GBS Intelligence Agent

Automated competitor & client intelligence for GBS / Finance Transformation consulting. Weekly newsletter + interactive dashboard — fully automated via cron.

Built by Moritz Richter · Business Consulting Finance

---

What it does

TLCA (Thought Leadership Collection Agent, the internal name used throughout the code) monitors public sources for strategic signals relevant to Global Business Services (GBS), Global Capability Centers (GCC), and Agentic AI — and delivers a curated weekly intelligence brief automatically.

Google Alerts (RSS feeds)
        ↓
intel_ingest_stdlib.py     →  SQLite database
        ↓
summarize_new_links.py     →  Claude Haiku (relevance scoring + summarization)
        ↓
weekly_newsletter_output.py →  HTML newsletter (score ≥ 3 only)
generate_dashboard.py       →  Intelligence Explorer (interactive HTML)
send_newsletter.py          →  Gmail → inbox

Every Monday morning, automatically:

• Ingests 50+ RSS feeds across competitors and client companies
• Scores each article 1–3 for strategic relevance using Claude
• Renders a newsletter with only the highest-signal articles
• Builds an interactive dashboard for deeper exploration
• Sends the newsletter by email
• Archives output by calendar week

---

Intelligence Explorer

The dashboard (intelligence_explorer.html) is a standalone dark-theme HTML file — no server needed, just open in a browser.

Features:

• Card grid with competitor + client signal separation
• Competitor cards grouped by cluster: MBB · Big4 · Accenture
• Filter by type, topic tag (GBS / GCC / Agentic AI / Operating Model), calendar week
• Full-text search across titles, summaries, companies
• Click-to-expand modal with full summary + source link
• Relevance badges: ★ HIGH · MED · LOW
• Keyboard shortcuts: / search · Esc reset

---

Relevance Scoring

Each article is scored by Claude before summarizing:

Score	Label	Meaning	Newsletter
3	★ HIGH	Direct strategic signal: new offering, GBS/AI publication, C-suite move, M&A	✅
2	MED	Indirect signal: market commentary, trend piece	Explorer only
1	LOW	Weak signal: generic news	Explorer only
0	SKIP	Job post, stock tracker, ETF filing, economic indicator	Discarded

---

Setup

1. Clone and install

git clone https://github.com/YOUR_USERNAME/tlca.git
cd tlca
pip3 install -r requirements.txt
chmod +x run_weekly.sh

2. Configure API keys

export ANTHROPIC_API_KEY="sk-ant-..."          # https://console.anthropic.com
export GMAIL_USER="your@gmail.com"
export GMAIL_APP_PASSWORD="xxxx xxxx xxxx xxxx"  # https://myaccount.google.com/apppasswords
export INTEL_RECIPIENTS="recipient@email.com"

Add to ~/.zshrc or ~/.bashrc for persistence.

The weekly runner stops on any failed pipeline stage or feed fetch, so it will not send a partial newsletter. The edition counter is committed only after the email is sent successfully. Set INTEL_FAIL_ON_FEED_ERROR=0 only when a partial run is acceptable.

3. Configure feeds

cp sources.example.json sources.json

Edit sources.json:

1. Go to google.com/alerts
2. Create alerts with Deliver to: RSS feed
3. Copy feed URLs into sources.json
4. Add company names to competitor_sources or client_sources

4. Run

bash run_weekly.sh

Output in output/CW_YYYY_WW/:

• newsletter_full.html — full email (open in browser to preview)
• newsletter_block.html — article blocks only
• newsletter_block.txt — plain text
• client_signals.html — client signals only
• intelligence_explorer.html — interactive dashboard

5. Schedule (every Monday 07:30)

(crontab -l 2>/dev/null; echo "30 5 * * 1 cd /path/to/tlca && bash run_weekly.sh >> logs/weekly.log 2>&1") | crontab -

Adjust UTC offset for your timezone.

---

Customization

Add a company to track

In sources.json:

{
  "source": "CompanyName_Finance",
  "url": "https://www.google.com/alerts/feeds/YOUR_ID/FEED_ID"
}

Add "CompanyName" to competitor_sources or client_sources.

Add blocked domains (noise filter)

"blocked_domains": ["gurufocus.com", "marketbeat.com"]

Change newsletter score threshold

export INTEL_MIN_SCORE=2   # include MED signals (default: 3)

Add your photo to the dashboard footer

cp your-photo.jpg photo.jpg
export INTEL_PHOTO_PATH="photo.jpg"

---

Environment Variables

Variable	Default	Description
ANTHROPIC_API_KEY	—	Required. Claude API key
GMAIL_USER	your@gmail.com	Gmail sender
GMAIL_APP_PASSWORD	—	Gmail App Password
INTEL_RECIPIENTS	recipient@email.com	Comma-separated recipients
INTEL_DB_PATH	intel.db	SQLite database
INTEL_SOURCES_PATH	sources.json	Feed and tagging configuration
INTEL_FAIL_ON_FEED_ERROR	1	Stop before delivery when a feed fails
INTEL_WINDOW_DAYS	7	Newsletter lookback window
INTEL_MIN_SCORE	3	Minimum relevance score for newsletter
INTEL_MAX_SUMMARIZE	30	Max articles to summarize per run
INTEL_DASHBOARD_WEEKS	4	Explorer lookback in weeks
CLAUDE_MODEL	claude-haiku-4-5-20251001	Claude model
INTEL_PHOTO_PATH	photo.jpg	Footer photo path

---

Stack

• Python 3.10+ — no heavy ML dependencies
• SQLite — lightweight local database
• Claude Haiku (Anthropic) — fast, cheap summarization + scoring
• Gmail SMTP — newsletter delivery
• Google Alerts — RSS feed source (free)
• Vanilla HTML/JS — dashboard, no framework needed

---

Suggested Alert Queries

Competitor monitoring:

site:mckinsey.com ("operating model" OR "global business services" OR "agentic ai")
site:deloitte.com ("finance transformation" OR "shared services" OR "GBS")

Client monitoring:

"CompanyName" "finance transformation" OR "CFO" OR "operating model"
"CompanyName" "shared services" OR "GBS" OR "global business services"

---

License

MIT — free to use, modify, and distribute.

---

Built as a personal project to automate competitive intelligence for GBS and finance transformation consulting. Contributions welcome.