News Discovery
AI-Curated Market Intelligence
Schedule
Mon/Thu
Sources
6 source types
Articles/Cycle
200+
Dedup
3-layer intelligence
01
Overview
The most comprehensive news intelligence agent in the platform. Scans 6 independent source types — real-time web search, Google results, and 19 curated RSS feeds — to discover 200+ articles per cycle. A 3-layer deduplication system eliminates noise, then each surviving article undergoes full-text extraction and deep relevance scoring against a proprietary taxonomy covering both industry/portfolio and firm/practice dimensions. Consultants receive only the highest-signal articles matched to their specific coverage areas. The dual-scoring model ensures that articles are both commercially relevant to the companies an MD covers and strategically aligned with active BCG practice areas — eliminating the keyword-match noise that plagues simpler systems. A two-model strategy uses fast, cost-efficient filtering for the initial volume, reserving deep analysis for articles that survive deduplication. Results are automatically routed to the BD Radar briefing with per-MD relevance matching, ensuring every Managing Director sees exactly the intelligence that matters to their portfolio.
02
How It Works
A 4-stage pipeline using a multi-model strategy to balance speed and depth. High-volume filtering handles noise removal at scale; deep scoring is reserved for articles that survive deduplication.
1
Multi-source discovery
Parallel search across real-time web, Google, and 19 curated RSS feeds for tracked company news
2
Intelligent deduplication
3-layer noise removal: URL normalization, historical record check, and AI-powered semantic dedup
3
Full-text extraction and scoring
Deep article extraction with multi-dimensional relevance scoring across industry and practice taxonomies
4
Intelligence delivery
Scored articles categorized and routed to the intelligence hub for BD Radar integration
External APIs
Perplexity Sonar Pro
Serper (Google Search API)
Firecrawl (full-text extraction)
Claude
Groq
Airtable
03
The Manual Alternative
Without this agent, a consultant or research analyst would need to perform the following research manually for every intelligence cycle:
Research TaskManual EffortAutomated
Scan 6 source types for company news2-3 hours< 5 minutes
Deduplicate articles across sources45 minutesAutomatic (3-layer)
Extract full article text for analysis1-2 hoursAutomatic (Firecrawl)
Score relevance per Managing DirectorSubjective / skippedDual-taxonomy AI scoring
Route articles to the right recipients30 minutesAutomatic per-MD matching
Total manual effort per cycle: 5-7 hours of analyst time. The News Discovery agent completes the same work in under 5 minutes with greater depth, consistency, and zero coverage gaps.