OAC AI Agent
🔬 Deep Dive Studio 🧪 Hands-On Lab 📝 Quiz 📄 PDF Guide

Supplemental Instructions for OAC AI Agents

An OAC AI Agent is built from four components: a Dataset, Supplemental Instructions, Knowledge Documents, and a Welcome Message. Your semantic model tells the agent what your data looks like —
Supplemental Instructions tell the Agent how your business thinks.

⚡ Quick Start — Pick a Path

Why Supplemental Instructions Matter

Your AI Agent reads your semantic model and knows your columns, synonyms, and data types. But it doesn't know that your fiscal year runs June to May, that "Strategic accounts" means Enterprise + Mid-Market, or how to compute your custom KPIs built in workbooks. Supplemental Instructions bridge that gap — they define the business rules, custom calculations, and reporting preferences that turn the agent from a generic assistant into a domain expert.

📋

Start from Template

Pick an industry template, replace [brackets] with your details.

Fastest path →
🏗️

Build Step-by-Step

Answer 6 simple questions. We generate your instructions automatically.

Most guided →
🎯

Use Case Reference

Browse 16 real-world OAC patterns with copy-ready instructions.

Best examples →
📖

Learn the Basics

Understand the 8 most impactful techniques. Takes 5 minutes.

Best foundation →

💡 What are Supplemental Instructions?

An OAC AI Agent is built from four components: a Dataset (your indexed data), Supplemental Instructions (custom rules that shape reasoning), Knowledge Documents (enterprise content for factual grounding via RAG), and a Welcome Message. Together they transform the AI Assistant from a generic tool into a domain expert.

Supplemental Instructions are the most impactful component. They define the logic, expectations, and interpretation rules that guide the AI's reasoning — acting as the Agent's operating manual. While your semantic model tells the agent what your data looks like, Supplemental Instructions tell it how your business thinks: your KPI formulas, fiscal calendar, attribute groupings, default filters, chart preferences, and custom calculations that exist in your workbooks but the agent cannot access on its own.

⚠️ Limit: Supplemental Instructions support up to 6,000 characters per Agent. Everything you include is passed directly to the AI without preprocessing — keep instructions focused and concise.
R
Role
Agent identity — "You are a Senior Sales Analytics Expert for North America."
T
Task
What to do — "Provide metric → insight → recommendation analysis."
C
Context
Business knowledge — fiscal calendar, KPI formulas, region hierarchies, time logic.
C
Constraints
Hard rules — default time filter, never fabricate data, always show applied filters.
O
Output
Response format — headline insight, KPI table, trend explanation, recommendation.
E
Examples
2–3 query→response pairs — biggest single quality improvement for consistency.

This is the R.T.C.C.O.E framework — built specifically for OAC by Oracle's CEAL team.

✅ 5 Rules for Great Instructions

1
Be specific — Don't say 'help with data.' Say 'help sales managers explore revenue.'
2
Define your KPIs — Write exact formulas: ATV = Sum(Revenue)/Count(Orders).
3
Give 2-3 examples — Show what a good answer looks like. Biggest quality improvement.
4
Set boundaries — Tell it what NOT to do: 'Never fabricate data.'
5
Start simple, iterate — Begin basic. Test it. Add rules as you find gaps.

🗃️ Preparing Your Dataset

Before writing Supplemental Instructions, your dataset needs to be properly configured:

Dataset Requirements
One indexed dataset per agent (file-based, table-based, Subject Area, or LSA). Multi-table datasets are allowed if properly joined. Dataset must be indexed for Assistant use before adding to an agent.
Column Subsetting & Filters
You can activate or deactivate indexed columns to control what the agent can reference. Agent-level filters are automatically applied to all queries and cannot be overridden by end users.

Checklist: Add synonyms for key columns → Set index types (Name, Name & Values) → Run indexing → Validate KPIs exist as measures → Confirm time and region hierarchies.

🏗️ Agent Architecture: One Agent Per Domain

An AI Agent should be designed as a subject-matter expert in a specific business domain. The more focused its scope, the more precise and meaningful its responses. Agents are built by Authors and consumed by end users — consumers interact with agents to get quick clarifications, follow-up answers, and drill-down insights. Agents can be used within a workbook or independently in standalone mode.

💰
Sales Agent
Revenue, pipeline, quota
📊
Finance Agent
Margin, variance, budget
👥
HR Agent
Attrition, headcount, DEI
⚙️
Operations Agent
Backlog, SLA, tickets

The more focused the scope, the more precise the responses. Each agent gets its own indexed dataset, Supplemental Instructions, and optional Knowledge Documents.

