Skip to main content

​
Overview

Earnings Transcripts is Finance Agent’s core feature for analyzing quarterly earnings calls from 2020-2025. Get word-for-word executive commentary, financial metrics, and forward guidance directly from the source that professional analysts use. Unlike generic LLMs that summarize web content, Finance Agent retrieves actual transcript chunks from your target companies and quarters, giving you precise answers with exact citations.

Word-for-Word Accuracy

Every answer is backed by actual transcript excerpts, not paraphrased summaries

Smart Planning

AI agent plans research approach before searching, ensuring comprehensive coverage

Multi-Quarter Analysis

Compare metrics across quarters automatically with parallel search

Citation Tracking

Every fact includes [TC-N] citations linking back to the exact source

​
How It Works

When you ask a question about earnings calls, Finance Agent:
  1. Plans the Search - Generates focused keyword phrases optimized for semantic search (not verbose questions)
  2. Parallel Retrieval - Searches all target quarters simultaneously using hybrid vector + keyword search
  3. Answer Generation - Synthesizes findings with precise [TC-N] citation markers
  4. Quality Evaluation - Assesses completeness and iterates if gaps are found (up to 2 iterations)

​
What You Can Ask

  • “What was Apple’s Q4 2024 revenue?”
  • “Show me Datadog’s billings for the last 3 quarters”
  • “What did Microsoft say about Azure growth in Q3 2024?”
  • “What guidance did management provide for Q1 2025?”
  • “How did executives explain the margin compression?”
  • “What did the CEO say about the competitive landscape?”
  • “What questions did analysts ask about AI investments?”
  • “How did management respond to concerns about deceleration?”
  • “What did executives say about customer acquisition costs?”
  • “Compare MSFTandMSFT and GOOGL cloud revenue growth”
  • “What did AAPLandAAPL and NVDA say about supply chain issues?”
  • “Compare operating margins for METAandMETA and SNAP”

​
Query Examples

​
Single Metric Lookup

Question: What was Datadog's Q3 2024 billings?

Answer:
Datadog reported **$748M** in billings for Q3 2024, up **21%** year-over-year [TC-1]. 
This represents an acceleration from Q2 2024's **$690M** in billings [TC-2].

**You might also ask:**
- What was Datadog's RPO (remaining performance obligations) in Q3 2024?
- How did Datadog's customer count change in Q3 2024?
- What guidance did Datadog provide for Q4 2024 billings?

Question: Show me Apple's iPhone revenue for the last 3 quarters

Answer:
Here's Apple's iPhone revenue trend:

| Quarter  | iPhone Revenue | YoY Growth |
|----------|----------------|------------|
| Q4 2024  | **$46.2B**     | **+6%** [TC-1] |
| Q3 2024  | **$39.3B**     | **-1%** [TC-4] |
| Q2 2024  | **$45.9B**     | **+0%** [TC-7] |

Management noted that Q4 2024's growth was driven by strong demand for iPhone 15 Pro models [TC-2].

​
Comparative Analysis

Question: Compare Microsoft and Google cloud revenue growth in Q3 2024

Answer:
## Cloud Revenue Comparison - Q3 2024

| Company   | Cloud Revenue | YoY Growth | QoQ Growth |
|-----------|---------------|------------|------------|
| Microsoft | **$31.8B** (Azure) | **+33%** [TC-1] | **+3%** [TC-1] |
| Google    | **$11.4B** (GCP) | **+35%** [TC-8] | **+5%** [TC-9] |

**Key Insights:**
- Both companies saw acceleration vs Q2 2024 [TC-3][TC-10]
- Microsoft cited AI workloads driving Azure growth [TC-2]
- Google highlighted new enterprise customer wins in GCP [TC-11]

​
Features

Combines vector embeddings (70%) with keyword matching (30%) for optimal recall:
  • Vector search: Captures semantic similarity (“revenue” matches “sales”, “income”)
  • Keyword search: Ensures exact terms are prioritized (“billings” won’t miss “billings”)

​
Planning-Driven Retrieval

Before searching, the agent generates:
  • Sub-questions: What specific data is needed?
  • Search queries: Dense keyword phrases optimized for embedding search
Example: 
User question: "What is Datadog's revenue per employee?"

Agent plan:
  Sub-questions:
    - What was Datadog's total revenue?
    - How many employees does Datadog have?
  
  Search queries:
    - "total revenue subscription professional services"
    - "number of employees workforce headcount"

​
Citation Format

Every answer includes [TC-N] markers that link to source transcripts:
  • [TC-1] = First transcript chunk retrieved
  • [TC-5] = Fifth transcript chunk
  • Click any citation to see the full excerpt with company, quarter, and year metadata

​
Multi-Ticker Synthesis

When comparing companies, Finance Agent:
  1. Runs parallel searches for each ticker
  2. Renumbers citations to avoid conflicts (AAPL gets TC-1 to TC-10, MSFT gets TC-11 to TC-20)
  3. Synthesizes findings into a unified comparative answer

​
Coverage

  • Time Range: 2020-2025 (coverage varies by company)
  • Companies: S&P 500 and major tech companies
  • Quarters: Q1, Q2, Q3, Q4 (fiscal quarters, varies by company fiscal year)
  • Database: PostgreSQL with pgvector for semantic search
Quarter availability varies by company. Use specific quarter references (“Q3 2024”) or relative terms (“last 3 quarters”, “latest quarter”) to get the most recent data available.

​
Technical Details

​
Architecture

  • Chunk Size: 1000 characters with 200-character overlap
  • Embedding Model: all-MiniLM-L6-v2 (384 dimensions)
  • Chunks per Query: Up to 15 per quarter, 20 shown to LLM per iteration
  • Max Iterations: 2 (initial generation + 1 refinement)
  • Quality Threshold: 85% confidence to stop early

​
Retrieval Flow

User Question
  ↓
Plan Generation (sub-questions + search queries)
  ↓
Parallel Hybrid Search (vector + keyword across all target quarters)
  ↓
Answer Generation (with [TC-N] citations)
  ↓
Quality Evaluation
  ↓
[If gaps found] → Replan with additional queries → Search again
  ↓
Final Answer

​
Best Practices

1

Use Ticker Symbols

Always include $TICKER format for companies (e.g. $AAPL, $MSFT). This ensures accurate company identification.
2

Specify Time Periods

Be specific about quarters:
  • “Q3 2024” = exact quarter
  • “last 3 quarters” = most recent 3 available
  • “latest quarter” = most recent available
3

Request Specific Metrics

The more specific your question, the better:
  • Good: “What was RPO growth in Q3 2024?”
  • Less good: “Tell me about performance”
4

Use Follow-Up Suggestions

Every answer ends with 3 suggested follow-up questions to help you explore related topics.

​
Limitations

  • No Real-Time Data: Transcripts are from published earnings calls, typically 1-2 quarters behind current date
  • Fiscal Year Variations: Companies define fiscal years differently (Apple’s Q4 is Oct-Dec, Microsoft’s Q4 is Jul-Sep)
  • Coverage Gaps: Not all companies have complete historical data back to 2020
  • No Stock Prices: Focus is on earnings call commentary, not market data
  • SEC Filings - For annual 10-K data (balance sheets, CEO compensation, audited financials)
  • Real-Time News - For breaking developments not yet in earnings calls
  • Financial Screener - For screening stocks by fundamentals (coming soon)

Build docs developers (and LLMs) love

Get started for freeTalk to us