Oracle

Full profile

Which AI this covers: This profile covers AI built into Oracle’s finance platforms, including established machine learning, generative features and agents that can take actions in the ledger, reconciliation and consolidation processes. It does not cover general-purpose assistants a team might use alongside Oracle. The categories are defined in What We Mean by AI.

At a glance

Oracle describes Financial Consolidation and Close as a subscription-based consolidation and reporting solution built for and deployed on Oracle Cloud, with built-in functionality for consolidation and close process tasks, including currency translation, automated cash flow, task management and supplemental schedule data. Its Account Reconciliation service is described as managing balance sheet and consolidation system reconciliations, including preparation and review assignments and workflow. In Fusion Cloud ERP, Oracle’s latest release documentation describes the Ledger Agentic Application as a unified accounting workspace that combines conversational inquiry, proactive financial monitoring and accounting reasoning.

Vendor-stated capabilities

Everything in this section is Oracle’s own description of its products, read in Oracle’s documentation and release notes on October 12, 2026. A vendor’s description establishes what is offered, not what it achieves.

Close, reconciliation and consolidation

Financial Consolidation and Close (Fusion Cloud EPM). Oracle states that the service provides out-of-the-box translations, consolidation, eliminations and adjustments, with an optional setting that tracks intercompany elimination data separately by data source. Consolidation and Task Manager are enabled by default when an application is created; Supplemental Data and Enterprise Journals are optional, and Enterprise Journals manages the preparation and approval of manual journal entries for inclusion in general ledger posting. Oracle lists Enterprise Journals capabilities that include a single point of entry for manual journals, workflow for approval routing, and validation of entries for thresholds, matching balances and format before posting.

Account Reconciliation (Fusion Cloud EPM). Reconciliation Compliance manages preparation and review assignments, controls the preparation and review workflow, and provides visibility into reconciliation status and possible risk conditions. Transaction Matching automates high-volume reconciliations; Oracle describes it as an optional licensed feature, with a user workflow that includes running auto match, confirming suggested matches and creating manual matches.

Narrative Reporting (Fusion Cloud EPM). Narrative Reporting appears in this profile through its AI features, which generate narrative summaries for management reports and answer questions about report data.

Close features in Fusion Cloud ERP. Oracle’s General Ledger documentation describes a Close Monitor that shows period close status across General Ledger, Payables, Receivables, Assets, Projects and Costing for related ledgers, with a manager assigned to each ledger and ledger set. Intercompany reconciliation reports help reconcile intercompany receivables and payables between pairs of provider and receiver legal entities. The ledger offers reporting-only and balance-transfer consolidation, and Oracle describes a third method, Financial Management consolidations, for complex structures such as joint ventures and minority interest holdings.

AI features

Oracle groups AI in Fusion Applications into three technologies: predictive AI using statistical machine learning, generative AI using large language models, and agentic AI, whose agents can trigger and run tasks and interact with enterprise systems based on AI-driven decisions. In Cloud EPM, it describes five routes to AI: predictive features embedded in EPM processes, generative features in the EPM interface, EPM Assistants that customers build as agents in Fusion AI Agent Studio from Oracle templates, Microsoft Office and Excel interaction through Smart View and its extensions, and customer-built integrations that call EPM REST APIs from external AI tools.

The table classifies each feature using the categories in What We Mean by AI. The kind of AI, the suggests-or-acts column and the who-controls-it column are CoE classifications of Oracle’s descriptions, not Oracle’s labels.

Who controls it. Oracle changes these features through its releases. Fusion Cloud Applications receive quarterly updates, some features arrive enabled and immediately available to end users, and features delivered disabled can later be enabled automatically when an opt-in expires. Cloud EPM features carry monthly release numbers, and Oracle publishes a What’s New document for each monthly EPM update; the October 2026 (26.10) update was scheduled for test environments on October 2, 2026 and is scheduled for production environments on October 16, 2026, and is not applied where the customer has asked to skip it. Customers can enable or disable Fusion features that use AI; predictive features also need configuration because the model is trained on the customer’s data. In Financial Consolidation and Close, an administrator turns generative AI on in the application settings. For agents, Oracle states that the models available in AI Agent Studio can vary by release and that older models are deprecated and retired. Where customers connect external AI tools to EPM through Oracle’s APIs, Oracle states that the customer owns model selection, prompt design, security boundaries, auditability and output review.

