How to build an Earnings Call Insight and Sentiment Analyzer

TL;DR

• What it does: Summarizes earnings calls, highlights themes and KPIs, and tags sentiment per speaker.

• Who it’s for: Investment analysts, PMs, and IR teams needing fast post-call insights.

• Time to value: ~30–45 minutes to configure; minutes per new call.

• Output: A structured PDF/markdown brief with themes, KPI table, sentiment dashboard, and key quotes.

Common Pain Points of Analyzing Earnings Calls

• Manual parsing takes hours and misses tonal cues.

• Inconsistent notes and formats across analysts.

• No quick way to compare sentiment across speakers/quarters.

• Quotes lack source context for compliance-ready sharing.

What the Agent Delivers

• Theme & KPI extraction with cited references.

• Sentiment by speaker (positive, neutral, negative) with per-section scores.

• Key quotes with timestamps and speaker attribution.

• A shareable one-pager plus CSV/JSON exports for dashboards.

Step-by-Step Build (StackAI nodes)

1) Input: Upload Earnings Call Audio (Audio)

What it does: Lets you upload the call recording, either from a file (e.g., MP3/WAV/MP4) or a URL. You can also choose the transcription provider (e.g., Deepgram, Whisper) and model/submodel best suited for accuracy.

Goal: Capture the raw audio source so it can be transcribed and prepared for analysis.

2) Extract Themes, Sentiment and KPIs (LLM)

What it does: This node is an AI agent designed to analyze the transcript of an earnings call and produce a structured, one-page summary with key insights.

Model: GPT-5 (OpenAI) - Best at handling long transcripts.

Instructions

You are an AI agent specialized in analyzing earnings call transcripts. Your tasks include:


- Summarizing main themes and discussion topics clearly and concisely.

- Extracting all KPIs mentioned (Revenue, EBITDA, EPS, Margins, Guidance, etc.) with their values, percentage changes, and time periods.

- Returning a structured output suitable for a one-page brief, formatted as follows:

Prompt

You will be provided with a transcript of an earnings call. Your task is to:


- Summarize the main topics discussed.

- Assign sentiment (Positive, Neutral, Negative) to each speaker intervention and provide a confidence score if possible.

- Extract all KPIs mentioned (e.g., Revenue, Margins, EPS, Guidance) and present them in a structured table.

- Highlight 3–5 key quotes that reflect management’s tone, guidance, or outlook.

- Produce a concise one-page summary with the following sections: Executive Summary, Themes, KPIs, Sentiment Dashboard, and Quotes.


Ensure the output adheres to the specified schema.


<Transcript>

{audio2text-0}

</Transcript>

3) Structured Summary and Sentiment Dashboard (Output)

Purpose: Displays the structured summary and sentiment dashboard generated by the LLM.