The AI that decides what to build next
ProductOracle is a product decision engine for product leaders that ingests customer + revenue + usage signals and produces a ranked, strategy-aware roadmap with probabilistic impact and explainable rationale.
Revenue-coupled prioritization from multi-source signal -> ranked bets -> weekly decision pack.
Positioning: turn messy product signals into a ranked, revenue-tied roadmap with explainable EV and assumptions.
npm i
npm run dev- Landing + dashboard MVP is live and demoable
- CSV/TXT signal ingestion and explainable prioritization are working
- Strategy Profiles + capacity constraints are scaffolded
- Weekly decision pack export (
.md) is working - Optional Gemma AI synthesis is integrated (via Google Gemini API free tier)
- Uses Next.js 15 App Router, TypeScript, Tailwind CSS, shadcn/ui, lucide-react
For AI-powered prioritization insights on the dashboard, add your free Google AI Studio API key to .env.local:
GEMINI_API_KEY=your_key_here
Get a free key at https://aistudio.google.com/app/apikey. Then enable "Use Gemma AI" when processing signals. No local server required.
In 30 days, ProductOracle should show:
- 2 real connectors live (support + CRM or analytics)
- EV-based ranking under capacity constraints
- Weekly decision pack used in real planning cadence
- At least 1 shipped bet with predicted vs actual outcome logged
The moat is not the model. It is:
- Outcome feedback loop (prediction -> actual -> calibration)
- Company-specific objective function + priors
- Institutional memory graph (decision history + evidence + rationale)
- Cross-source normalization and dedupe quality over time
- Week 1: Strategy profiles + capacity-constrained optimizer
- Week 2: Connectors v1 (support + CRM) with normalized schema
- Week 3: Probabilistic EV (risk-adjusted EV + sensitivity)
- Week 4: Outcomes + calibration + weekly pack v2
See docs/EXECUTION-PLAN.md for full narrative and demo script.