Skip to content
Eduardo Moreno

Case study

Credit risk strategy for a ~1M-account card portfolio

Scoring models in production, risk segmentation and customer strategies for the credit card business of a major U.S. bank.

Client
BBVA USA
Period
Aug 2020 – Sep 2021
Role
Consultant, Data Science & Strategy at Management Solutions
Where
Chicago, with regular on-site visits to Houston
  • ~1Mcredit card accounts

Context

My first major engagement at Management Solutions was for the credit card department of BBVA USA, whose office was in Houston. I worked from Chicago, traveled regularly to the client, and reported directly to the Director of the Credit Card department.

Challenge

Understand and manage risk better across a portfolio of about one million credit card accounts, spread across different card networks and customer segments.

My role

I owned the risk models and the analysis behind the strategies derived from them, and took part in the decisions on how to act on each segment.

What I did

  • Scoring in production. Developed creditworthiness scoring models combining internal machine-learning scores with FICO bureau scores (TRIAD). The models were deployed to production.
  • Risk segmentation. Designed classification matrices crossing the internal and FICO scores to segment customers by risk level.
  • Customer strategies. Participated in decision-making on the strategies derived from the models, including targeted interest-rate reductions for specific segments of the risk matrix.
  • Delinquency visibility. Built roll-rate matrices tracking transitions between 30, 60, 90 and 120+ days past due, to support recovery and charge-off decisions.
  • Early warning. Deployed an early-warning trigger system to flag customers at risk of default.
  • Next-best action. Developed decision-tree models that determined the action to take with each customer based on their behavior.
  • Data. Prepared and processed customer data from the bank’s databases using SQL, Spark and Hive.

Results

  • Scoring models deployed to production across the portfolio.
  • Significantly improved visibility into past-due accounts, supporting recovery and charge-off decisions.
  • Strategy results from the risk matrices and decision-tree models were presented to the bank’s C-level committee.
  • On the strength of this work, the engagement was extended to the acquiring bank after the acquisition (see the migration case).
  • I received a top-tier (A) performance rating as Consultant, a rating rarely awarded at the firm.

Technologies

  • SQL
  • Spark
  • Hive
  • Machine learning
  • FICO (TRIAD)
  • Decision trees