Automating document processing with Pega GenAI
Pega GenAI™ helps you automate Case processing by analyzing documents and extracting data from them. Your application captures information from attachments, so users enter less data by hand and make fewer errors. Users can then focus on work that needs human judgment.
For example, in a claims intake process, customers upload invoices, medical bills, and handwritten accident statements. Pega GenAI reads these documents and fills in the claim details, so an adjuster starts work with the data already in place.
Pega Platform™ offers two ways to process documents with Pega GenAI:
- DocAI Connect Rules, which process attachments as part of a Case Lifecycle
- Agents, which analyze documents that users attach during a conversation
As a Lead System Architect (LSA), you decide which approach fits each business process and how much of the work the application completes without manual review.
DocAI Connect automation in a Case Type
A DocAI Connect Rule extracts field values from documents that users upload during a Case Lifecycle. Automated extraction reduces manual data entry and helps users resolve Cases faster. Use this approach when the Case needs specific data that users provide by attaching a document. To use a DocAI Connect Rule in a Case Type, add a DocAI Connect automation Step to the Case Lifecycle.
Note: To use DocAI Connect Rules in your application, turn on the EnableDocumentAgent release toggle.
A DocAI Connect Rule can use optical character recognition (OCR) to read a document and return a confidence score for each extracted value. When every field has a high confidence score, the Case continues through straight-through processing. When one or more fields have a low confidence score, the Case routes the document to a person for review.
The following figure shows how DocAI Connect uses confidence scores to determine whether a Case continues through straight-through processing or requires manual review:
A DocAI Connect Rule combines OCR with large language models (LLMs). When you turn on OCR preprocessing and confidence scores, the Rule processes each document in four steps:
- A user uploads a document to a Case or in a conversation with an Agent.
- The Smart Chunking Service reads the text by using Amazon Textract OCR and returns the text line by line.
- An extraction LLM receives the list of fields to extract, the OCR text, and the original document. The LLM returns a value for each field.
- An assessment LLM checks each value against the document and returns a confidence score.
Note: The Smart Chunking Service runs on Pega Cloud®, outside the Pega GenAI Gateway. OCR preprocessing and confidence scores work only on Pega Cloud. On-premises systems support only basic extraction.
For more information about DocAI Connect Rule configuration, see Creating a DocAI Connect Rule.
Document processing in an Agent conversation
When your application includes a conversational Agent, users can attach a document while they chat with the Agent. The Agent analyzes the document and answers questions about it. If the content matches the Case Data Model, the Agent also fills in related fields. For example, a user can ask the Agent to find unusual amounts in a long expense report.
Agents support document attachments by default without any additional configuration. Use this approach for one-off analysis outside the Case Lifecycle. This option works only in Pega Constellation™ applications. For more information about document processing with Agents, see Document processing by Agents.
Note: The supported file types depend on the LLM that the Agent uses. For more information, see Model limitation for attachment support by Agents.
Check your knowledge with the following interaction:
This Topic is available in the following Module:
If you are having problems with your training, please review the Pega Academy Support FAQs.
Want to help us improve this content?