← Portfolio
Proprietary

From SPSS Modeler to Python: propensity models at Bci

Industry
Banking · insurance and cards
Type of work
Predictive modeling & migration
Period
2016

Confidential project: code and data are not public.

Architecture

Teradata customer data Training Python + XGBooston team laptops 20+ models cross/upsell,activation, reactivation Scoring Python scriptson Airflow Teradata scores Before: SPSS Modeler · from 2016: Python, XGBoost and Airflow

Goal

Move modeling off SPSS Modeler and onto Python, to build more propensity models, faster, for insurance and cards.

What I did

  • Showed how to migrate from SPSS Modeler to Python and set the initial foundations of the new workflow.
  • With the team we built more than 20 XGBoost models for cross/upsell, activation and reactivation.
  • Scoring ran as Python scripts orchestrated by Airflow, reading from and writing to Teradata.

Results

20+
propensity models
> 0,70
typical model AUC

Stack

PythonXGBoostAirflowTeradata