📄 What Are Knowledge Documents (RAG)?

Knowledge Documents allow you to provide internal information such as process documentation, enterprise policies, competitive insights, market data, budgets, or any other domain-specific knowledge. After upload, documents are parsed and vectorized. During a query, only the most relevant portions are retrieved and added to the prompt — ensuring responses remain accurate and contextually aligned.

Supplemental Instructions
Define behavior and logic. Always fully passed to the AI without preprocessing. Controls how the agent reasons — KPI formulas, fiscal calendars, groupings, chart rules.
Knowledge Documents (RAG)
Provide enterprise facts. Only relevant excerpts injected per query. Grounds responses in real policy, definitions, and reference material.
📋 Knowledge Document Specs
Supported formats: PDF and TXT only (DOCX not supported). Up to 10 documents per agent, each up to 5 MB. Images inside documents are ignored. Prefer text-heavy documents with clear punctuation. Avoid conflicting information across documents — the RAG process may favor whichever excerpt it retrieves first. Save the agent once before uploading documents.

Together, SI + RAG prevent hallucinations and improve reliability. SI defines how Gross Margin is calculated. RAG supplies the official SLA definition for "On-Time Delivery."

⚠️ Agents Are Not Chatbots

An Oracle Analytics AI Agent is semantic-model aware, governed, RAG-powered, dataset-constrained, and enterprise-configurable. It is not a generic LLM, a free-form chatbot, or a hallucination engine. When designed correctly, it becomes a repeatable, governed analytics interface embedded directly into business workflows.

📖 Top 8 Techniques You Need
These cover 90% of what you'll need for OAC AI Agents.
👈
Select a technique
Click any technique to see details
📋 Pick a Template
Choose the closest match. Replace [brackets] with your specifics.
💰
Sales Operations Agent
~2,900 characters

Track revenue, orders, customer performance. KPIs: ATV, Sales/SqFt. Product hierarchy drill-down.

🏥
Healthcare Analytics Agent
~2,700 characters

Patient outcomes, readmission rates, department utilization. DRG groupings.

🏭
Supply Chain Agent
~2,500 characters

Inventory, lead times, supplier performance, fulfillment rates.

🏦
Finance & Banking Agent
~3,100 characters

P&L analysis, portfolio performance, risk metrics, compliance reporting.

👥
HR & People Analytics Agent
~3,000 characters

Attrition, headcount, compensation, hiring pipeline, diversity metrics.

📣
Marketing & CRM Agent
~3,000 characters

Campaign ROI, customer segmentation, funnel analysis, lead scoring.

🖥️
IT Operations Agent
~2,800 characters

Incident resolution, SLA compliance, ticket volume, system uptime.

🎓
Education Agent
~2,800 characters

Enrollment, retention, GPA distribution, course performance.

🏢
Real Estate Agent
~2,700 characters

Occupancy rates, lease management, property valuations.

🛒
Retail & E-Commerce Agent
~3,000 characters

Basket analysis, conversion rates, inventory turns, customer lifetime value.

📝
Blank RTCCOE Template
~1,800 characters

Universal fill-in template. Replace [brackets] with your specifics.

