Work through 12 exercises using 4 real datasets. Each exercise uses the exact instructions and sample utterances from the OAC Agents Technical Guidance — practice building Supplemental Instructions that tell the Agent how your business thinks.
OAC instance with BI Content Creator role + "Create and Edit AI Agents" permission enabled via Console → Roles and Permissions
Console → Extensions and Enrichments → Generative AI. Select a Gen AI Service under AI Assistant Features.
Upload all 4 datasets below into OAC. Right-click → Inspect → Search → Index for Assistants → Run Now.
Home Page → Create → AI Agent. Attach the dataset indicated for each exercise. Add SI in the Supplemental Instructions box.
Each exercise specifies which dataset to use. You'll create up to 4 agents, one per dataset. Download and upload these into your OAC instance.
| Dataset | Key Columns | Used In Exercises |
|---|---|---|
| SOL.xlsx | Order Priority, Product Category, Product Sub Category, Product Name, Sales, Profit, Shipping Cost, City, Customer Name, Ship Mode, Order Date | 1–6, 9 |
| Retail_Sales.xlsx | category, sales_revenue, transactions, selling_area_sqft, store_name, region | 7 |
| Sample_Superstore_-_Rate.xlsx | Sales, Shipping Rate, Segment, Category, Region, Profit, Quantity | 8 |
| GP_Sales_Locations.csv | LATITUDE, LONGITUDE, Sales, Cat | 10–12 |
Classify 'Order Priority' into 2 groups. The column has values: Low, Medium, High, Critical, Not Specified. Map them to "Regular" and "Important".
1. "Compare Average Profit for Regular and Important Orders"
2. "Revenue for Important Orders"
3. "Important vs Regular by Shipping Cost"
Agent groups Low+Medium+Not Specified as "Regular" and High+Critical as "Important" and shows comparison.
Create a product hierarchy drill-down: Product Category → Product Sub Category → Product Name.
1. "Profit by Product Category" → then say "Drill Down" → then "Drill Down" again
2. Should navigate: Category → Sub Category → Product Name
3. At Product Name level, drill down should say "most detailed level"
When users ask generic questions without specifying a metric, default to "Sales" (Revenue).
1. "Compare Cities for furniture product category" → Sales should be picked up automatically
2. "Show performance by region" → Should default to Sales
3. "Profit by Product Category" → Should use Profit (user specified it)
Auto-apply City = "Toronto" filter on every query, even when the user doesn't ask for it.
1. "Revenue by Shipping Mode" → Should auto-filter to Toronto
2. "Sales by Product Category" → Check that City: Toronto appears
Specify default visualization types: Table for categories, Area Chart for time trends.
1. "Sales by Product Categories" → Without instruction: Bar chart. With instruction: Table
2. "Sales trend by month" → Should show Area Chart
Create a shortcut for "Monthly Summary" / "Monthly Report" that auto-generates a specific table.
Month is hardcoded in this example. In production, you'd update this value when the period changes.
1. "Monthly Report" → Should produce Table with Sales, COUNT(DISTINCT Order ID), Shipping Cost filtered to Oct 2025
2. "Performance this month" → Same output
Create custom KPI calculations: Average Transaction Value (ATV) and Sales per Sq. Ft using columns from the Retail Sales dataset.
Create a new AI Agent and attach Retail_Sales.xlsx as the dataset. Index it first.
1. "ATV Per Category" → Should compute Sum(sales_revenue) / Sum(transactions)
2. "Sales per Square Feet by Region" → Should compute Sum(sales_revenue) / Sum(selling_area_sqft)
Define default aggregation rules per metric. Without this, the agent might Sum a Shipping Rate instead of Averaging it — silently producing wrong numbers.
Create a new AI Agent and attach Sample_Superstore_-_Rate.xlsx as the dataset.
1. "Shipping Rate by Segment" → Should automatically use AVG (not Sum)
2. "Sales by Category" → Should use Sum
3. "Average Sales by Region" → User said "average" — should override Sum default
Standardize ranking behavior: default to Top/Bottom 5, default metric Sales, map keywords like "Best" and "Worst".
1. "Show Top Customers" → Top 5 customers by Sales
2. "5 Worst Profitable Product Sub categories" → Bottom 5 by Profit ascending
3. "Best 3 cities" → Top 3 by Sales
Map a non-standard fiscal calendar (FY = Jun 1 – May 31) so "current quarter" uses fiscal logic, not Gregorian. Note: Update hardcoded values when your fiscal periods change.
1. "Show Revenue for Current Quarter" → Should filter to FY2026 Q2
2. "Compare Previous and Current Fiscal Years Profits" → FY 2025 vs FY 2026
Enable location-aware queries using LATITUDE and LONGITUDE columns for distance-based filtering.
Create a new AI Agent and attach GP_Sales_Locations.csv.
1. "Map the Sales" → Should show Map using LATITUDE, LONGITUDE
2. "Show Sales in Tennessee" → then "Change Chart Type to Map"
3. "Show me sales near Nashville" → Should use 50-mile default radius on Map
Add named shop coordinates via Supplemental Instructions so users can say "near Shop-ABC" and the agent resolves coordinates. In production, shop locations could also be added as a Knowledge Document (RAG).
1. "Show Sales near Shop ABC" → Should resolve to Chicago (41.8781°N, 87.6298°W), 50-mile radius, Map
2. "Show Sales within 300 miles of Shop XYZ" → Should resolve to NYC coordinates, 300-mile radius