Report Copilot: Prompting Patterns and Scenarios
14. Report Copilot: Prompting Patterns and Scenarios
14.1 Effective Prompts
Good prompts are clear, specific, and include business context. Here are common patterns:
14.2 Scenario 1: Summarization (One Row Per Group)
Prompt: "Count active claims by peril type. Group by peril. Include only open claims."
Why it works: Specifies the grouping field, the grain (one row per peril), and the filter (open claims).
14.3 Scenario 2: Drill-Down with Joins
Prompt: "For each policy, show the policy number, insured name, and total premium. Include only in-force policies."
Why it works: Names the output columns, the anchor table (policies), and implies a join to insureds for readable names.
14.4 Scenario 3: Top-N Ranking
Prompt: "Top 10 insureds by total claims count. Sort from highest to lowest."
Why it works: Specifies rank, metric, sort order, and the entity of interest.
14.5 Scenario 4: Time Series
Prompt: "Count new claims by month for the last 12 months. Group by month."
Why it works: Specifies the time grain, date range, and the metric.
14.6 Scenario 5: Multi-Condition Filtering
Prompt: "Show all claims with loss amount greater than $50,000, claim status = Open, and loss date in 2024."
Why it works: Lists all conditions explicitly and avoids ambiguous phrasing like "big claims."
14.7 Tips for Better Results
Be explicit about columns - Name which fields should appear in the output.
Specify aggregations - "Count," "Sum," "Average" are clearer than "analyze."
Mention filters upfront - "Only active policies" is better than relying on defaults.
Use domain terminology - "Policies," "claims," "premiums," etc., are recognized terms.