Oracle Fusion AI for Finance

AI Tools Hub

Many large companies in India run their accounts on Oracle Fusion Cloud ERP. Over the last two years Oracle has added a lot of AI to it: features that read supplier invoices, agents that watch the general ledger, tools that write the first draft of a variance note. In this lesson we go through what each one does, how a finance user works with it, which controls keep it safe, and how your company switches it on.

Oracle releases Fusion updates every quarter, named like 26A, 26B, 26C and 26D for the four updates of 2026. Features change fast, so everything here is checked against Oracle’s own pages and is correct as of November 2026.

What you will learn

The three kinds of AI in Oracle Fusion: predictive, generative and agents
What the Payables, Expenses, Ledger, Payments, Cash Processing, Planning and Reconciliation agents do
How a typical invoice, expense and ledger review works with AI, with Indian examples
Where AI helps with reports, narratives and forecasting
The approvals, roles and data controls around these features
How your administrator switches them on

Three kinds of AI in Oracle Fusion

Oracle groups its AI into three kinds. It helps to know which kind you are using, because they behave differently.

Kind	What it does	Example in Fusion
Predictive AI	Learns from past data and predicts a value	Intelligent account combination defaulting, which predicts and fills the account code on an invoice
Generative AI	Writes text, summaries or drafts	Management reporting narratives, which write a draft summary for a report
AI agents	Carry out a piece of work across several steps, and ask you when they need a decision	Payables Agent, which takes an invoice from email all the way to approval

Oracle also now talks about agentic applications. An agentic application is a full workspace built around agents for one job, like the Ledger workspace for accountants. In October 2026, Oracle announced 25 agentic applications powered by Fusion Claw (introduced in September 2026) across ERP, HR, supply chain and customer experience, and said they are available now. Ledger is one of the finance ones.

The finance agents, one by one

Here is the list from Oracle’s AI for ERP page, in plain words.

Agent	What it helps with
Payables Agent	Takes in invoices from email, portals, EDI or e-invoicing, and PDFs. Extracts the data, matches it to purchase orders and receipts, creates the accounting, applies tax, policy and fraud checks, and routes for approval and payment.
Expenses Agent	Extracts expense details, matches receipts to card charges, checks expenses against policy, handles exceptions and schedules submission. Employees can answer its questions in natural language.
Ledger Agent	Answers natural language questions about balances and journals, runs monitoring prompts that flag problems, explains them with details, and can create adjustment journals.
Payments Agent	Helps manage early payment, virtual cards and financing options, connects with bank systems, and watches acknowledgements and exceptions.
Cash Processing Agent	Creates receipts from bank statement lines and matches remittance advices to receipts across many formats.
Reconciliation Agent (Oracle EPM Account Reconciliation)	Helps accountants check account balances continuously, find exceptions, prepare reconciliation explanations and link supporting journals.
Planning Agent	Gives trend and variance analysis through natural language, runs predictions on Fusion data and guides “what if” simulations.
Profitability and Cost Management Agent	A chat interface to build, change, calculate and analyse profitability and cost models.

Remember: Oracle says its AI agents are built into Fusion Applications at no additional cost. Building completely new agents, or using third party agents or outside language models, can need a separate subscription. Check with your Oracle account team before you plan anything custom.

Payables: invoice processing with the Payables Agent

Accounts payable is where most finance teams see AI first, because invoices arrive in every format. Oracle’s 26B release note describes the Payables Agent for Invoice Ingestion, Compliance, and Control like this:

Invoices arrive through streams. A stream is a continuous flow of documents from one sender, file type and channel, like one email address.
Document IO, the default reading engine, uses large language models to read each document and map the data to invoice fields. It replaces the older Intelligent Document Recognition engine for this job.
The Streams page shows a File Capture Status for each file, which tells you whether the invoice was created automatically or needs a manual review.
The Invoice List page is the main place to work. By default it shows the invoices that need your attention. You can ask the agent questions about the invoices you have filtered.
For a flagged invoice, you can cancel it, correct data that was captured wrongly, or release the flag if it was a false alarm.
For a new supplier layout that is not read well, there is a one time Training step.

