UNIQUEst™ Predictive Platform
Data & Features
Our model ingests 47 quantitative features organized into five meta-categories that predict breakout likelihood.
Funding Momentum
- • Round velocity and size progression
- • Investor tier and reputation scores
- • Syndicate composition metrics
- • Follow-on participation rates
Talent Density
- • Hiring velocity by function
- • Engineering ratio and seniority
- • Previous company pedigree
- • Advisor and board composition
Network Strength
- • Graph centrality measures
- • Co-investor density patterns
- • Advisor connection quality
- • Structural hole positions
Founder Signal
- • Prior exit and founding experience
- • Domain expertise depth
- • Technical contribution history
- • Thought leadership metrics
Market Timing
- • Category growth trajectories
- • Competitive landscape shifts
- • Regulatory tailwinds/headwinds
- • Enterprise adoption signals
47 Total Features
Comprehensive feature set engineered to capture both direct signals and second-order effects that predict breakout potential.
Model Stack
Tabular Features → XGBoost
Gradient-boosted decision trees process 35 tabular features (funding, hiring, market metrics). Optimized for precision at top percentiles with SHAP interpretability.
Network Structure → GraphSAGE
Graph neural network learns embeddings from investor-company-founder networks. Captures centrality, clustering, and structural positions that tabular models miss.
Interpretability
Every score includes SHAP explanations showing feature contributions. Risk flags surface potential concerns for human review.
Example interpretability output: "Company X scores 82nd percentile. Primary drivers: high investor tier (+12 points), rapid hiring velocity (+9 points), strong network centrality (+7 points). Risk flag: single previous founder exit, recommend additional reference checks."
Backtests & Monitors
Model calibrated on historical breakout cohorts with quarterly retraining. Performance metrics and drift monitoring ensure ongoing predictive accuracy. Detailed methodology available under NDA for LP review.