Soubhagya Dubey

Senior Data Analyst specializing in credit risk analytics, portfolio optimization, and lending strategy for retail fintech. ~3 years building data pipelines, risk models, and decision frameworks for unsecured lending at scale.

Designed Expected Credit

Loss provisioning model and tenure-based pricing strategies.

Proven impact: 2x risk differentiation on 43K+ borrowers, +10% portfolio growth, +15% conversion improvement. Expert in Python, SQL (Snowflake), XGBoost, and decision-tree-based risk segmentation.

Senior Data Analyst — Credit Risk & Portfolio Analytics at Ganit Business Solutions - Kissht (2024-07 – Present)

Award: "The Prodigy". Unsecured personal-loan book: ₹7,154 Cr

• ECL Provisioning Model — Upgraded Expected Credit Loss model for 7,154 Cr unsecured personal-loan book. Previous method used coarse DPD-bucket level (flat 4.3% PD). Designed decision-tree upgrade splitting borrowers on paydown progression, bureau score, risk tags, enquiries, and residual tenure.

Impact: PD spread widened to 3.4%–15.8% range, surfaced ~ 20 Cr hidden risk, enabled borrower-level audit-defensible provisioning. Productionised on Snowflake (300K+ loans monthly).

• Risk Segmentation & Repeat Policy — Led rule-based segmentation of 43,560 repeat borrowers achieving 2x risk differentiation (1.6% vs 3.1% 3M30 DPD rate). Designed tenure-migration logic moving customers from 3M to 6M loans based on repayment progression and bureau movement. Owned ticket-size optimisation replacing flat caps with tenure-wise FOIR and bureau-score-band sizing.

Impact: repeat conversion +15% with portfolio risk held flat (±0.3%); identified ~7,600 for top-up expansion.

• Pricing & Portfolio Growth — Engineered targeted loan offers using V27/V43 model bands and DPD patterns, expanding addressable base by 40K+ users (+12% portfolio growth). Designed tenure-wise ticket sizing using FOIR thresholds and bureau insights (+15% conversions). Converted negative-margin high-risk pools to 17% margin by launching low-ticket 12-month loans. Shifted 95% eligible users from 3M to 6M tenure (+10% volumes, +18% CLTV).
• Geographic Risk & Collections Strategy — Implemented pincode-level risk tagging into underwriting policy, reducing bounce rates -2%. Built DPD-segmented collection strategy for 500K+ active loans tracking T0-T180 payment flags, EMI buckets, and POS metrics. Set up daily PSI/KS monitoring on portfolio health with vintage curves (MOB1-MOB12) and auto-rollback triggers.
• Analytics Infrastructure — Architected end-to-end Snowflake pipelines processing 300K+ loans monthly across Ring and Kissht, optimising query runtime from 45+ min to <10 min via temp-table architecture. Built approval funnel analytics (200K+ applications monthly), identifying bottlenecks that improved conversion rates +12% (new) / +16% (repeat).

Data Analyst — Fintech Lending Product at Zappian (Credmudra) (2023-05 – 2024-06)

• Led analytics department for new fintech product, establishing KPI infrastructure supporting marketing and risk strategies for 50K+ monthly leads.
• Conducted CIBIL data EDA for lead segmentation, designing targeted marketing campaigns that improved conversion rates +12% across credit score bands.
• Built data warehouse using Pentaho and ClickHouse, streamlining analytical workflows and reducing report generation time from 3 hours to 15 minutes.

Bachelor of Technology in Mechanical Engineering – Gyan Ganga Institute of Technology and Sciences (2016 – 2020)