Introduction
Learn how to craft effective prompts for Lume AI to achieve accurate and efficient results in financial recasting, data transformation, and workflow automation. This guide covers key principles and practical examples to help you get the most out of AI-driven edits.Key Principles
1. Be Specific & Clear
- State exactly what you want to achieve
- Include all necessary parameters
- Avoid ambiguous or vague instructions
- ❌ “Put ‘John Doe’”
- ✅ “Place ‘John Doe’ as the default value for this target field.”
- ❌ “Make this sound better.”
- ✅ “Rewrite this loan summary to be more professional and concise, while keeping key financial metrics.”
2. Contextual Awareness
Provide Sufficient Background
- AI edits are more effective when given relevant context
- Include details like industry jargon, audience type, or document purpose
- A huge lever is providing an example
- ❌ “Split the address”
- ✅ “We are provided with a full address containing Street, City, State, and Zip. Please extract the street name. The addresses may vary in format, so please account for these differences.”
Leverage Data-Driven Insights
- If applicable, reference specific data points or sources for accuracy
3. Iterative Refinement
Break Down Complex Edits
Complex edits should be broken down into clear, structured steps within a single prompt. Example: “Please process this list of lenders by:- Extracting the first name
- Adding a unique id to the end of each name
- Including any additional name-related fields present”
Feedback & Adjustments
- If an AI edit isn’t perfect, refine your prompt
- Use structured feedback to improve results
4. Structural & Formatting Enhancements
Use Formatting Directives
- Specify whether edits should include bullet points, tables, or numbered lists
Ensure Logical Flow
- Indicate if content should be reordered for coherence
Best Practices
Do:
- Start with a clear objective
- Provide sample data when possible
- Specify output format requirements
- Break complex tasks into steps
Don’t:
- Use vague instructions
- Assume context is understood
- Request multiple transformations in one prompt
- Skip providing examples for complex tasks
Troubleshooting
If you’re not getting desired results:- Add more specific examples
- Break down the request into smaller steps
- Clarify any assumptions
- Include edge cases you want to handle