AI Is Coming to ERP. Oracle EBS Clients Do Not Have to Wait for a Migration.

The most immediate opportunity: asking questions of EBS data

One of the clearest developments is Oracle’s work around natural-language access to EBS information.

Oracle has published an approach for enabling natural-language queries against E-Business Suite 12.2 using Oracle APEX and OCI Generative AI. Instead of navigating multiple forms, running a predefined report or asking an analyst to write a query, a permitted user could ask a business question in everyday language and receive information based on EBS data. (Oracle Blog)

In June 2026, Oracle announced further updates to its EBS Natural Language Query solution, including an additional Leasing and Finance AI domain. This is significant because it shows that AI around EBS is moving beyond a broad technical demonstration toward increasingly business-specific applications. (Oracle Blog)

Consider the questions business teams ask every day:

• Which customer invoices are overdue by more than 60 days?
• Which suppliers have the highest number of invoice holds?
• What assets are approaching the end of their useful lives?
• Which departments exceeded budget this quarter?
• Where are purchase orders waiting for approval?
• Which items have experienced repeated shortages?
• What changed in working capital compared with the prior month?

In many EBS environments, answering these questions requires a combination of reports, exports, SQL queries, spreadsheets and support from technical teams.

Natural-language access does not automatically resolve the underlying data, security or reporting challenges. But it can create a simpler interface between business users and enterprise information.

The value is not the chatbot itself.

The value is reducing the time between a business question and a reliable answer.

AI should complement—not bypass—ERP controls

The excitement around generative AI creates a genuine risk: organizations may attempt to place an intelligent interface over data and processes that are not sufficiently controlled.

An AI assistant should not become an alternative route around:

• Role-based access
• Approval limits
• Segregation of duties
• Data-security policies
• Financial controls
• Audit requirements
• Validated business rules

For ERP environments, an answer that sounds convincing but ignores access controls or business definitions is worse than no answer at all.

A finance user asking for “revenue” may mean invoiced revenue, recognized revenue, booked orders or cash received. An operations leader asking about “inventory available” may mean on-hand quantity, available-to-promise quantity or inventory net of allocated demand.

Before introducing AI, the organization must establish:

1. Which information the assistant can access
2. How business terms will be defined
3. Which source is authoritative
4. How user permissions will be applied
5. How the result will be validated
6. Whether the AI can recommend, initiate or complete an action

AI works best when it is layered onto a governed ERP environment—not used to conceal an ungoverned one.

Five practical AI opportunities for EBS clients

1. Natural-language reporting and enquiry

This is likely the most accessible starting point.

An AI-enabled enquiry layer can help users locate information, summarize results and explore trends without replacing existing transactional processes.

Potential applications include:

• Accounts payable and receivable enquiries
• Purchase-order status
• Asset and depreciation information
• Expense analysis
• Inventory availability
• Order status
• Project-cost summaries
• Budget-versus-actual explanations

The first use cases should be narrow, permission-aware and easy to validate. A controlled finance or procurement domain is often more practical than attempting to create one assistant for the entire enterprise.

2. ERP support and knowledge assistance

Many EBS support teams hold years of knowledge across:

• Service tickets
• Configuration documents
• Operating procedures
• Job aids
• Technical specifications
• Oracle support notes
• Testing evidence
• Enhancement histories

Finding the correct answer can depend heavily on a small number of experienced people.

A retrieval-based AI assistant can help support teams search approved internal documentation, summarize prior resolutions and locate relevant procedures more quickly. It can also help business users answer routine “how do I?” questions before opening a service request.

This does not replace experienced functional and technical consultants. It allows them to spend less time finding known information and more time addressing exceptions, process improvements and complex issues.

3. Intelligent document processing

EBS processes often begin outside EBS.

Invoices, purchase orders, contracts, remittance documents, expense receipts and supplier communications may arrive through email, portals, scanned files or shared folders.

AI and document-processing services can help:

• Extract invoice information
• Classify documents
• Compare invoice and purchase-order data
• Identify missing fields
• flag possible duplicates
• Route exceptions
• Draft explanations for reviewers

The strongest design keeps EBS as the controlled transaction system while using AI to reduce the manual effort required to prepare, validate and route information.

4. Exception detection and operational summaries

ERP teams do not need AI to approve every transaction. They may benefit more from AI that helps them understand what requires attention.

Examples include:

• Summarizing invoice holds
• Grouping recurring integration failures
• Explaining unusual spending patterns
• Identifying aging approval queues
• Highlighting transactions requiring review
• Summarizing daily or weekly operational exceptions
• Drafting management commentary based on validated data

This moves the user experience from “review every record” toward “focus on the records most likely to matter.”

Predictive and generative techniques can work together here. A model or business rule identifies the exception; generative AI explains it in practical language.

5. Connecting EBS to AI-enabled workflows

Some of the highest-value applications will require AI to interact with systems beyond EBS.

