Finance Operations Solution Architect & Developer
Intelligent Document Processing: Invoices
Document AI Extraction With GL Classification And Human Confirmation
Built an end-to-end invoice intake pipeline on Salesforce that uploads PDFs, extracts fields via Data 360 Document AI, normalizes and GL-classifies line items with an Agentforce prompt template, and presents editable results in a custom LWC for AP review.
Document AI Screen Flow Agentforce Prompt Templates Apex LWC Named Credentials
Sequence Diagram
Challenge
Accounts payable teams retyped invoice headers, vendor details, and line items from PDFs into Salesforce, a slow, error-prone process with no consistent GL coding and little auditability when extraction or classification failed mid-flow.
Key Outcomes
- Reduced manual invoice data entry for AP reviewers
- Standardized line-item GL classification against a shared expense chart
- Added human-in-the-loop confirmation before accepting extracted invoice data
- Surfaced extraction and normalization failures with explicit Flow error screens
View Approach Details
- Designed a Screen Flow on an Agentforce Studio flexipage for PDF upload, extraction, normalization, parse, and confirmation with dedicated error screens on failure.
- Implemented Apex invocable callouts to the Data 360 Document AI extract-data API using Named Credentials and an Invoice_Processing IDP configuration.
- Used an Agentforce prompt template to unwrap typed Document AI JSON into flat invoice schema and classify each line item to a standard GL expense chart.
- Built Apex normalizer and parser classes so Flow receives flow-friendly fields and a human-readable line-item display, with HTML-entity and markdown fence handling for prompt responses.
- Delivered an invoiceConfirmation LWC for in-flow review and edit of headers, amounts, payment details, and GL codes before returning confirmed JSON to the Flow.