OAN AI is the intelligence layer inside every building block. RAG, agents, tools, prompts, and observability, grounded in your own data and plugged into the models you already trust. You are not buying another AI platform. You are getting AI that just works inside the process.
The AI buying cycle in finance looks the same everywhere. Big promises on a demo dataset, a heavy platform to stand up, and a fragile integration to write yourself.
You have sat through a dozen AI platform pitches. Every one of them was impressive on a generic dataset, and none of them could tell you what was in the invoice sitting on your desk.
A new tenancy, a new data pipeline, a new integration project, a new ML team. By the time the agent is live, the problem it was supposed to solve has moved on.
They answer questions nobody asked, take actions nobody approved, and sit outside the process your team actually runs. Governance and audit become an afterthought.
You already have OCI Generative AI or Azure OpenAI or an enterprise LLM agreement. The last thing you need is another vendor telling you which model to use.
Three principles drive how we built the OAN AI layer. Everything else on this page follows from them.
Extraction, workflow, content, and every other block has AI inside it where it makes the work faster, cleaner, or safer. Not a separate module to buy, configure, or integrate later.
Every answer and every action is grounded in your real Oracle Database, your WebCenter Content, and your transactional history. Retrieval-augmented by default, on the data you already own.
OCI Generative AI, Azure OpenAI, OpenAI, Anthropic, or open source. Bring your own tenancy, your own contract, your own governance. We plug in, we do not replace.
Six primitives shared across every product. When we ship a new OAN agent or a custom one for you, it inherits all of them on day one.
Every question is answered against your real data first. OAN retrieves the right invoices, contracts, vendor history, and policy snippets before the model ever sees the prompt, so answers are specific to you, not generic.
OAN agents are scoped to a real finance task (classify this exception, verify this milestone, screen this supplier) and operate inside OAN workflows, not on top of them. They are audited, versioned, and always human-reviewable for critical decisions.
The functions an agent can call are the OAN building blocks themselves: Capture, Workflow, Content, ERP Integration, User Management. Agents do not invent actions. They use the same primitives that humans use inside the platform.
Every prompt template is versioned, testable, and reviewable. Changes go through the same release process as the rest of the platform. No rogue prompts hiding in a notebook somewhere.
Token usage, latency, cost, model, prompt version, and the retrieved context for every AI call. Traces roll up per product, per customer, per workflow, so you always know what the AI did and what it cost.
PII redaction, policy enforcement, rate limits, and output validation happen before the model sees a request and after it returns. The guardrail layer is part of every building block, not bolted on to each use case.
OAN AI is model-neutral by design. Use your existing OCI Gen AI tenancy, your Azure OpenAI contract, your OpenAI account, Anthropic Claude, or self-hosted open-source models. We do not tell you which one to use. We plug in, respect your governance, and let you route the right model to the right task.
Native for Oracle customers, runs in your OCI tenancy.
Use your existing Azure tenancy and contract.
Direct API access for the latest frontier models.
Via direct API or AWS Bedrock if you prefer.
Self-hosted on OCI or your own infrastructure.
Cheap model for extraction. Reasoning model for exception analysis. Fast model for chat. OAN routes each task to the right model for the job and tracks cost per product, per customer, and per workflow.
Every OAN AI call is retrieval-augmented by default. Before a prompt reaches the model, OAN pulls the right invoices, contracts, vendor history, workflow state, and policy snippets from your own Oracle Database and WebCenter Content. The answer the model gives back is grounded in what is actually true for your business.
Every one of these runs on the same AI stack. Same RAG, same primitives, same observability, same pluggable model layer.
Reads any invoice format, extracts header and line items, predicts the right GL code based on your history, and explains every exception in plain English.
See the productMatches complex remittances against open AR, explains deductions, and suggests resolutions drawn from your customer payment history.
See the productScores vendor risk against sanctions, fraud signals, and historical behavior. Detects near-duplicates across your vendor master automatically.
See the productReads PDF, email, EDI, and portal orders, validates against master data, and pushes clean orders straight into Oracle EBS or Fusion.