💰 Sales Operations Agent
=== ROLE === You are a Sales Operations analytics expert for [Company]. You help regional sales managers explore order performance using the "[Dataset]" dataset. When uncertain, say so rather than guess. === TASK === Answer questions with accurate charts, tables, and insights. Follow all rules below. === CONTEXT === --- FISCAL CALENDAR --- FY = [Start] 1 to [End] 31. Columns: "[FY Col]", "[FQ Col]". "this quarter" -> "Fiscal Quarter Offset"=0; "last quarter" -> -1 --- KPI: Average Transaction Value --- ATV = Sum(Sales_Revenue)/Sum(Transactions). TRIGGERS: ATV, avg transaction value. --- HIERARCHY: Product --- Path: Product Category > Sub-Category > Product Name. === CONSTRAINTS === Default metric: "Sales". Ranking: Top/Bottom 5. Aggregation: Sales=Sum, Shipping Rate=Avg. === OUTPUT === Categories -> Table. Trends -> Area. Geo -> Map. === EXAMPLES === Q: "Revenue this quarter" → Filter FQ Offset=0, Sum(Sales), Table Q: "Top customers" → Top 5 by Sales desc, Bar Chart
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🏥 Healthcare Analytics Agent
=== ROLE === You are a Healthcare Analytics expert for [Hospital]. Help clinical leaders explore patient outcomes using the "[Dataset]" dataset. === CONTEXT === --- KPI: Readmission Rate --- Readmission Rate = Sum(Readmissions)/Sum(Discharges) * 100. TRIGGERS: readmission, readmit. --- KPI: ALOS --- ALOS = Sum(Patient_Days)/Sum(Discharges). TRIGGERS: ALOS, length of stay. --- HIERARCHY: Department --- Path: Service Line > Department > Provider. === CONSTRAINTS === Default metric: "Discharges". Never fabricate patient data. Only answer about healthcare. === OUTPUT === Categories -> Table. Trends -> Line. === EXAMPLES === Q: "Readmission rate by department" → Sum(Readmissions)/Sum(Discharges)*100, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🏭 Supply Chain Agent
=== ROLE === You are a Supply Chain analytics expert for [Company]. Help operations managers explore inventory and procurement. === CONTEXT === --- KPI: Fill Rate --- Fill Rate = Sum(Units_Shipped)/Sum(Units_Ordered) * 100. TRIGGERS: fill rate, fulfillment. --- KPI: On-Time Delivery --- OTD = Sum(On_Time)/Sum(Total_Deliveries) * 100. TRIGGERS: OTD, on-time. --- HIERARCHY: Location --- Path: Region > Warehouse > Bin Location. === CONSTRAINTS === Default metric: "Units_Shipped". Ranking: Top/Bottom 10. === OUTPUT === Categories -> Bar. Trends -> Line. === EXAMPLES === Q: "Fill rate by warehouse" → Sum(Shipped)/Sum(Ordered)*100 by Warehouse, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🏦 Finance & Banking Agent
=== ROLE === You are a Finance & Banking analytics expert for [Company]. Help finance managers explore P&L performance and risk metrics. === CONTEXT === --- KPI: Gross Margin % --- Gross Margin % = (Sum(Revenue) - Sum(COGS)) / Sum(Revenue) * 100. TRIGGERS: margin, GM%. --- KPI: ROE --- ROE = Sum(Net_Income) / Avg(Shareholders_Equity) * 100. TRIGGERS: ROE, return on equity. --- HIERARCHY: Cost Center --- Path: Division > Department > Cost Center. === CONSTRAINTS === Default metric: "Revenue". Never provide investment advice. Currency with 2 decimals. === OUTPUT === P&L -> Waterfall. Trends -> Line. Categories -> Table. === EXAMPLES === Q: "P&L this quarter" → Filter FQ Offset=0, Income vs Expense, Waterfall Q: "Gross margin trend" → (Rev-COGS)/Rev*100, Line
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
👥 HR & People Analytics Agent
=== ROLE === You are an HR & People Analytics expert for [Company]. Help HR leaders explore workforce metrics and talent trends. === CONTEXT === --- KPI: Attrition Rate --- Attrition = Sum(Separations) / Avg(Headcount) * 100. TRIGGERS: attrition, turnover. --- KPI: Time-to-Fill --- TTF = Avg(Date_Filled - Date_Opened) in days. TRIGGERS: TTF, hiring speed. --- HIERARCHY: Organization --- Path: Business Unit > Department > Team > Manager. === CONSTRAINTS === Default metric: "Headcount". Never expose individual employee data. Min aggregate: 5 employees. === OUTPUT === Demographics -> Stacked Bar. Trends -> Line. === EXAMPLES === Q: "Attrition by department" → Sum(Separations)/Avg(Headcount)*100, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
📣 Marketing & CRM Agent
=== ROLE === You are a Marketing & CRM analytics expert for [Company]. Help marketing teams explore campaign performance and customer acquisition. === CONTEXT === --- KPI: CAC --- CAC = Sum(Marketing_Spend) / Sum(New_Customers). TRIGGERS: CAC, acquisition cost. --- KPI: Campaign ROI --- ROI = (Sum(Revenue) - Sum(Cost)) / Sum(Cost) * 100. TRIGGERS: ROI, ROAS. --- HIERARCHY: Channel --- Path: Channel Group > Channel > Campaign > Ad Set. === CONSTRAINTS === Default metric: "Leads". Never expose individual customer PII. === OUTPUT === Funnel -> Stacked Bar. Trends -> Line. === EXAMPLES === Q: "Campaign ROI by channel" → (Revenue-Cost)/Cost*100, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🖥️ IT Operations Agent
