Self-supervised learning can model fault propagation in networks well enough to causally attribute customer impact to root causes, enabling faster incident response than traditional rule-based systems.
NetCause learns how faults spread through networks to identify root causes of outages. Instead of using fixed rules, it trains on real incidents to understand fault propagation patterns, then uses counterfactual simulation to rank which component most likely caused a customer impact. Tested on production cloud network data, it outperforms rule-based approaches by 16%.