See the productAnswers questions grounded in your contracts, invoices, workflows, and policies. Works across every OAN product, with full audit trail.
See the productScreens suppliers and payments across bank verification, sanctions, PEP, identity, and email risk in parallel. 35-second decisions.
See the productYour unique process may not match any OAN product out of the box. That is fine. Because the AI layer and the building blocks are the same ones our shipped products run on, we can assemble a bespoke agent for your workflow in weeks, not quarters. Every custom agent inherits the whole platform on day one.
A custom agent that verifies every invoice against the matching MSA or SOW at approval time, flagging milestone, rate, or term mismatches before money moves.
A continuous monitoring agent that scores transactions against historical patterns per cost center or GL account, surfacing anomalies for controller review.
A custom triage agent that reads expense submissions, cross-references your policy, and routes exceptions to the right approver with reasoning attached.
A reconciliation agent that handles intercompany matching across subsidiaries, with human-in-the-loop for anything above materiality thresholds.
Your bespoke agent is not a separate code base. It is a new configuration on top of the same building blocks, the same AI primitives, and the same observability that every OAN product runs on. When the platform gets better, your custom agent gets better with it.
Every AI call is traced, logged, priced, and audited. The same audit trail your team uses for human actions.
Prompt version, retrieved context, model used, tokens, latency, and cost captured for every AI invocation.
AI spend rolls up by OAN product, customer, workflow, and user. Finance can see exactly where the budget is going.
Prompts are code. Every change is reviewed, tested against an eval set, and can be rolled back.
Every agent action goes into the same audit log as every human action. One place to answer the question “what happened?”
The questions we hear most often when teams are evaluating the OAN AI layer against standalone platforms, custom stacks, or Oracle Fusion Agent.
No. OAN AI is the intelligence layer already inside every OAN building block. When you buy AP Automation, Cash Application, Vendor Management, or any other OAN product, you get the AI that runs inside them. There is no separate AI platform license, no parallel agent framework to stand up, and no additional team required to operate it.
They solve different problems. Oracle Fusion Agents sit inside Oracle Fusion ERP and operate on Fusion data and processes. OAN AI sits inside OAN products (AP, Cash App, Vendor Management, and so on) and operates on the data and processes those products handle. If you run both, they coexist. OAN AI is not a replacement for Fusion Agent and is not positioned as a competitor.
Yes. OAN is model-neutral. You can plug OAN AI into OCI Generative AI, Azure OpenAI, OpenAI, Anthropic Claude, or self-hosted open-source models (Llama, Mistral). We use your tenancy, your contract, and your governance. If you already have a preferred provider, we use it. If you want us to recommend one, we will.
A guardrail layer runs before every prompt. PII redaction, policy enforcement, and output validation are configured per use case and enforced automatically. Your data stays in your Oracle Database and WebCenter Content; only the minimum context required for a given task is sent to the model, and nothing is used for training.
That is exactly how the stack is designed. Because the AI layer and the building blocks (Capture, Workflow, Content, Integration, Security) are the same ones the shipped products use, we can assemble a custom agent on top of them in weeks, not quarters. The custom agent inherits all the platform primitives (RAG, prompts, tools, observability, guardrails) on day one.
Three layers. First, retrieval-augmented generation grounds every answer in your real data before the model sees the prompt. Second, output validation checks structured responses against expected schemas and policies. Third, for any decision that moves money or affects the ledger, a human is always in the loop. Agents recommend or draft, people approve.
Observability is built in. Every AI call logs token usage, latency, cost, model used, prompt version, and the context it retrieved. Costs roll up by product, customer, workflow, or user, so finance can see exactly where AI spend is going. Every agent action is also captured in the same audit log as human actions.
No. OAN AI is operated as part of the platform. We handle prompt versioning, eval, model routing, cost optimization, and guardrail updates. Your team configures the use cases that matter to you. The platform handles the rest.
A 60-minute briefing tailored to your stack, your models, and one real use case we can demonstrate against your own process. No generic demo, no slideware.
No commitment. A working walkthrough with your team.