Users open it from the Payables Agent entry in the Navigator. The old Payables work area stays available, so a team can move across slowly.

Read this chart from top to bottom. When a diamond answers No, follow the side note.

A supplier invoice with the Payables Agent

Worked example: a price difference at Kapoor Textiles

Kapoor Textiles Ltd buys dyes from Mehta Chemicals. The purchase order and the invoice say this:

Item	Quantity	Rate	Value before GST	GST at 18%	Total
Purchase order	500 kg	₹240	₹1,20,000	₹21,600	₹1,41,600
Goods receipt	500 kg
Supplier invoice	500 kg	₹250	₹1,25,000	₹22,500	₹1,47,500

Quantity matches, but the rate is ₹10 higher. The invoice is ₹5,900 more than the PO in total. The agent’s matching step catches this, and the invoice does not move to payment. The AP executive, Neha, opens it from the Invoice List, sees the difference, and asks the buyer. The buyer confirms the old rate of ₹240 is agreed, so Neha asks Mehta Chemicals for a credit note of ₹5,900.

Notice who made the decision. The agent found the problem. A person decided what to do.

Tip: Oracle’s setup advice says supplier names and addresses in Fusion must match exactly what is on the invoices, and your PO number format must match what suppliers print. Clean your supplier master before you start, or the agent will send many invoices for manual review.

For Indian users, Oracle’s AI feature list also includes IDR Adaptive Learning for India HSN Codes (release 25A), a predictive feature in invoice capture. Ask your admin whether it is switched on for your business unit.

Expenses: the Expenses Agent

Expense reports are small but many. The Expenses Agent tries to turn them into one flow: receipt in, details extracted, matched to the card charge, checked against policy, submitted.

From Oracle’s release notes, recent pieces include:

Expenses Agent for Email-Based Expense Completion (26B): employees complete expenses through email.
Expense Policy Inquiry Enhancements Using Expenses Agent (26B): employees ask policy questions.
Expenses Agent for Cash Advance Application and Expenses Agent for Cost Allocations and Additional Information (26C).
Expenses Agent for Itemization and Attendees (26D).

Worked example: a hotel bill over the limit

Arjun, a sales manager in Mumbai, stays two nights in Delhi for a client meeting. His company policy allows ₹8,000 per night for hotels in metro cities.

Expense	Amount	Policy limit	Result
Hotel, 2 nights at ₹9,500	₹19,000	₹16,000	Over by ₹3,000
Cab, airport to hotel and back	₹2,400	No limit, receipt needed	Within policy
Client dinner	₹4,800	Attendees required	Needs attendee names

The agent reads the receipts, matches them to his corporate card charges, checks policy, and asks Arjun for what is missing: a reason for the higher hotel rate and the names of the dinner guests. Arjun replies that the policy hotels were full because of a trade fair. His manager still approves or rejects it. The agent does not decide the exception.

Read the chart below the same way, top to bottom.

An expense report with the Expenses Agent

General ledger and close: Ledger Agent and the Ledger workspace

The Ledger Agent (26A and 26B) is meant to move accountants from chasing reports to getting told when something needs action. Oracle’s 26B release note says it:

Runs monitoring prompts at a frequency you choose, like daily, and shows an insight only when a condition is met. Some prompts come ready made, for example one that flags manual journals posted without approvals.
Answers natural language questions about balances, variances, journals and subledger details from Payables and Receivables.
Explains exceptions, variances and reconciliation differences, with suggested next steps.

In 26C, Oracle added the Ledger Agentic Application, a single Ledger Workspace where accountants ask questions, monitor, see close status across General Ledger, Payables, Receivables and other subledgers, and download large results as Excel files. Oracle encourages existing Ledger Agent users to plan their move to it in 26D.

