Someone needs a new laptop. They open a requisition, and a question freezes them. Does this need a quote? Is there a preferred supplier? What is the spending limit before a second approval kicks in? The policy holding every answer lives in a PDF on a shared drive nobody opens. So they guess. Either the requisition bounces back, or a non compliant purchase slips right through. Multiply that single moment by a few thousand a year, and you have the exact problem the Procurement Policy Advisor was built to solve.

It is one of the AI agents Oracle now ships inside Fusion. It also happens to be a great first agent to learn, because the payoff is obvious and the setup stays contained. So here is what it does, why it matters, and how to actually stand one up.

What it is

The Procurement Policy Advisor is a conversational agent living inside the Redwood Self Service Procurement application. While someone builds a requisition, they can ask it plain questions about products, services, and policy, and it answers from your company's own documents. Not a generic chatbot bolted on the side. It reads the policy files you feed it and responds in the moment, right where the work happens.

Two outcomes drive the whole thing. It streamlines access to policy and process documents at the point of creating a requisition, so nobody has to leave the screen to chase a rule. And it improves compliance by handing over policy guidance as people submit requisitions, catching the guesswork before it turns into a problem.

The mechanism is easy to picture. The agent has a document tool, which holds your attached and published procurement policy and guideline documents. Someone asks a question, the agent pulls its answer from those documents. Good documents in, good guidance out.

Why it matters more than it looks

Procurement runs on rules most requesters never learn, because learning them was never their job. The result is a slow, error prone middle layer where buyers answer the same policy questions on repeat and non compliant requisitions get caught late, if they get caught at all. The advisor collapses that layer. It drops the rulebook inside the requisition, phrased as a conversation, sitting right there the instant a question comes up. That is a real change in how compliance happens. Guidance in the moment instead of enforcement after the fact.

There is a timing argument too. Oracle has moved agent creation into one place, AI Agent Studio, and it flat out recommends that anyone who built agents before update 25C rebuild them there. Getting fluent in this setup is turning into a core Fusion skill, not a party trick.

The build, start to finish

The whole thing has a clear shape. Turn on the platform. Grant yourself access. Create the tool that holds the policy documents. Wire it into an agent. Surface it to users. Here is the path, drawn from Oracle's own configuration guidance.

Start with the profile options. In Setup and Maintenance, open Manage Administrator Profile Values and set the enabling profiles to Y. This flow leans on three: ORA_HCM_VBCS_PWA_ENABLED, ORA_PER_GUIDED_JOURNEYS_SETUP_REDWOOD_ENABLED, and ORA_PER_AGENT_TASK_TYPE_GUIDED_JOURNEYS_ENABLED. Nothing works until these are on.

Next, the security role. In the Security Console, create a role, say AI_AGENT_ADMIN_ROLE, and assign the privileges that let a user manage and use agents, including HRC_MANAGE_AI_AGENT_PRIV and HRC_ACCESS_AI_AGENT_CHAT_PRIV. Give that role to yourself so the rest of the setup is even possible.

Then build the policy tool. Head to My Client Groups, click Show More, and open the Configure Agents page for your area. On the Tools tab, add a new tool, name it something clear like Procurement Policies Tool, and attach the documents that make up your policy. This is the most important content step in the entire build. The agent can only ever be as good as the documents behind it. Curate them ruthlessly. A tight set of current, well written policy files beats a heap of outdated ones every time.

Create the agent from the template next. Open AI Agent Studio from the Navigator under Tools, find the Procurement Policy Advisor template, and choose how to create it. Copy Template is the quick path. It builds the agent team and its artifacts automatically and drops you on the canvas to adjust. Use Template is the guided path, walking through each piece one at a time. For a first build, Copy Template gets you working fast, then refine.

On the canvas, attach the policy tool you made so the agent knows where to read from. Head to the Security tab and set which roles can reach the agent. That controls who actually gets the advisor inside their requisition flow.

Last, surface it with a guided journey. The advisor reaches users through a guided journey on the Redwood Self Service Procurement page, enabled through Visual Builder Studio. This is the step that plants the agent inside the requisition experience instead of leaving it a standalone tool. Configure the journey so the advisor appears where requesters naturally reach for help.

Then test as a requester. Open the Redwood requisition flow and throw real questions at it, the ones buyers field every week. Does this category need a quote. What is the limit for this expense type. Who is the preferred supplier for laptops. Check the answers against your documents, then tighten the document set wherever a reply came back vague or wrong.

A migration note worth knowing

If the organization built a policy agent before update 25C using the older RAG setup, do not start from scratch. Oracle gives you a migration path. Go to My Client Groups, then Show More, then Configure RAG Agents, open the Tools tab, and hit Migrate on the existing document tool. Give it a unique name and code, select the family and product, then find that migrated tool in AI Agent Studio and attach it to a new agent built from the template. You keep the document work already done and simply move it onto the current platform.

Ways to make it genuinely useful

The default use, answering policy questions during requisition creation, already earns its keep. A few ways to push further.

Encode the tribal knowledge, not just the policy. The official document is the obvious input. The higher value input is the unwritten stuff living in your buyers' heads. Which supplier for which category. Which purchases always need extra documentation. The thresholds people forget. Capture that as a clean guidance document and the advisor becomes a standing answer to questions that used to require a human every single time.

Keep the documents alive. An agent pointed at a stale policy gives confidently wrong answers, which is worse than no agent at all. Treat the document set as something to review on a schedule, especially after quarterly updates or policy changes. The setup is a one time job. The document upkeep is the real ongoing work, and it is what separates an advisor people trust from one they learn to ignore.

Watch what people ask. The questions users bring to the advisor draw a live map of where your policies are fuzzy. Same question, over and over, on one topic? The underlying policy or the document explaining it needs work. The agent quietly becomes a feedback loop on your own rules.

The honest limits

This is early technology and it deserves a clear eyed look. The advisor answers from the documents you give it, so its quality is entirely a function of that content. It guides, it does not enforce, so it sits alongside approval rules rather than replacing them. And the setup touches profile options, security roles, and Visual Builder Studio, so it is a genuine configuration effort, not a switch. None of that dents the value. It just means the work lives in the preparation, which is precisely where a good consultant earns their place.

The Procurement Policy Advisor is a small, sharp preview of where Fusion is heading. Expertise dropped straight into the moment of work, phrased as a conversation. Learning to build one well is a skill worth having now.