← Portfolio
Proprietary

Experience Lighthouse: text mining for customer success at Bci

Industry
Banking · customer experience
Type of work
Text mining & classification
Period
2018

Confidential project: code and data are not public.

Architecture

Emails customers Calls transcribed Support cases and complaints Classic text mining counts, TF-IDFand truncated SVD 10,000+ labels with experienceteams Multi-label XGBoost service, productand sentiment Power BI cross-analysis Customer success Bci · 2018 · Python and XGBoost, scheduled with Airflow

Goal

Give the customer success team an organized read of what customers said across every channel: which service or product they talked about, and in what tone.

What I did

  • We brought together experiences from several channels: emails, transcribed calls and support cases.
  • We used classic text mining: word counts, TF-IDF and truncated SVD.
  • I coordinated manual labeling with the experience teams: more than 10,000 labeled interactions to train the models.
  • We trained multi-label XGBoost models in Python to classify each experience by service, product and sentiment, with runs scheduled in Airflow.
  • We put the results in Power BI so the team could cross channels, products and tone.

Results

10.000+
interactions labeled with the experience teams

Stack

PythonTF-IDFXGBoostAirflowPower BI