Customer-added models in AI Agent Studio. Oracle lists these providers for models a customer adds with its own credentials: OpenAI (through Microsoft Azure), Gemini Enterprise Agent Platform (through Google Vertex AI), Google Gemini (through direct API), Anthropic (through Google Vertex AI), Oracle Cloud Infrastructure, and OpenAI (through direct API). Oracle states that it updates this list monthly and may remove deprecated models, that a request for a new model is in most cases completed within two to four weeks, and that the customer is responsible for provider credentials, costs, model lifecycle changes and migrations. The pages read do not say that Oracle’s own finance features use models from these providers.

Models and data handling for Fusion Applications, as Oracle describes them. Oracle states that its machine learning models are retrained every day on new or changed data, that the large language model used by Fusion Applications is a foundational model that is not further trained or fine-tuned, that an organization’s feedback data is isolated from shared models, that Oracle gathers product telemetry, and that it records whether users accept or change the model’s output. For Fusion Applications generative AI data, Oracle states that the retention period is set through a profile option to a value between 90 and 365 days, that older data is deleted automatically, and that a default of 90 days applies if no value is set. These pages do not say whether the same settings apply to Cloud EPM’s own generative features. Oracle states that the models EPM uses outside AI Agent Studio, including in the business process interface, are listed in its OCI Generative AI documentation for GPT-OSS, while EPM Assistants use the models offered in AI Agent Studio. Oracle states that all agent runs in AI Agent Studio are monitored and that a trace view shows each session step by step, including which tools were called. Oracle also states that it completed the Cloud Security Alliance’s AI Consensus Assessments Initiative Questionnaire and lists an ISO/IEC 42001:2023 certification for an AI management system supporting its SaaS applications.

Subprocessors. Oracle’s Services Privacy Policy states that Oracle maintains lists of Oracle affiliates and subprocessors that may process customers’ personal information, available to customers through My Oracle Support.

Direction Oracle has stated. Oracle’s 26D release document says future Ledger Agentic App releases will focus on policy-driven recommendations and autonomous execution of accounting actions within organizational policies and controls. The same document states that it is not a commitment to deliver functionality and should not be relied on in making purchasing decisions.

Independent market evidence

SEC filings. Oracle reports in its Form 10-K for the fiscal year ended May 31, 2026 that Fusion Cloud ERP, which includes Fusion Cloud EPM, is designed to provide financial, project and risk management solutions on a single data and security model. Oracle reports in the same filing that its applications incorporate embedded AI-driven automation, analytics and generative AI capabilities. These are the company’s own statements, made in a filing with the SEC; they describe strategy and risk, not product performance.

The risk factors most relevant to a finance customer, as Oracle reports them:

• Oracle states that its AI products may not operate as anticipated, that it relies on partners and suppliers to produce some of them, and that AI models may contain errors and draw on incomplete or inaccurate data, producing inaccurate outputs.
• Oracle states that flaws in its offerings could affect its customers’ ability to ensure accuracy in financial processes and reporting.
• Oracle states that malicious attacks can lead, and have led, to the compromise of personal information or confidential data of Oracle or its customers.
• Oracle notes that EU AI Act obligations are being implemented in phases through 2030 and that evolving AI law could require it to adjust or limit its offerings or its use of AI in some jurisdictions.

Analyst placements. None are included here. No analyst’s own public page was read for this profile; under our method a vendor’s account of a placement would appear only as a vendor-stated claim.

Regulators and standards bodies. No regulator or standards-body publication concerning these Oracle products was identified in the sources read for this profile.

Independently verified outcomes

No independently verified outcomes were found. Two web searches for independent studies or evaluations of outcomes from Financial Consolidation and Close, Account Reconciliation and Oracle’s ERP finance agents, and a read of the 10-K, returned only Oracle’s own materials, including customer surveys Oracle conducted, and secondary compilations that could not be traced to primary evidence. No regulator site was searched. Oracle’s customer surveys and case studies are vendor-stated and are not used here.

