How the score is built

From raw alternate data to an explainable score — and exactly which data each dimension needs from the GST / UPI / Account Aggregator / EPFO rails.

Rails-native, not rail-rebuilding. Consented data access is now solved infrastructure — the Unified Lending Interface exposes 130+ data services to 60+ lenders, and the Account Aggregator has crossed 100M linked accounts. This engine is the explainable scoring layer that sits on top: each feature below names the exact ULI / AA / GSTN / EPFO service it would consume on integration. No pipes to rebuild.

1. STEP 1
   Raw alternate data
   GST / UPI / AA bank / EPFO feeds (12 months)
2. STEP 2
   Feature engineering
   14 ratios & trends, grouped into 7 dimensions
3. STEP 3
   WOE binning
   each feature → a trained bin with a Weight of Evidence
4. STEP 4
   Points
   points = factor × coefficient × WOE (factor = PDO/ln2 = 28.85)
5. STEP 5
   Sub-score 0–100
   scale a dimension's summed points between its worst & best achievable

Each feature value is mapped to a trained WOE bin → contributes points = factor × coefficient × WOE. A dimension's sub-score (0–100) scales its summed points between the worst and best achievable for that dimension; the composite (300–900) is the baseline plus all feature points. Every weight is visible on the Model & Weights page.