5.1 AI in the Supplier Invoices Process: IDR, Document IO, the Payables Agent and Intelligent Defaulting

Oracle's exam page lists Understand the role of AI in the Supplier Invoices process as its own objective, and it says the credential covers AI/ML, generative AI and agentic AI. This section covers the four AI capabilities that touch supplier invoices, and the configuration of the one you will most often meet in an implementation: Intelligent Document Recognition (IDR).

1. Intelligent Document Recognition (IDR)

IDR is Oracle's built-in invoice recognition service. It extracts invoice information from emailed documents, creates invoices and imports them into Payables. It also learns from your corrections, so recognition for each supplier improves over time.

End-to-end flow

1. The IDR inbox receives emails with invoice attachments. You get a designated address when you subscribe. Oracle recommends that suppliers send to your corporate mailbox and that you set a mailbox rule to forward to the Oracle address, rather than giving suppliers the Oracle address.
2. IDR processes the attachments and extracts header and line information. Invoices can arrive as a single page, multiple pages, several invoices in one attachment, or several attachments in one email. Oracle's imaging guidance suggests images such as TIFF at a minimum of 300 dpi.
3. Invoices appear in the Scanned Information tile on the Invoices landing page, grouped by age (0–7, 8–14 and 15+ days):
   • Incomplete: the invoice needs attention.
   • Not Validated: all expected fields were recognized.
4. Users correct Incomplete invoices in the Interactive Viewer by selecting the right information on the image, with the mouse or the keyboard.
5. Adaptive learning records the corrections and applies them to that supplier's future invoices. It covers the header, the invoice lines (description, quantity, unit price, UOM and amount) and the routing attributes. Adaptive learning can also be shared between environments.

Configuring IDR

Use Setup and Maintenance > Financials > Payables > Manage Intelligent Document Recognition Options. On first use the buttons show Activate and Activate and Close. Activating starts IDR and begins supplier synchronization. Then:

• the status changes to Active;
• the page shows the unique email address Oracle assigned to you;
• the buttons change to Save and Save and Close;
• suppliers resynchronize every 24 hours, so supplier or PO format changes take up to 24 hours to reach IDR. Allow for this when testing.

The options cover four decisions:

• how to derive the business unit;
• how to recognize invoice lines;
• how to handle tax and freight;
• how to handle service and utility suppliers.

For the business unit, Override Default Business Unit can use the sender's email address or the recipient email address, each with an Email to Business Unit Mapping table. Enter only sender addresses or only recipient addresses in the mapping, to match your choice.

Routing attributes in the email subject

You can put up to four attributes in the email subject, after an underscore. Text before the first underscore is ignored. The attributes can be recorded on the invoice or used to route incomplete invoices. Routing Attribute 1 can also override the default business unit on an unmatched scanned invoice.

Oracle's example subject is Invoice-1234 attached_Urgent_Supply chain related_Plant-1_Bin#1. It gives:

• Routing attribute 1 = Urgent
• Routing attribute 2 = Supply chain related
• Routing attribute 3 = Plant-1
• Routing attribute 4 = Bin#1

The position decides the attribute number, so senders must always use the same order.

Routing incomplete invoices

Incomplete invoices are routed by the BPM task FinApIncompleteInvoiceHold. Its rule set IncompleteInvoiceRuleSet contains JobRoleAssignmentRule, which by default sends them to users with roles based on Accounts Payable Specialist and Accounts Payable Supervisor. You can change or add rules, for example to route Urgent invoices to one person. To do so you need the Manage Financial Applications Workflow Rules privilege, and you work in the BPM Worklist through Manage Task Configurations for Financials. Users need View Scanned Payables Invoice to receive routed invoices, and Edit or Create Payables Invoice to complete them.

2. Document IO: generative AI recognition (26A)

In 26A, Oracle introduced Document IO, a generative AI and LLM-based document ingestion agent. It replaces the IDR backend engine for recognizing supplier invoice attributes. Oracle describes this as a technology upgrade only:

• no change to invoice submission, email endpoints or processing configuration;
• all previous adaptive training is kept;
• better contextual understanding, more file types and languages;
• less dependence on recognizing the supplier first.

In 26A it was available to select customers through Oracle's early-adopter program. Oracle's readiness tips:

• onboard high-volume image suppliers first;
• make supplier names and addresses match the invoices exactly;
• make PO number formats match Procurement;
• exclude supplier types, such as internal suppliers, where needed.

3. The Payables Agent (agentic AI)

The Payables Agent for Invoice Ingestion, Compliance, and Control (26A, extended in 26B to 26D) aims at near-touchless invoice processing. It:

• ingests invoices from email, portals, EDI or e-invoicing and PDFs, in many languages;
• normalizes the data, matches to POs and receipts, and defaults attributes, accounting and tax;
• detects anomalies;
• can turn natural-language standard operating procedures into system rules.

Its experience has four parts:

• Configuration: capture endpoints and streams; compliance and control settings for anomalies and defaulting.
• Streams: real-time monitoring of ingestion.
• Training: upload or review invoice layouts, and import or export training data between instances.
• Invoice List: a single place for exceptions and queries.

Oracle positions it as embedded with no extra subscription, with human oversight kept in the process.

4. Intelligent Account Combination Defaulting (predictive AI)

For invoice lines with no PO, AI and machine-learning models from Oracle AI Apps predict each segment of the distribution combination from historical patterns. They are refreshed as users correct the values.

• A segment is defaulted only if its prediction exceeds a dynamically calculated confidence threshold. Other segments stay blank, and the account search window opens so the user can complete them.
• Oracle recommends at least 3 months of invoice history.

Activation steps:

1. Create the Business Intelligence user.
2. Connect AI Apps to Business Intelligence.
3. Train the models and download the evaluation report.
4. Opt in using the promotion code in that report.

This is different from Transaction Account Definition (section 5.3). TAD derives accounts from your rules. Intelligent defaulting predicts them from your history.