Summary
Overview
Work History
Education
Skills
Timeline
Generic

Soumya Ghosh

Bengaluru

Summary

I currently lead a four-member Data Science team in Amex. We have end-to-end ownership of the models for the one of the major marketing channels in Amex which brings in 10% of new customers annually. We develop, validate and track these models through their life cycles. We also conduct research on Machine Learning techniques to improve their performances. My role requires 50% individual contribution work and 50% people management work.

Overview

9
9
years of professional experience

Work History

Lead Data Scientist

American Express
01.2017 - Current
  • Neural Networks and Factorisation Machines: Developed Field Aware Factorisation Machines (FFM) model to capture feature interactions and improve Top-N performance of travel recommendation system by 65% vs benchmark
  • Researched two tower Neural Network model with trained user and item latent feature embeddings
  • Uplift modelling: Built the first offer models on Amex's product page using Uplift Modelling framework to target offer sensitive segments leading to a cost saving of $20 M
  • Researched Meta-Learners like S, T, X and R learner and Uplift Gradient Boosted Trees
  • Feature selection was done using K-Bins KL Divergence method
  • Improved uplift metrics Qini by 23% over Meta-learners
  • Gaussian Mixture Modelling: Boosted conversion for the Paid Search Channel in UK market by 12% vs rule-based segmentation, by using soft clustering technique Gaussian Mixture Models to segment geolocations based on demographics information
  • Market Mix Modelling: Analysed marketing investments using Market Mix modelling techniques
  • Engineered features to capture market lag effect and diminishing return in marketing activity
  • Analysed response curves to find more opportunities to maximise ROI, developed multivariate regression SUR (Seemingly Unrelated Regression) at product X channel level
  • Research used by marketing to increase investments in Media by 30% (~15$ M) in 2021
  • Response Model: Led the re-development for one of the major marketing channels in Amex with estimated increase in annual revenue by $17M
  • Selected features for XGBoost using Information Value, Mutual Information and backward selection
  • Tuned hyper-parameters using Bayesian Optimisation and model interpretation using SHAP and PDP
  • Managed stakeholders from strategy, tech and governance and compliance teams
  • Image classification: Generated insights on the performances of successful Amex ads leading to an estimated growth in conversion by 8% in co-brand channels
  • Used Google Vision API for object detection and text extraction from Amex online ads
  • Used template matching to detect which product was being advertised on image
  • And transfer learning on ImageNet to build a model on the click through rate of the ad

Analyst

Dunnhumby
01.2016 - 07.2017
  • Approximate Nearest Neighbour Search: Used Approximate Nearest Neighbour (ANNOY) to match products across different vendors and improved on the limitations speed of KD-Trees by 400%
  • Topic Modelling / Market Basket Analysis: Used LDA (Latent Dirichlet Allocation) to find groups of products frequently bought together
  • Analysis was used for marketing campaigns for product bundles leading to an incremental revenue of $11M

Education

B.Tech + M.Tech (Dual Degree) - Mechanical Engineering

IIT Kharagpur
01.2016

Skills

  • Python
  • Spark
  • SQL
  • Machine Learning
  • Recommendation Systems
  • Uplift Modelling
  • Neural Networks
  • Model Interpretability
  • Data Science Research
  • Analytics
  • Project Management
  • Stakeholder Management
  • Governance

Timeline

Lead Data Scientist

American Express
01.2017 - Current

Analyst

Dunnhumby
01.2016 - 07.2017

B.Tech + M.Tech (Dual Degree) - Mechanical Engineering

IIT Kharagpur

Similar Profiles

CREATE PROFILE
Soumya Ghosh