You'll learn how to design IA solutions using Work Queues, Session Variables, and more, and understand the criteria for evaluating ML for automation to create proper solution design. Once you’ve learned how to create reusable IA templates from best practices, you’ll see how they reduce the time needed to bring a solution into production. The book then takes you through the BP Control Room and management aspects of an IA solution, introducing you to the unique management concerns IA presents compared to RPA, due to the uncertainty introduced by model predictions and a constantly evolving regulatory environment that restricts how IA can be used. The book highlights IA’s impact on the wider automation context through user permissions, security, deployments, and BP's Robotic Operating Model, and concludes by recreating a real-world intelligent automation processes in BP. This book is not just practical; it’s also enriched by real-life experience and the insights distilled from the authors’ research at MIT, which examined over 70 IA use cases.
Author(s): James Man
Publisher: Packt Publishing Pvt Ltd
Year: 2023
Language: English
Pages: 994
Intelligent Automation with Blue Prism
Contributors
About the author
About the reviewer
Preface
Who this book is for
What this book covers
To get the most out of this book
Download the example code files
Conventions used
Get in touch
Share Your Thoughts
Download a free PDF copy of this book
Part 1:Connecting Blue Prism to ML Models
1
Machine Learning as a Service: the Digital Exchange and Web APIs
Technical requirements
Using the DX
Accessing the DX
Machine learning web API fundamentals
An overview of MLaaS on the DX
Vendor selection
Examples
Example 1 – AWS Comprehend for text entity extraction, key phrase extraction, and sentiment analysis
Example 2 – Azure Form Recognizer for invoice extraction
Example 3 – GCP Cloud Vision batch OCR processing
Summary
2
Predicting from Command Prompt and PowerShell
Technical requirements
Command-line basics
Output streams
Output redirection
Blocking versus non-blocking execution
Predicting from the command line using Utility – Environment
Example 1 – Running a program using the Start Process
Example 2 - Running a program using Run Process Until Ended
Getting prediction results back into BP
Example 3 – Saving STDOUT and STDERR as Data Items
Example 4 – Redirecting an internal command (dir) to files
Timing out long-running predictions
Example 5 – PS script timeout
DX VBOs – Utility - PowerShell and Script Execution VBO
Utility – PowerShell
Script Execution VBO
Example 7 – Calling a Python program
Summary
3
Code Stages
Technical requirements
Setting up ML.NET in BP
Adding references and namespaces to BP
Example 1 – preparation work before BP
Porting ML.NET C# into a Code Stage
Global Code
Example 2 - porting the source code into BP
Improving BP integration
Example 3 - refactoring
Summary
Part 2:Designing IA Solutions
4
Reviewing Predictions and Human in the Loop
Technical requirements
Why should we review predictions?
Reduce business risk
Stay ahead of regulatory concerns
What does HITL mean in the context of IA?
What criteria can be used to trigger human intervention?
Random sampling
Thresholding
How can we share prediction data between prediction reviewers and BP?
Reviewing predictions through shared folders
Summary
5
IA Process and Work Queue Designs for HITL
Technical requirements
Single-Process, single-Work Queue designs
Asynchronous (non-blocking) reviews
Synchronous (blocking) reviews
Multiple-Process, single-Work Queue designs
Independent manual review logic
Multiple-process, multiple-work queue designs
Fully independent manual reviews
Separating ML predictions and manual reviews into their own Processes and Work Queues
Design comparison
Design 1 – asynchronous reviews (one Process, one Work Queue)
Design 2 – synchronous (polling) reviews (one Process, one Work Queue)
Design 3 – independent HITL review logic (two Processes, one Work Queue)
Design 4 – fully independent HITL reviews (two Processes, two Work Queues)
Design 5 – full separation (three Processes, three Work Queues)
Summary
6
Reusable IA Components
Technical requirements
IA session control
Forcing HITL review
Disabling HITL review
Forcing review data recreation
Example 1 – three IA Session Variables
ML prediction kill switch
Example 2 – kill switch
ML model versioning
Two different ways of calling web APIs
Calling a web API using an Object when a new endpoint is provided
Calling a web API using an Object when the vendor reuses an existing endpoint
Example 3 – versioning ML endpoints manually
Calling Web API Services
New ML model evaluation
Example 4 – new ML model evaluation Process template
Reusable IA components review
Summary
7
IA Templates and Utility – IA Object
Technical requirements
Object – Utility – IA
Random Integer in Range
Random Decimal in Range
Run Process Read Stdout Stderr with Timeout
File to Base64
Threshold Excel to Collection
Get Threshold by Label
Object Overview
Process templates
Single-Process, single-Work Queue, synchronous review Process template
Single-Process, single-Work Queue, asynchronous review Process template
Three-Process, three-Work Queue, asynchronous review Process template
Summary
Part 3:Control Room and Management
8
The LAM, User Roles, and MTE
Technical requirements
IA User Roles and Permissions
ML Auditor
ML Deployer
ML Reviewer
A User Role comparison
MTEs
MTE for the ML Auditor and ML Reviewer User Roles
MTE limitations
An updated LAM template
Summary
9
ML Deployments and Database Operations
ML deployments and rollbacks
Web API deployment strategies
Script deployment strategies
Code Stage deployment strategies
Database operations
Table growth maintenance
Extracting ML prediction data from the database
Exporting reviewed prediction data from the database
Summary
10
IA’s Impact on the Robotic Operating Model
Strategy
Future of Work Vision
Business case and value
Governance, Risk, and Controls
Workforce
Building your organizational model
Adopting new ways of thinking and working
Roles and career paths
Design
Assessment and Prioritization
Requirements Design
Development
Methodology and Teamwork
Delivery Controls
Testing and Quality Assurance
Operations
Deploy and Release
Support model
Summary
Part 4:Real-Life Scenarios and Other Blue Prism Products
11
Processing Refunds
Technical requirements
ML model background information
EC model
Entity recognition model
Generative AI model
ML model summary
Solution design
Email classification model
Entity recognition model
Generative AI model
Solution design diagram
Implementation
Example 1 – Creating the solution structure from IA templates
Example 2 – Implementing the IA details in Process 1
Summary
12
Power Service Interruptions
Technical requirements
ML model background information
Outage prediction model
Customer complaints model
ML model summary
Solution design
Handling model deployments
Example 1 – Outage prediction model deployment
Example 2 – Customer complaint model deployment
Example 3 – Rollback customer complaint model deployment
Exporting data for audit
Example 4 – Exporting OP model data through SQL
Example 5 – Exporting customer complaint model data through SQL
Summary
13
Other Intelligent Blue Prism Products and Future IA Trends
Decipher IDP
How is Decipher related to IA?
Using Decipher
Next steps
Document Automation
How is Document Automation related to IA?
Using Document Automation
Next steps
Decision
How is Decision related to IA?
Using Decision
Next steps
Interact
How is Interact related to IA?
Using Interact
Next steps
Future IA trends
Improved AI product integration
Democratized ML using LLMs
AI ethics and safety
Summary
Appendix
IA Risk Management
Socio-organizational IA risks
Operational IA risks
IA risk mitigation measures
Index
Why subscribe?
Other Books You May Enjoy
Packt is searching for authors like you
Share Your Thoughts
Download a free PDF copy of this book
