Case Study: BRED minimizes ATM downtime and predicts failures with Dataiku

BRED Minimizing ATM Downtime with Machine Learning

BRED, a retail bank serving over one million customers, needed to minimize downtime across its more than 600 ATMs by turning raw, highly imbalanced outage data into actionable maintenance insights. To meet this challenge, The Data Factory — BRED’s data center of excellence — built a production monitoring and predictive-maintenance system using Dataiku.

Using Dataiku, the team combined clustering, survival analysis, and time-series models to classify failure patterns (finding more than 70% of ATMs rarely fail and varied outage-duration distributions such as ~20% >15 minutes and ~40% <2 minutes) and implemented thresholded alerts and dashboards to avoid false alarms and prioritize repairs. The Dataiku-powered solution runs in production across 600+ ATMs, provides geography- and team-level maintenance granularity, reduced unnecessary dispatches, and gave BRED measurable visibility to cut downtime while the team continues model optimization and automation.

BRED

Bertrand Ring

Head of Data Service