Limits to know, as of the 26B and 26C notes:

Conversations are in English only.
A closed conversation cannot be reopened.
Questions outside GL balances, journals, close monitoring and supported subledger details are treated as off topic.
Average daily balance and budget balance questions are not supported yet.
Answers only show data your role allows you to see.

Worked example: freight cost jumps in October

The accountant at Kapoor Textiles has a monitoring prompt on freight expense. It flags this:

Account	September 2026	October 2026	Change
Freight outward	₹5,00,000	₹8,00,000	₹3,00,000 (60%)

She asks, “Explain the increase in freight outward for October.” The explanation lists the journals behind the balance. Two journals of ₹1,50,000 each have the same transporter, the same reference and the same date. One is a duplicate of a provision. She checks the transporter’s statement, confirms only one bill exists, and reverses the duplicate journal through the normal journal approval. The real increase is ₹1,50,000, which matches the extra Diwali shipments.

Follow the chart from top to bottom.

A ledger exception from alert to fix

Common mistake: Accepting the explanation without opening the journals. The explanation points you to where to look. You still need the transporter statement, the invoice or the contract before you change anything.

For bank work, Oracle’s 26D feature list includes Cash Processing Agent for Bank Statement Reconciliation, and a Payables Period Close Workspace agentic application for the payables close. These are in the latest quarterly update, so check whether your company has applied it.

Reconciliations in Oracle EPM

Many Oracle customers do account reconciliations in Oracle Fusion Cloud EPM (Enterprise Performance Management), in a module called Account Reconciliation. It already has transaction matching and auto reconciliation, which match high volumes of lines by rules.

In June 2026, Oracle added Account Reconciliation AI Assistants, which give a chat experience through AI Agent Studio:

Assistant	What it does
Transaction Matching Assistant for Administrators	Runs jobs like Auto Match, Run Auto Alert, Train Match Prediction, Archive and Purge Transactions
Account Reconciliation Assistant for Administrators	Manages periods, runs the predicted attribute jobs and checks job status
Account Reconciliation Assistant for Users	Shows reconciliation details, adds and displays comments
Account Reconciliation Run Report Assistant for Users	Runs the Reconciliation Compliance and Transaction Matching reports

These are for customers with Cloud Enterprise EPM or Cloud EDM who also have AI Agent Studio. Setup needs a Service Administrator, a link between the Fusion and EPM identity domains, and a confidential application in EPM. Oracle also says that EPM environments not on the August 2026 (26.08) update or later by the end of January 2027 will lose access to generative AI features.

Reports, narratives and forecasting

Writing commentary is slow work in every close. Oracle has several generative features for it. Oracle marks each of these as generative AI:

Management reporting narratives in ERP: draft narrative summaries for reports.
Generated narratives in Narrative Reporting and EPM Reports: draft text that describes exceptions, compares two periods and finds the top contributors behind a number. Oracle says these are only available on Cloud EPM Enterprise, not Standard licences, and the drafts need a person to review them.
Generative narrative reports in EPM: narrative reports with conditional text and note summaries.
IPM Insights summary: a narrative summary of one insight or a group of insights. IPM Insights itself watches EPM data and alerts you to anomalies, bias and hidden patterns.

For forecasting, the main tools are:

Predictive cash forecasting in ERP: short and medium term cash forecasts, including customer receipts based on how customers usually pay.
Predictive Planning, Auto Predict and Advanced Predictions in EPM Planning: predictions from your history, scheduled predictions, and forecasts that use several drivers.
Risk analysis: Monte Carlo simulation, which runs a model many times with different inputs to show how likely each outcome is.
Planning Agent: variance and trend questions in natural language, plus “what if” scenarios.

Tip: Every one of these narratives is a draft. Read each number in the draft against the report it describes. A draft that says “revenue rose due to higher volumes” when the real reason was a price increase will mislead the CFO.

