Case study · Banking
Retention & cash analytics
Churn models, customer segments and branch cash forecasts for retail banks, from feature store to monitored production.
- Client
- Retail banks (several engagements)
- Result
- Led a team of four across several banks
- Stack
- Python, scikit-learn, Time-series forecasting, Plotly Dash, Apache Superset
The problem
The banks were losing depositors and cardholders without warning, targeting campaigns broadly, and planning branch and ATM cash on rules of thumb that risked both cash-outs and idle cash.
The work
- Deposit and credit-card churn models covering features, training, evaluation and monitoring.
- RFM and clustering segments for cardholders and depositors, used with marketing teams for targeting and uplift tests.
- Cash forecasts for branch and ATM networks using classical and machine-learning time-series methods.
- Executive dashboards in Plotly Dash and Apache Superset: portfolio health, churn drivers, cohorts and campaign ROI.
- A model registry, CI checks and handover runbooks for the banks’ operations teams.
Outcome
Churn scores fed the banks’ retention playbooks, segments drove campaign targeting, and the cash forecasts cut cash-out risk and logistics cost. I led a team of four data scientists across these engagements.
The client isn’t named, to respect confidentiality. I can share more detail on a call where permitted.
Talk about a similar projectMore work
Claim-denial prediction
A model that flags insurance claims likely to be denied before they are submitted, served to internal apps through a monitored API.
Ran in production with monitoring
Support assistant in Microsoft Teams
A Teams and telephony assistant that answers support questions from the company’s own documents, using an OCR and entity-extraction pipeline underneath.
Live in a corporate Teams tenant