Oracle Integration can connect applications and data sources across Oracle and non-Oracle environments. Oracle now describes Oracle Integration as a foundation for giving AI agents access to enterprise data and controlled automation. Its current capabilities include converting trusted integrations into tools that agents can call while maintaining governed, observable workflows. (Oracle Doc)

For an EBS client, this could support processes such as:

• Receiving a customer request and retrieving the related order
• Checking EBS invoice status and drafting a response
• Comparing supplier information across EBS and an external portal
• Starting an approval workflow based on an identified exception
• Creating a service ticket with relevant ERP context attached
• Combining ERP, CRM and operational information into one summary

The important design principle is that the AI should not directly improvise changes to core ERP data.

It should work through authenticated APIs, integrations, business rules and approval workflows.

The rise of enterprise AI agents

The market is now moving from AI that only answers questions toward AI that can use tools and participate in multistep workflows.

In March 2026, Oracle made Enterprise AI Agents generally available within OCI Generative AI. The platform supports agent workflows, function calling, file search, managed vector stores, memory and natural-language-to-SQL capabilities. (Oracle Doc)

Oracle’s July 2026 Integration release also expanded agentic AI functionality, including new AI-agent operations, knowledge-base enhancements and human-in-the-loop approval patterns. (Oracle Doc)

For ERP leaders, “agentic AI” should not be interpreted as giving an autonomous bot unrestricted authority over financial or operational systems.

A more realistic enterprise model is:

1. The agent receives a business request.
2. It retrieves permitted information.
3. It calls an approved integration or service.
4. It prepares a recommendation or transaction.
5. A human reviews the action where required.
6. The controlled system executes and records it.

This model can make processes faster without removing accountability.

Do not begin with the technology

A common mistake is to purchase an AI platform and then search for a business problem.

EBS clients should begin by identifying processes where people currently spend significant time:

• Finding information
• Re-entering data
• Reading documents
• Reconciling systems
• Explaining exceptions
• Preparing routine reports
• Searching support documentation
• Following up on approvals

A good first AI use case should have:

• A clearly defined user
• A repeatable business problem
• Reliable source data
• Measurable manual effort
• Manageable security requirements
• A way to verify the output
• A human escalation path
• A business owner willing to participate

“Build an AI assistant for EBS” is not a sufficiently defined use case.

“Help the accounts payable team summarize invoice holds and identify the next required action” is much closer.

A practical roadmap for EBS organizations

Step 1: Stabilize the EBS foundation

Review:

• EBS release and database readiness
• Patch levels
• Security
• Customizations
• Interfaces
• APIs
• Reporting dependencies
• Data quality
• Role design

AI will expose weaknesses in the underlying environment more quickly. It will not correct them automatically.

Step 2: Identify high-friction work

Interview finance, procurement, supply-chain, operations and support users.

Look for activities involving:

• Repetitive searches
• Manual summaries
• Spreadsheet-based reconciliation
• High-volume document review
• Recurring support questions
• Delayed decision-making

Step 3: Select one contained use case

Choose a domain with a clear owner and measurable baseline.

Good examples include:

• AP enquiry
• Purchase-order status
• Fixed-assets support
• ERP knowledge search
• Invoice exception summarization
• Integration-error triage

Step 4: Establish security and governance

Define:

• Data access
• Model access
• Prompt and response logging
• Retention
• Sensitive-data handling
• Human review
• Testing
• Error escalation
• Acceptable use

Step 5: Pilot before scaling

Measure:

• Time saved
• Accuracy
• Adoption
• Number of escalations
• Reduced support effort
• Faster response times
• User confidence
• Control exceptions

The purpose of a pilot is not simply to prove that the technology functions. It is to determine whether the use case produces enough operational value to justify production deployment.

AI does not force an ERP decision—but it should inform one

Adding AI around EBS does not remove the need for a long-term application strategy.

Some organizations will eventually move to Oracle Fusion Cloud Applications. Others will continue operating EBS for a considerable period. Many will maintain hybrid environments combining EBS, cloud applications, third-party platforms, data services and custom solutions.

AI can provide value in each model.

It can also reveal where the organization’s real limitations reside.

The biggest constraint may not be EBS itself. It may be:

• Fragmented integrations
• Inconsistent master data
• Undocumented customizations
• Limited reporting
• Weak process ownership
• Excessive manual work
• An inflexible support model

Those findings can help leadership make a more informed modernization decision.

Modernization does not always begin with replacement

For EBS clients, the emerging AI opportunity is not about adding a conversational interface to every process.

It is about selectively improving how people interact with enterprise systems:

• Making information easier to access
• Reducing repetitive work
• Accelerating support
• Explaining exceptions
• Connecting fragmented processes
• Improving the speed of decision-making

Oracle’s continued EBS support horizon gives organizations room to modernize deliberately. At the same time, developments in natural-language enquiry, OCI Generative AI, enterprise agents and Oracle Integration provide new options for improving existing environments. (Oracle Doc)

The organizations that benefit most will not be those that adopt the most AI.

They will be those that choose the right processes, protect the integrity of their ERP controls and connect AI to measurable business outcomes.