Watsons
Watsons are automated product analysts that replaces manual data pulls with AI-driven analysis.
Core trick
Agents pull product data, run 87 predefined analyses, and rank anomalies with ranking weights tuned by PM feedback.
Description
Watsons are automated product analysts that replaces manual data pulls with AI-driven analysis. A PM selects from a catalog of 87 analyses across 13 categories (retention, monetization, engagement, etc.) and Watsons run them end to end: pulls the data, applies the right statistical methodology, generates charts, and surfaces a ranked feed of insights worth attention, not raw numbers.
They also continuously monitor KPIs for anomalies and triggers qualitative research (competitor analysis, voice-of-customer, unmet-need discovery) when a quantitative finding warrants explanation. An Insight Engine scores, deduplicates, and ranks everything into one prioritized feed so a startup can open it Monday morning and immediately see what needs attention.
A reinforcement learning loop captures PM feedback on every surfaced insight and continuously tunes the ranking weights so the system gets better at predicting what deserves attention over time.
README excerpt
# Watsons: an army of automated Product Analysts An AI-powered product analytics system that transforms raw product data into a prioritized feed of actionable business insights. Watsons orchestrates [Gemini Managed Agents](https://ai.google.dev/gemini-api/docs/agents) for quantitative analysis and the [Gemini Deep Research Agent](https://ai.google.dev/gemini-api/docs/interactions/deep-research) for qualitative user and market research, delivering a unified "what needs attention" stream for product managers. --- ## How It Works Watsons replaces the manual analyst workflow of pulling data, computing metrics, and interpreting results. A PM selects an analysis (or an entire category), and Watsons runs it end-to-end: pulls data, executes the methodology, produces charts, and returns **ranked insights worth attention** rather than raw numbers. The system operates in three modes: | Mode | Description | |------|-------------| | **Single Analysis** | Run one specific analysis (e.g., cohort retention curves) with configurable parameters | | **Group Run** | Run an entire category of analyses (e.g., all 12 Monetization analyses) with a shared data snapshot | | **Continuous Monitoring** | Scheduled KPI anomaly detection that alerts only on genuine statistical breaches | Additionally, a qualitative research layer uses the Gemini Deep Research Agent to answer "why" questions — what users want, what they're saying, and which competitors they're switching to. --- ## Architecture ```