Propensity Model

Authored by Ameex Technologies on 04 Sep 2019

Problem Statement

  • The client is a leading ecommerce company selling baby nutrition products
  • The client wants to leverage user website behaviour to improve their business
  • The client wants to know who the potential customers are to be targeted

Our Approach

  • Data from BigQuery is fetched and moved to google cloud storage
  • Data processing tasks like handling missing values, handling categorical variables are done using dask parallel processing
  • Feature engineering was done based on users journey from the past 30 days
  • Model is developed using LighGBM and trained using cross validation technique to prevent overfitting

Visualization - Propensity Model


A framework to provide propensity score and insights that will be used to find potential customers who are very likely to enroll and who are less likely to enroll


  • Model predicted 82.76% data accurately in training data and 78.27% data accurately in testing data
  • Model gives the propensity score which is used to determine a user’s likelihood to enrol
  • Propensity score indicates the probability of conversion for that user, 0 being extremely unlikely to enrol and 1 extremely likely to enrol. Propensity score can be used by marketing team to optimise channels/campaigns based on users who are highly likely to convert

Case Studies