Controls and approvals

Oracle’s own words in its October 2026 announcement are that people define the objectives, authority and accountability, and agentic applications work within the policies and controls each organisation sets. In practice, these are the controls you will see:

Control	How it shows up
Roles and data access	Ledger answers are limited to the data your role allows. The Payables Invoice List needs business unit data access, or users see no invoices.
Separate privileges	Payables has separate privileges to manage or only view capture, streams and training.
Approvals	Invoices are routed for approval. Expense exceptions go to the manager. Journals follow your journal approval rules.
Monitoring	Ledger has a ready made prompt for manual journals posted without approvals.
Access reviews	Risk Management has an Access Certification Advisor that gives a keep or remove recommendation for each user and role, which the certifier accepts or overrides. Sometimes it gives no recommendation.
Region	The Payables Agent note says its generative AI services run in the Oracle Cloud Infrastructure Commercial Realm (OC1), and other environments get them when Oracle rolls them out there.

How to switch these features on

You do not switch AI on from your own screen. Your Oracle administrator does it, usually after each quarterly update. Each feature in Oracle’s “What’s New” release notes has an action to enable column. In the Financials 26B notes you will see values like these:

Action to enable	What it means	26B example
Setup Required	The feature is delivered but needs configuration first	Ledger Agent, Payables Agent, Payments Agent
Opt In plus Setup	The admin opts in to the feature, then configures it	Expenses Agent for Email-Based Expense Completion
Service Request Required	You must raise a service request with Oracle	Cash Processing from Bank Statements and Remittance Advices

Follow this chart from top to bottom.

Switching on an Oracle AI feature

For the Ledger Agent, the 26B note lists setup steps such as adding the Ledger Inquiry Assistant role to a job role, assigning data access sets, and setting a profile option for REST access. If Oracle managed parts are missing, the Payables Agent shows a message asking you to log a service request and name the agent in its title.

Common mistakes

Expecting a feature you have not enabled. Most AI features say Setup Required. If your admin has not set them up, you will not see them, even after the update.
Dirty supplier master data. Names, addresses and PO formats that do not match the invoices send everything to manual review.
Treating a narrative draft as final. Generative text can sound right and still be wrong. Check every number.
Forgetting role setup. Users without business unit data access see an empty Invoice List and think the agent is broken.
Letting exceptions be released without a reason. When you release a flagged invoice, write why. Your auditor will ask.
Asking the Ledger Agent off topic questions. It is built for balances, journals, close monitoring and supported subledger details, in English.

Quick recap

Oracle Fusion has predictive AI, generative AI and AI agents. Oracle says its agents are built in at no additional cost.
The main finance agents are Payables, Expenses, Ledger, Payments, Cash Processing, Reconciliation and Planning.
The agent does the reading, matching and checking. People decide exceptions and approve.
Narratives and forecasts are drafts that you review.
Features arrive in quarterly updates, and most need setup, opt in or a service request before you can use them.

Try it yourself

Use the Kapoor Textiles invoice from this lesson.

Suppose the supplier invoice says 520 kg at ₹240 instead, but the goods receipt still says 500 kg. Work out the invoice value before GST, the GST at 18% and the total.
Which check would catch the problem, price or quantity?
Write the two line note you would add when you put this invoice on hold.

Answer for step 1: 520 × ₹240 is ₹1,24,800. GST at 18% is ₹22,464. The total is ₹1,47,264. The quantity check catches it, because only 500 kg was received.

Where to go next

SAP Joule for Finance and Accounting: how the same ideas work in SAP
AI for Month-End Close: A Practical Guide: safe use of any AI tool across the close
Three-Way Match in Procure to Pay: the matching logic the Payables Agent follows
Accounting Systems SAP & Oracle: the basics of both ERPs

Sources

These are the official Oracle pages used for this lesson, checked in November 2026.