Pricing and packaging

Oracle publishes list prices in its Fusion Cloud Service Global Price List. The figures below are list prices in US dollars from the list dated October 8, 2026; negotiated enterprise terms may differ, and the list does not show implementation, partner or internal costs. Oracle’s list states that the standard term for the ERP, EPM and Fusion AI Agents cloud subscription services priced here is three years (two other sections of the list, Permitting and Licensing and the PaaS Integration Cloud Service for Oracle SaaS, note a one-year term) and that the list is subject to change without notice.

Packaging, as documented. The generative AI settings, Transaction Matching Assistance and EPM Assistants are documented for EPM Enterprise rather than EPM Standard. EPM Assistants also require a Fusion Applications environment with AI Agent Studio, while the AI features in the EPM interface do not. Transaction Matching itself is an optional licensed feature of Account Reconciliation. The price list defines a custom AI agent as one created with, or modified from, Oracle-delivered agents "included with" the customer’s cloud service, for example by adding a tool or bringing the customer’s own model. Fusion AI Units are consumed against a pool: Oracle sums the units used each calendar month and decrements them from the pool, depleted units must be replenished or the consuming usage stopped, and unused units are forfeited at the end of the services period without refund.

What a given AI feature consumes, and which agents fall inside the base subscription rather than the separately priced AI services, is not set out in the material read; that is a question for the order form.

Known limitations

Documented constraints, from Oracle’s own material:

• Edition and mode. The Enable AI option in Financial Consolidation and Close is available only for EPM Enterprise subscriptions, and for Insights only when Hybrid mode is enabled. Transaction Matching Assistance runs only on EPM Enterprise environments and predicts one-to-one matches only.
• Update level. Beginning in July 2026, generative AI in Cloud EPM works only on environments running the 26.04 update or later; environments not upgraded lose access.
• Region. Generative AI in Cloud EPM is not yet available in all OCI regions, and the Enable AI option does not appear where AI is unavailable in the home region.
• Where model calls go. Oracle’s table of OpenAI endpoint locations for Fusion AI agents covers 27 data center locations. For example, it lists USA endpoints for data centers in Australia, Canada, Japan and Brazil, and Europe (EEA and Switzerland) endpoints for data centers in the United Kingdom, Germany, India and Singapore; a customer should read the row for its own data center.
• Language and scope. The Ledger Agentic Application’s conversational experience is available in English, and automatic exception resolution covers journal import and posting exceptions for missing currency rates and inactive or end-dated segment values. The Narrative Reporting Assistant analyzes only the first grid of a report.
• Output reliability. Oracle notes that the summaries generated for IPM Insights in Financial Consolidation and Close may not always be factual, accurate or appropriate, and that the user is responsible for reviewing them.
• Release documents are not commitments. Oracle’s release documentation disclaims any commitment to deliver the functionality it describes.
• Irreversible setup choices. In Financial Consolidation and Close, an application feature cannot be disabled once enabled.

No independent evidence of limitations, such as audit findings or regulator commentary specific to these features, was found in the sources read.

CoE analysis

Oracle’s finance AI spans three of the four categories the CoE uses: established matching models, generative summaries and narratives, and agents that post corrections, change period status or update approvals. The practical consequence is that a finance team running Oracle cannot answer "do we use AI in the close?" with one answer; it has to answer feature by feature. Five things are worth establishing before relying on these features.

What is on, and who turned it on. Some 26D features arrive enabled, the Payables Period Close Workspace needs no enabling step, and opt-in expirations can switch features on in later updates. Oracle’s own 26D material describes the Cash Processing Agent both as enabled by default and as needing setup. An inventory of enabled AI features should therefore be refreshed with each quarterly Fusion update and each monthly EPM update, not once. The close and consolidation playbook sets out that inventory as the first phase of a rollout.

Which features act, and on whose authority. The Ledger Agentic Application can apply policy-approved corrections, the Cash Processing Agent can reconcile and create matching rules once approved, and EPM Assistants can update approval status. Each of these is an actor in the close that needs a documented scope, an owner and a record of the limits it operates within, as argued in Your AI Agent Is an Unlisted Signer. Oracle states that Ledger Agentic Application responses are based on the data the user is authorized to access. That makes the access design of the people using the agent part of the agent’s control.

