Kolton Dupey · 2025-07-23 · 10 min read
Automating Earnings Call Intelligence with Python, GPT-4, and Twitter Integration
Automation
This post walks through an end-to-end automation pipeline I built that
scrapes earnings call transcripts from MarketBeat, summarizes them using
OpenAI's GPT-4, extracts sentiment and strategic insights, and
optionally posts the result to Twitter. The goal is to streamline
financial content production while maintaining a high level of
analytical depth and structure.
This pipeline is modular, highly extensible, and designed for analysts,
investors, and creators who want to scale earnings call coverage
efficiently.
📊 Overview
The pipeline is built from two primary components:
- 🛰️ Transcript Monitor and Processor
- 🧠 Transcript Summary Generator with Twitter Integration
These components work independently and together. One prepares the data
efficiently; the other distills and distributes insights.
🛰️ Transcript Monitor and Processor
The first module handles all the mechanics of transcript discovery,
parsing, optimization, and storage. It performs the following steps on a
scheduled loop:
⏰ Monitor for New Transcripts
Using
requests and BeautifulSoup, the script scrapes MarketBeat's
earnings transcript listings. It pulls the ticker, logo, published date,
and transcript URL. Transcripts are only processed if they haven't
already been seen, and checks run during defined market hours (6 AM to 8
PM ET).🧩 Parse and Structure Conversations
Once a transcript URL is fetched, the HTML content is parsed into
speaker-message pairs. The script identifies speakers by class names and
div structure and intelligently handles operator sections that contain
multiple embedded Q&A exchanges. This parsing logic is tuned to
MarketBeat's structure but can be adapted to other sources with minimal
changes.
🧹 Clean and Optimize Transcript Text
Raw earnings call transcripts are verbose. To reduce API token usage and
improve the quality of summarization, I created a transcript optimizer
that performs several transformations:
- Removes greetings, filler phrases, and redundant language
- Normalizes spacing and punctuation
- Capitalizes cleaned statements
- Consolidates the transcript into a single continuous message block
Both the full and optimized transcripts are saved as JSON to a
timestamped directory for traceability and reuse.
📉 Analyze Token Usage and Cost
Using OpenAI's
tiktoken library, the script counts the tokens required
for both full and optimized payloads. This is used to estimate GPT-4
cost per transcript and quantify the benefits of cleaning. In testing,
optimization typically reduces token count by 25 to 40 percent.🧠 Transcript Summary Generator
The second module is responsible for generating a structured summary
from the optimized transcript and optionally posting it to Twitter. It
is flexible enough to handle manual input, individual files, or batch
processing.
📝 Structured Prompt Engineering
The summary request sent to GPT-4 uses a strict format that includes:
- Ticker and headline
- Overall sentiment (e.g., Neutral, Slightly Positive)
- Management guidance (1 sentence)
- A six-sentence transcript summary
- Three key pros
- Three key cons
This format balances clarity, structure, and nuance. It enables
cross-platform reuse and ensures consistency regardless of the company
or industry.
🔍 Summary Generation
The OpenAI API call is made using
gpt-4-turbo-preview, with a moderate
temperature for creativity. The summary includes forward-looking
statements, KPIs, risks, and strategic context. It's designed to help
readers understand the core of the earnings call without needing to read
the full transcript.🐦 Optional Twitter Posting
If enabled, the script uses a custom
tweet_poster module to
authenticate with the Twitter API and post the summary as a single
tweet. It handles rate limiting, client initialization, and error
capture. Delays can be configured between tweets during batch runs.🔄 Batch Processing and Flexibility
The summary analyzer supports multiple workflows:
- Manual input from clipboard or terminal
- JSON file path input
- Entire folder processing with optional Twitter output
This flexibility makes the system usable for live monitoring, editorial
work, or retrospective coverage.
🛠️ Technologies Used
- Python 3
- BeautifulSoup and requests for web scraping
- re and tiktoken for text normalization and token estimation
- OpenAI GPT-4 Turbo for summarization
- JSON and datetime for structured storage
- Twitter API via tweepy or requests
📣 Example Output
🚨 $NKE Nike Inc. 03/20/2025 Earnings 🚨
Sentiment: Slightly Negative
Guidance: Management expects Q4 revenues to be down
in the mid-teens range and gross margins to decrease approximately 400
to 500 basis-points.
Transcript Summary: Nike Inc. reported a challenging
fiscal 2025 3rd-quarter, with revenues down 9% on a reported basis and
7% on a currency-neutral basis, highlighting the difficulties faced in
various markets, particularly in Greater China. Despite these
headwinds, the company is aggressively pushing forward with its 'Win
Now' strategy, focusing on reigniting brand momentum through sport,
stabilizing the business, and repositioning its product portfolio
towards performance sports. The leadership emphasized efforts to
diversify and innovate within their product lines, notably in running
and training, and to reduce reliance on classic franchises that have
seen decelerated growth. Additionally, Nike is undertaking significant
measures to clean up the marketplace, particularly in digital and
wholesale channels, aiming for a healthier, full-price business model.
The company also highlighted extensive brand and marketing activities
during key sports events to reconnect with consumers and drive
engagement.
Pros:
- Aggressive strategy to reignite brand momentum through sport.
- Focus on performance sports and product diversification away from declining classic franchises.
- Significant brand and marketing activities demonstrating strong engagement in key sports events.
Cons:
- Revenue down 9% on a reported basis and 7% on a currency-neutral basis.
- Gross margins declined 330 basis-points due to higher markdowns and inventory challenges.
- Digital and Greater China market challenges, with aggressive but necessary marketplace clean-up actions impacting short-term performance.
#earnings #earningsreport
✅ Conclusion
This pipeline automates what is traditionally a time-intensive process.
From scraping and cleaning transcripts to summarizing and sharing key
insights, the workflow reduces manual effort while maintaining fidelity
and structure. The result is a scalable, cost-effective approach to
earnings call analysis that can power newsletters, investor dashboards,
or automated market commentary.
If you're building something similar or want access to the source code,
feel free to reach out.
– Kolton Dupey
Automation Developer | Salesforce & API Integrator
Want this built for your business?
This post comes from a real engagement. Dealing with something similar? Book a 15-minute call and I'll tell you what it would take.
Related posts
Automation
Google Maps Lead Generation: Automate Business Discovery with n8n (Step-by-Step Guide)
This workflow searches Google Maps for businesses in your niche, resolves each listing to its website, scrapes the site with polite rate-limits, extracts valid emails + company info, de-duplicates, and appends the results to Google Sheets—perfect for outbound campaigns.
7 min read · 2025-08-31
Read post →
Automation
The Ultimate Guide to Enterprise Document Automation: A Deep Dive into Advanced DocuSign Integration
Explore a sophisticated Python-based DocuSign automation system with over 2,400 lines of production-ready code that transforms manual document workflows into highly efficient, scalable operations for enterprise environments.
15 min read · 2025-08-22
Read post →
Automation
From Click to CRM - Automating Lead Flow from WhatConverts to Salesforce with AI-Powered Precision
Discover how to build an intelligent automation system that captures leads from WhatConverts and seamlessly integrates them into Salesforce with AI-enhanced data processing using Zapier.
5 min read · 2025-07-14
Read post →