Talk to us

OAN AI automation

The AI is already inside the platform

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.

  • Model-neutral
  • Grounded in your data
  • Production observability
  • Custom-ready
Agents and actions
OAN AssistRisk AgentException Agent
AI runtime primitives
RAGToolsPromptsMemoryGuardrailsObservability
Model layer (pluggable)
OCI Gen AIAzure OpenAIOpenAIClaudeOpen-source
Grounded in
Oracle DBWebCenterERP data

We know what you have been sold

Every vendor is selling you AI. Almost none of it fits your 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.

Every AI demo looked great, but none of them used our invoices.

Finance directorAI demos that never touched your data

Each AI pilot needs a new tenancy, pipeline and team before it starts.

IT directorEvery pilot needs its own infrastructure

The agents act outside our process, and nobody approved what they did.

Process ownerAgents that work in isolation

We already have an LLM agreement and do not need another AI contract.

CIOAnother AI contract you do not need

The AI runtime

The primitives that every OAN agent runs on

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.

How agents work

  • Retrieval-augmented generationRAGEvery 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.
  • Purpose-built agents, not generic botsAgentsOAN 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.
  • Every building block is a toolToolsThe 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.

Control and audit

  • Prompts are versioned codePromptsEvery 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.
  • Full visibility into every callObservabilityToken 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.
  • Policy, PII, and safety before the promptGuardrailsPII 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.

Our approach

AI should live where the work happens

Three principles drive how we built the OAN AI layer. Everything else on this page follows from them.

AI belongs inside the building block

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.

Grounded in your data, not a demo dataset

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.

Plugged into the models you already trust

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.

Inside the building blockNot a separate module
Grounded in your dataOracle Database, WebCenter Content, ERP history
Your modelsYour tenancy, contract and governance

Model neutral

Bring the model you already trust

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.

  • 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
Oracle Cloud Generative AINative for Oracle customers, runs in your OCI tenancy.
Azure OpenAI ServiceUse your existing Azure tenancy and contract.
OpenAI PlatformDirect API access for the latest frontier models.
Anthropic ClaudeVia direct API or AWS Bedrock if you prefer.
Open-source modelsSelf-hosted on OCI or your own infrastructure.

Grounded in your data

The model is replaceable. Your data is not

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.

Your data never leaves the platform

Oracle Database, WebCenter Content, and ERP transactions stay where they are. Only the minimum context needed for a given task is sent to the model.

No training on customer data

OAN does not train models on your data, and the providers we plug into honor the same contract.

Retrieval that knows your schema

Retrieval is built for finance. Invoices know about POs, contracts know about milestones, vendors know about payments. Not generic vector search.

Retrieval-augmented flowIs INV-2847 in line with the Acme MSA?
  • From APInvoice INV-2847
  • From WebCenterMSA-2024-001
  • From Oracle DBMilestone history
  • PolicyPayment terms policy
User asksOAN retrievesModel callGrounded answer
In line with contract. Milestone 3 signed off Mar 5. Amount matches the $48K milestone value.

AI in action

Real AI at work inside OAN products

Every one of these runs on the same AI stack. Same RAG, same primitives, same observability, same pluggable model layer.

01 / 06 · AP AUTOMATION

Invoice extraction and GL coding

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 AP automation

Custom AI solutions

Custom agents, built on the same platform

Your 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.

  • Contract compliance agent

    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.

  • Anomaly detection for high-risk accounts

    A continuous monitoring agent that scores transactions against historical patterns per cost center or GL account, surfacing anomalies for controller review.

  • Policy-aware expense triage

    A custom triage agent that reads expense submissions, cross-references your policy, and routes exceptions to the right approver with reasoning attached.

  • Multi-entity reconciliation agent

    A reconciliation agent that handles intercompany matching across subsidiaries, with human-in-the-loop for anything above materiality thresholds.

The custom agent inherits the platform

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.

Your agent
Configuration on top
AI layer
RAGPromptsToolsGuardrailsObservability
Blocks
CaptureWorkflowContentIntegrationSecurity

Observability & governance

You always know what the AI did, and what it cost

Every AI call is traced, logged, priced, and audited. The same audit trail your team uses for human actions.

  • Full trace per callPrompt version, retrieved context, model used, tokens, latency, and cost captured for every AI invocation
  • Cost by product and customerAI spend rolls up by OAN product, customer, workflow, and user. Finance can see exactly where the budget is going
  • Versioned prompts and evalPrompts are code. Every change is reviewed, tested against an eval set, and can be rolled back
  • Unified audit logEvery agent action goes into the same audit log as every human action. One place to answer the question “what happened?”

FAQ

Frequently asked questions

The questions we hear most often when teams are evaluating the OAN AI layer against standalone platforms, custom stacks, or Oracle Fusion Agent.

Is OAN AI a standalone AI platform that I need to license separately?

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.

How is this different from Oracle Fusion Agent or Fusion AI Agents?

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.

Which AI models does OAN support, and can we use our existing OCI Gen AI or Azure OpenAI tenancy?

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.

How does OAN handle PII and sensitive data when making AI calls?

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.

What if we want a custom AI agent for a process that is not in any OAN product today?

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.

How do you prevent hallucinations in something as high-stakes as finance?

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.

How do we know what the AI cost us, and what it did?

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 every human action.

Do we need a machine learning team to run this?

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.

One AI layer. Every OAN product.

Let us show you AI grounded in your data

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.

Tom Korbecki, Founder, oAppsNET
Tom KorbeckiFounder, oAppsNET