Whether review can catch the error. Transaction Matching Assistance leaves the decision with the user and attaches a confidence score. A confidence score is the system’s view of its own output; a reviewer confirming high-confidence matches in bulk is not yet a control that would detect a wrong match. Review Is a Control Only When It Can Catch the Error and Treat AI Output Like an Unsupported Journal Entry set out what evidence that review needs.

What changes underneath. Oracle’s models in AI Agent Studio vary by release and older ones are retired, while the customer owns model lifecycle for any model it adds. For generative output used in management reporting narratives or consolidation summaries, finance should know which model and configuration produced a given result. Internal Control Over Generative AI, Applied to the Close treats this as part of the evidence a generative AI control should retain, and A Chart Is a Claim applies the same discipline to the narrative that accompanies a number.

What assurance covers. Oracle’s 10-K warns that flaws in its offerings could affect customers’ accuracy in financial reporting. Whether Oracle provides assurance reports that cover these AI features, and how Oracle evidences the agent traces it describes, are questions to put to Oracle rather than assumptions.

This analysis does not rank Oracle against other vendors, and suitability depends on the organization’s own processes, edition and controls.

Community Observations

Not yet published. Community Observations are added after a licensing review.

Member input under review

None yet.

Relevant diligence questions

1. Which AI features in our Oracle environment are enabled today, which were delivered enabled by default, which opt-in expirations are scheduled in coming updates, and which monthly EPM updates have we taken or skipped?
2. For the Ledger Agentic Application, the Cash Processing Agent, the Payables Period Close Workspace and any EPM Assistants we deploy, what actions can each take without a person’s confirmation, under which identity and role, and where is each action logged?
3. Does Oracle provide service organization control reports for the Fusion Cloud ERP and Cloud EPM services we use; if so, do they cover these AI features and agent runtimes, for which period, and which complementary user entity controls apply to them?
4. Which model produced a given generative output or agent action, how is a model change or retirement notified to us, and how do we retain the prompt, input, output and model version as evidence?
5. Where is our data processed when an AI feature or agent calls a model, including the model endpoint location for our data center and the subprocessors involved, and what retention period applies to generative AI data in Fusion Applications and in Cloud EPM?
6. Which AI features and agents are included in our current subscription, which consume Fusion AI Units or need the separately priced AI services, and how are consumption, depletion and forfeiture shown to us?

Evidence gaps and unanswered questions

• AI unit consumption. The price list refers to a Fusion AI Unit Actions Table that defines what consumes AI units; it was not located in the price list and was not read.
• Base subscription versus AI services. The material read does not state which Oracle-delivered agents are included in the base ERP or EPM subscription and which need the AI Agents, AI Units or Agentic Applications services.
• Assurance coverage. Whether Oracle provides service organization control reports that cover the AI features and agents was not established. The ISO/IEC 42001 certificate and its scope were not read.
• Subprocessor list. Oracle’s subprocessor list sits in My Oracle Support behind a customer login and was not read.
• Feature enablement. How Consolidation Job Analytics, the Consolidation Assistant, the Reporting Agent and Reconciliation Assignment Assistance are enabled is not stated in the pages read.
• Feature-by-process matrix. Oracle’s summary table of EPM AI features by business process uses images that could not be read as text; the per-process pages were used instead.
• Gated material. Oracle’s adoption guides on its customer community, its knowledge-base article on supported models and its known-issues documents sit behind a community or support login and were not read.
• Independent outcomes and analyst placements. None were found or included; see the sections above.
• EPM generative AI data location and retention. The endpoint table and retention settings read cover Fusion Applications; where Cloud EPM’s own generative features process data, and for how long it is kept, was not mapped.

Right of reply

Organizations referenced on this page may respond under our Editorial Independence & Right of Reply policy.

Review status

This profile was last reviewed on October 12, 2026. Vendor-stated information is rechecked monthly. It is not legal, accounting or procurement advice.