when multiple bots
update the training file at the exact same time, the last one will overwrite others’
training data. In practice, this does not pose any problems.
• The Document Understanding Process is currently not performing any cleanup
of its Temp folder (where it stores split or downloaded files).
• By design, the job ends successfully even if the processing of all classified
documents ends with exceptions.
• If you are updating to IntelligentOCR 6.5.0 from an older version, you will have
to add to the workflow SetTransactionStatus a new namespace manually:
“UiPath.IntelligentOCR.Exceptions”. 6.5.0 Introduced this breaking change but
creating a new template of 22.10 or later would have this solved.