Lyra AI Assistant is designed to understand natural language, but how you ask a question directly affects the quality, precision, and usefulness of the results.
This guide shows you how to structure prompts that consistently produce clear tables, actionable insights, and meaningful analysis — without needing technical query syntax.
The Mental Model: Think Like an Analyst
The best Lyra prompts mirror how an analyst would frame a request:
What am I analyzing?
Over what time period?
Where does it apply?
What constraints or thresholds matter?
How should results be presented?
You don’t need to include all five every time — but the more context you give, the better Lyra can optimize the response.
Core Prompt Building Blocks
1. Subject (What You’re Analyzing)
Start with a clear subject:
Borrowers
Lenders
Properties
Transactions
Markets or geographies
Examples
“Show mortgage transactions…”
“Analyze borrower activity…”
“Who are the top private lenders…”
2. Timeframe (Almost Always Include This)
Timeframes dramatically improve precision and performance.
You can use:
Relative: “last 12 months”, “last 90 days”
Fixed: “Q1 2024”, “2023”
Ranges: “January 1, 2023 to December 31, 2024”
Examples
“in the last 12 months”
“during Q3 2024”
“since 2022”
If you omit a timeframe, Lyra will infer one — but explicit is better.
3. Location (When Geography Matters)
Lyra understands locations naturally:
States
Counties
Cities
MSAs
ZIP codes
Radius searches (“within 5 miles of…”)
Examples
“in Harris County, TX”
“within the Phoenix MSA”
“near 123 Main St within 10 miles”
You do not need county codes or FIPS values.
4. Filters & Thresholds
Filters narrow results and surface meaningful patterns.
Common filters include:
Loan amount (“over $1M”)
Deal count (“at least 5 loans”)
Property type (“multifamily”, “commercial”)
Status (“active”, “unreleased mortgages”)
Examples
“loan amount greater than $500k”
“borrowers with 3+ active mortgages”
“properties with at least 10 residential units”
5. Output Control (Optional but Powerful)
You can explicitly request how results are returned.
Lyra supports:
Tables → structured comparison
Charts → trends and distributions
Lists → ranked or filtered entities
CSV downloads → offline analysis
Examples
“Show this as a table with columns for…”
“Visualize this in a bar chart”
“Export results as a CSV”
If you don’t specify, Lyra chooses a sensible default.
Example: Weak vs Strong Prompts
❌ Weak Prompt
“Show me lender activity.”
Too broad — unclear subject, timeframe, or geography.
✅ Strong Prompt
“Show private lending mortgage activity in the Chicago MSA over the last 12 months, grouped by lender and sorted by loan count.”
Clear subject, timeframe, location, grouping, and sorting.
Controlling Sorting, Grouping, and Limits
You can control result shape using natural language.
Sorting
“largest first”
“newest first”
“sorted by loan count descending”
Grouping
“group by lender”
“summarize by month”
“break down by county”
Limits
“top 25”
“limit to 50 results”
“show only borrowers with 5+ deals”
Asking for Links and Context
Lyra automatically includes:
Clickable company profile links
Clickable property profile links
You can also ask for:
“Include contact information if available”
“Show recent lenders used”
“Include most recent transaction date”
Using Follow-Up Questions (Multi-Step Analysis)
Lyra is designed for conversation, not one-shot queries.
You can build analysis step by step:
“Show me the top 50 private lenders in Florida by loan volume in 2024.”
“From those, which ones focus mostly on bridge loans?”
“For the top 5, show borrower concentration and any negative signals.”
Lyra maintains context and refines results without re-asking everything.
Common Mistakes to Avoid
❌ Don’t Guess Field Names
Avoid technical terms like consideration_amount or grantor_role.
Instead, say:
“loan amount”
“borrower”
“lender”
“transaction date”
❌ Don’t Overspecify
You don’t need:
County codes
Exact schema names
Database joins
Lyra handles normalization automatically.
❌ Don’t Combine Unrelated Requests
Break complex analysis into steps instead of one giant prompt.
❌ Don’t Ask for Impossible Data
Lyra can only analyze recorded data.
For example, future maturities require a recorded maturity date.
When Lyra Asks a Clarifying Question
If your request is ambiguous, Lyra may respond with options:
“I found 3 companies matching ‘ABC Properties’. Which one should I use?”
Just reply with:
“Use #1”
“The one active in Philadelphia”
This improves accuracy and avoids incorrect assumptions.
Quick Prompt Starters (Copy & Customize)
“Show me the top [number] private lenders in [location] over the last [timeframe].”
“Find borrowers in [market] with loan amounts over $[amount] and upcoming maturities.”
“Analyze wallet share for my top borrowers over the last [timeframe].”
“Compare lending activity for [company] versus competitors in [location].”
What to Read Next
Prompt Playbooks by Persona → Copy role-specific templates for underwriting, sales, and leadership
Agentic Workflows & Real-World Use Cases → See multi-step analysis in action
If you ever get stuck, ask Lyra:
“How should I phrase this?”
or
“What’s the best way to analyze this data?”
Lyra is built to help you refine the question — not just answer it.