Process Custom CSV and XML Files

The Payables Agent processes custom CSV and XML files through its invoice document processing flow.

Ingest Custom CSV and XML Files

The Payables Agent processes custom CSV and XML files through its invoice document processing flow. When a file arrives, the Ingestion Agent examines its structure and uses generative AI to map source columns or XML elements to Oracle Fusion Payables invoice attributes.

If document throttling is enabled and a file has a previously unseen structure that meets the configured criteria, it receives a New Format Detected status. You can review and refine the suggested mappings before subsequent files with the same structure are processed automatically.

Create and Train Invoices from Custom CSV and XML Files

You can process supplier invoice data in custom CSV and XML formats and use instruction-based learning to teach the Payables Agent how to map and transform source data into Oracle Fusion Payables invoice attributes.

Generative AI analyzes incoming file structures and identifies mappings between source attributes and Payables invoice attributes. You can review the generated mappings, provide instructions or corrections, define transformation rules, and save the learning for subsequent files with the same structure.

Train Structured Documents

Structured documents use instruction-based learning instead of annotation-based learning.

Instruction-based learning lets you describe, in natural language, how information in a source CSV or XML file should be mapped, derived, or transformed when creating a Payables invoice. The Payables Agent interprets these instructions and applies them to the document processing logic.

For example, you can provide instructions such as:

Map Number to Invoice Number.
Map Payment Terms to Payment Term.
Map Date to Invoice Date.
Prefix the invoice number with the year from the invoice date.

For XML files, instructions can reference complete or partial XML paths. You can train the agent without creating and maintaining custom transformation logic for each supplier-specific file structure.

Review the Source File

When training a custom CSV or XML document, you can access the source file from the training UI. Download and open the source file separately to review its original attributes and values while defining or validating instructions.

Use the Instruction Panel

The instruction panel organizes processing results into three sections to help you focus on mappings that require attention:

Available: Displays source attributes that haven't yet been mapped to Payables invoice attributes.
Needs Attention: Highlights required attributes that aren't mapped or mappings that require review.
Edit Instructions: Displays identified and mapped attributes with representative mapping instructions.

Use these sections to review the AI-generated results and provide mapping or transformation instructions as needed.

Map Source Attributes to Payables Invoice Attributes

You can use natural-language instructions to associate an attribute from the source file with a Payables invoice attribute.

For example:

Map SupplierInvoiceNumber to Invoice Number.

After you provide an instruction, the Payables Agent identifies the source and target attributes and updates the mapping. The Preview area displays the resulting Payables invoice attributes and values so that you can validate the outcome.

You can also enable Show All Attributes to review Payables invoice attributes regardless of whether they're currently mapped.

Review Source and Target Attributes

The training UI helps you determine which source and target attributes to use in your instructions.

You can:

Review source attributes that have already been captured.
Review source attributes that remain available for mapping.
View successfully mapped Payables invoice attributes and their source values.
Use Show All Attributes to review mapped and unmapped Payables attributes.

These capabilities help you identify the appropriate source and target attributes before defining a mapping.

Define Transformation Instructions

In addition to one-to-one attribute mapping, you can use instruction-based learning to transform invoice data before creating the invoice.

Examples include:

Split values: Split a source value into multiple Payables attributes.
Concatenate values: Combine multiple source values into a single target attribute.
Create invoice lines: Create freight or miscellaneous lines from information supplied elsewhere in the source document.
Convert values: Convert dates or other values into the required format.
Derive values: Derive information such as payment terms from other source information.
Apply conditional transformations: For example, convert positive amounts to negative values when processing credit memos.

Use reusable instructions to handle supplier-specific variations without repeating the same correction for every invoice.

Upload and Download Instructions

You can also manage learning through instruction documents.

You can:

Upload Word or text files containing mapping and transformation instructions.
Allow the Payables Agent to extract instructions from the uploaded document.
Replace existing instructions with uploaded instructions after review.
Review and edit the extracted instructions before applying them.
Download the current instructions for reference, offline review, or reuse.

Use instruction documents when mapping and transformation requirements are maintained externally or when you need to manage a larger set of instructions together.

Reuse Saved Learning

When you save reviewed mappings and instructions, the learning is associated with the detected document structure.

When another document with the same fingerprint is received, the Payables Agent reuses the saved mappings and transformation instructions. Processing can continue without repeating the training.

Review and train a new supplier or document format once to reduce manual intervention for subsequent documents with the same structure.