Building Data-Driven Client Relationship Management in Banking with Python

Meeting a banking client's needs with machine learning across multiple sales channels

Paul Hughes

Business Cases Data Science Machine-Learning Scientific Libraries (Numpy/Pandas/SciKit/...) Windows

See in schedule

This is a case study that documents how a small data science team in a big bank took on the challenge to transform a fragmented sales process into a data-driven one using Python and machine learning.

This talk outlines the various ways Python has been instrumental in delivering a production solution that serves advisers and relationship manager on a continuous basis.

The Challenge
- A bank has many clients with diverse needs and cost pressures mean fewer advisers resulting in reduced client coverage.
- Multiple sales channels and mixed service levels meant sales processes were uncoordinated and driven by heuristics and often very subjective.
- And... Excel sheets everywhere!

Solution
- Go data-driven!
- Learn from clients and understand product usage
- Empower and inform advisers and call centre agents
- Build a front-to-back sales process (no more Excels!)
- How? With Python!

The Python Bits
- Scikit learn machine learning pipelines that implement two distinct approaches to product affinity in banking and wealth management
- SQL Alchemy based API for data engineering and rapid prototyping of analytics
- Pandas and Jupyter for development and collaboration
- Luigi pipeline for daily processing of millions of transactions and engineering features
- Extracting features from text with NLP (Spacy)
- Delivering machine learning interpretability in production, e.g. with Random Forests and treeinterpreter
- A Python module that we built with all the reusable bits: building training and prediction datasets, developing pipelines, generating monitoring data and enabling explainability

Type: Talk (30 mins); Python level: Intermediate; Domain level: Intermediate