Media Watcher · Case study

Finding a new product inside an existing one.

Recognised a wider use for AML Watcher’s media technology and shaped it into a standalone intelligence product for monitoring coverage across languages.

My role: Technical Product Manager – Sole Product Owner
Company: Programmers Force
Period: Jun 2024 – Jun 2025

To a working prototype: 2 weeks; Reused the ingestion and NLP foundations built for AML Watcher.
Languages supported: 80+; Language coverage of the media analysis pipeline.
Levels of sentiment: 3; Separate analysis for an entity, a case and an entire article.

01 /

Context

While building AML Watcher's adverse media screening, I recognised that the underlying technology (real-time news ingestion, NLP entity extraction, multi-level sentiment analysis, and context-aware profiling) had applications far beyond compliance. PR teams, journalists, financial analysts, and government agencies all need real-time media intelligence. No AML product offered this standalone, and existing media monitoring tools lacked the analytical depth our infrastructure already provided.

02 /

The Problem

The adverse media capability was locked inside a compliance product. Its value was constrained to AML use cases, even though the technology could serve a much broader market. The opportunity was to repurpose the ingestion pipeline, NLP models, and sentiment architecture into a dedicated platform for any industry that needs to understand what the world is saying.

03 /

My Approach

I proposed, named, and championed the spin-off. The name "Media Watcher" captured the concept immediately. The architecture decisions I had made during the AML Watcher build were deliberately extensible, so the core pipeline could be repurposed without starting from zero. I delivered a functional prototype in two weeks, not a wireframe but a working system with source ingestion, entity extraction, sentiment scoring, and a live dashboard.

04 /

What Was Built

Three-Tier Sentiment Analysis Pipeline

The intellectual core. Raw text flows through language detection (80+ languages), NLP entity extraction, context-aware role identification, sentiment scoring via LLMs, and category assignment across 415+ categories. Sentiment operates at three levels: Entity-Level, where each person or organisation is scored independently; Case-Level, where scores are aggregated across a topic; and News-Level, which captures the overall article tone. This remains unmatched by competitors in either AML or media monitoring.

Global Source Coverage

Led integration research for 100,000+ sources across 230+ countries in 80+ languages. These include news outlets, Facebook, Twitter/X, Reddit, YouTube, TikTok, podcasts, broadcast channels, forums, and blogs. Defined source ingestion architecture, fetch frequencies, data normalisation, and custom source addition framework.

Intelligence Products

Three modules built on the sentiment pipeline. Media Intelligence handles narrative tracking, influence scoring, and influencer identification. Consumer Intelligence covers audience segmentation and behaviour tracking. Business Intelligence provides competitive landscape analysis, share-of-voice metrics, and strategic decision support.

1B+ Article Archive

More than 1 billion news articles spanning 25 years, each processed through the full AI pipeline including entity extraction, sentiment scoring, category assignment, and contextual tagging. A structured, searchable intelligence database for historical analysis.

Sub-200ms Alert System

Configurable triggers for keywords, sentiment shifts, volume spikes, or custom conditions. Delivered via Slack, WhatsApp, email, webhooks, and RSS feeds.

05 /

Impact

Zero-to-one product conceived, named, and delivered from existing technology
Working prototype in 2 weeks
100,000+ sources across 230+ countries in 80+ languages
Three-tier sentiment analysis unmatched by competitors
1 billion+ articles processed through AI pipeline
Sub-200ms alert delivery
Serves compliance teams, journalists, analysts, and policymakers in 20+ countries

06 /

Reflection

Media Watcher validated something I believe deeply: the best product ideas come from building, not brainstorming. I found the opportunity because I was deep in the technical work of adverse media screening and recognised its broader potential. The two-week prototype was possible because the architecture decisions I'd made months earlier were designed for extensibility. Zero-to-one products emerge from infrastructure you've already built, if you're paying attention.