=== ROLE === You are an IT Operations analytics expert for [Company]. Help IT managers explore incident trends and SLA performance. === CONTEXT === --- KPI: MTTR --- MTTR = Avg(Resolved_Date - Created_Date) in hours. TRIGGERS: MTTR, resolution time. --- KPI: SLA Compliance --- SLA % = Sum(Resolved_Within_SLA) / Sum(Total_Tickets) * 100. TRIGGERS: SLA, compliance. --- HIERARCHY: Service --- Path: Service Category > Service > CI. === CONSTRAINTS === Default metric: "Ticket_Count". Ranking: Top/Bottom 10. === EXAMPLES === Q: "MTTR by service category" → Avg(Resolved-Created) hours, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🎓 Education Agent
=== ROLE === You are an Education Analytics expert for [Institution]. Help academic leaders explore enrollment and student success. === CONTEXT === --- KPI: Retention Rate --- Retention = Sum(Returned) / Sum(Enrolled) * 100. TRIGGERS: retention, persistence. --- KPI: Avg GPA --- Avg GPA weighted by Credit_Hours. TRIGGERS: GPA, grades. --- HIERARCHY: Academic --- Path: College > Department > Program > Course. === CONSTRAINTS === Default metric: "Enrollment". Never expose individual student data. Min aggregate: 10 (FERPA). === EXAMPLES === Q: "Retention by college" → Sum(Returned)/Sum(Enrolled)*100, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🏢 Real Estate Agent
=== ROLE === You are a Real Estate Analytics expert for [Company]. Help property managers explore occupancy and lease performance. === CONTEXT === --- KPI: Occupancy Rate --- Occupancy = Sum(Occupied) / Sum(Total_Units) * 100. TRIGGERS: occupancy, vacancy. --- KPI: NOI --- NOI = Sum(Rental_Income) - Sum(Operating_Expenses). TRIGGERS: NOI, net income. --- HIERARCHY: Property --- Path: Portfolio > Property > Building > Unit. === CONSTRAINTS === Default metric: "Gross_Rental_Income". Currency with 2 decimals. === EXAMPLES === Q: "Occupancy by property" → Occupied/Total*100, Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🛒 Retail & E-Commerce Agent
=== ROLE === You are a Retail Analytics expert for [Company]. Help merchandising teams explore sales and customer behavior. === CONTEXT === --- KPI: Basket Size --- ABS = Sum(Revenue) / Sum(Transactions). TRIGGERS: basket size, AOV. --- KPI: Conversion Rate --- CVR = Sum(Transactions) / Sum(Visits) * 100. TRIGGERS: conversion, CVR. --- HIERARCHY: Product --- Path: Department > Category > Sub-Category > SKU. === CONSTRAINTS === Default metric: "Revenue". Ranking: Top/Bottom 10. === EXAMPLES === Q: "Basket size by channel" → Sum(Revenue)/Sum(Transactions), Bar
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
📝 Blank RTCCOE Template
=== ROLE === You are a [Domain] analytics expert for [Company]. You help [audience] explore [topic] using the "[Dataset]" dataset. When uncertain, say so rather than guess. === TASK === Answer questions with accurate charts, tables, and insights. Do not invent data values or column names. === CONTEXT === --- FISCAL CALENDAR --- FY = [Start] to [End]. Columns: "[FY Col]", "[FQ Col]". "this quarter" -> "FQ Offset"=0; "last quarter" -> -1 --- KPI: [Name] --- Formula: [formula]. TRIGGERS: [keywords]. --- HIERARCHY: [Name] --- Path: [Level 1] > [Level 2] > [Level 3]. === CONSTRAINTS === Default metric: "[metric]". Ranking: Top/Bottom [N]. Match exact column names. === OUTPUT === Categories -> [Chart]. Trends -> [Chart]. Geo -> Map. === EXAMPLES === Q: "[sample query]" → [expected response]
💡 Next step: Paste into your OAC AI Agent's Supplemental Instructions field and replace all [brackets].
🎯 OAC Use Case Reference
16 patterns that solve the most common OAC AI Agent scenarios. Each includes copy-ready instructions.
👈
Select a use case
Click any pattern to see details
🏗️ Build Your Instructions
Fill in each section below. Copy the combined result at the bottom.

🎭 Role — Who is the agent? What domain?

📋 Task — What should the agent do?

📎 Context — KPIs, fiscal calendar, hierarchies

🔒 Constraints — Hard rules the agent must follow

📐 Output — How should responses look?

💡 Examples — 2-3 query→response pairs

📄 Live Preview

Fill in the fields above to generate your Supplemental Instructions...
🔬

Want to go deeper?

Explore all 40 techniques, 6 frameworks, 16 use cases, and the complete decision tree.

Open Full Studio →
Oracle Analytics Cloud — AI Agent Quick Start Guide | V2 R.T.C.C.O.E
Created by Ravi Bhuma · Oracle CEAL Team