An AI-powered product analytics tool that combines a visual adoption dashboard with natural language PM insights and a prioritization recommendation — because charts show you what happened, but PMs need to know what to do next.
Product dashboards are everywhere. Actionable insights are not.
A PM staring at a feature adoption bar chart knows that feature_B has 93% adoption. What they don't know is whether that's good or bad, why the 7% who haven't adopted it matter, or what to do about it before next quarter's planning session.
The bottleneck in product analytics isn't data — it's the synthesis work between "here are the charts" and "here is what we should build next." This tool bridges that gap with an AI layer that reads the same metrics a PM would review and produces a specific, data-grounded prioritization recommendation in plain language.
- Product Managers preparing for quarterly planning who need a fast adoption health check
- Growth PMs investigating feature engagement and churn correlation
- Startup PMs without a dedicated data analyst who need self-serve adoption insights
- Enterprise PMs who want to pressure-test their prioritization with data before a roadmap review
Upload three CSVs (users, events, subscriptions) and get:
| Chart | What it shows |
|---|---|
| Feature Adoption Rate | % of users who have used each feature, color-coded green/yellow/red |
| Adoption vs Engagement Quadrant | 4-quadrant scatter plot: Core Features / Hidden Gems / Broad but Shallow / Underperforming |
| Cohort Retention Curves | Retention rate by signup month — reveals whether your product is improving for new users |
| Churn by Feature | Churn rate for users who adopted vs. never adopted each feature — identifies retention drivers |
| Output | What it tells you |
|---|---|
| Executive Summary | 2-3 sentence product adoption health snapshot |
| Key Insights | 4-5 data-grounded observations including non-obvious patterns |
| Warning Signals | Features or cohorts showing concerning trends |
| Q2 Priority Recommendation | Specific, opinionated recommendation referencing actual feature names and metrics |
| Quick Wins | 3 actions the team can take in the next 30 days |
Color-coded bar chart showing which features are thriving (green >60%), need attention (yellow 30-60%), or are underperforming (red <30%).
Every feature mapped to one of four strategic zones. Hidden gems (low adoption, high engagement) and underperformers (low adoption, low engagement) are immediately visible.
Retention curves by signup month reveal whether product improvements are actually working for new users — invisible in aggregate metrics.
Side-by-side churn rates for users who adopted vs. skipped each feature. The biggest gaps are your highest-ROI retention investments.
One click generates a full PM analysis: executive summary, key insights, warning signals, and a specific Q2 prioritization recommendation grounded in the actual metrics.
The adoption vs. engagement scatter plot is the core analytical view. Four quadrants, four strategies:
High Engagement │ HIDDEN GEMS │ CORE FEATURES
│ Low adoption, │ High adoption,
│ high engagement. │ high engagement.
│ Invest in │ Protect and
│ discoverability. │ deepen.
────────────────┼──────────────────────┼────────────────
Low Engagement │ UNDERPERFORMING │ BROAD BUT SHALLOW
│ Low adoption, │ High adoption,
│ low engagement. │ low engagement.
│ Investigate before │ UX or value gap —
│ investing. │ users don't return.
│ Low Adoption │ High Adoption
The AI recommendation synthesizes quadrant position with churn correlation to produce a prioritized hypothesis for the PM.
| Component | Technology | Reason |
|---|---|---|
| UI | Streamlit | Fast to ship; supports file upload and interactive charts |
| Charts | Plotly | Interactive, professional-grade visualizations |
| Data | Pandas | Standard data manipulation and aggregation |
| LLM | Claude Sonnet (Anthropic) | Strong analytical reasoning; reliable structured output |
| Dataset | Kaggle public dataset | Real-world SaaS data structure (users / events / subscriptions) |
| Language | Python 3.11 | Consistent with portfolio |
- Python 3.11+
- Anthropic API key
git clone https://github.com/AasthaSanghi91/saas-adoption-analyzer
cd saas-adoption-analyzer
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txtANTHROPIC_API_KEY=your_key_here
Download from Kaggle: SaaS Product Dashboard (MAU, feature usage, MRR)
Place users.csv, events.csv, and subscriptions.csv in the data/ folder.
streamlit run app.pyOpen http://localhost:8501. The sample dataset loads automatically — click "Generate PM Analysis" to see the AI recommendation.
The tool expects three CSVs matching a standard product analytics schema:
users.csv → user_id, signup_date, country, plan
events.csv → user_id, event_date, event_type
subscriptions.csv → user_id, subscription_status, revenue, churn_date
This mirrors how analytics data is structured in production systems (Amplitude, Mixpanel, Segment for events; Stripe for subscriptions). Swapping the CSVs for direct API connections would make this a live dashboard with minimal code changes.
See data/README.md for dataset source and column documentation.
The design choices in this tool reflect specific product tradeoffs. See the /decisions folder:
- Narrative vs. dashboard — why AI narrative + recommendation over charts alone; the PM workflow this supports
- Prioritization model — what signals the recommendation weighs; why "next quarter focus" not a ranked list; guardrails against AI overreach
- Dataset design — why three tables not a flat file; why cohort_month matters; what production data would look like
Real-time data connection. Replace CSV upload with direct Amplitude/Mixpanel/Segment API integration. The metrics layer requires only column mapping — the analytical logic is production-ready.
Account value weighting. Connect revenue data to weight adoption signals by account value. Low adoption among $500K Enterprise accounts is a higher-priority signal than low adoption among free users.
Configurable strategic context. Allow PMs to input current strategic priorities ("focused on Enterprise expansion this quarter") so the recommendation model filters its output accordingly.
Trend analysis. Compare current period vs. prior period for all metrics — surfaces whether adoption is improving or declining, not just where it stands today.
Segment drill-down. Click any feature in the quadrant chart to see adoption broken down by plan type, company size, and cohort — from overview to diagnosis in one click.
This tool is a direct productization of my experience building Oracle's centralized telemetry and predictive analytics framework, adopted by 50+ engineering teams across a $1B enterprise SaaS portfolio. The consistent failure mode that framework exposed: PMs had dashboards but not synthesis. Data was available; the interpretation wasn't. The AI narrative layer in this tool is the missing piece I wanted to build then.
Aastha Sanghi — Product Manager | Enterprise AI & SaaS
LinkedIn · Substack
Part of an AI PM portfolio series:
- RAG Trust Layer — Enterprise document Q&A with citation-based trust scoring
- Compliance Workflow Copilot — Contract clause analyzer with review checklist and risk scoring
- SaaS Feature Adoption Analyzer — this project