Root cause analysis (RCA)

Root cause analysis is the practice of working backward from an observed problem, a metric drop, a system outage, a spike in complaints, through the chain of contributing events and conditions to find the underlying cause, rather than stopping at the first plausible explanation or fixing only the visible symptom. Common techniques include the "five whys," repeatedly asking why a factor occurred until reaching something actionable, and fishbone (Ishikawa) diagrams that organize candidate causes into categories such as people, process, and systems.

In analytics work, RCA usually starts with decomposing a metric change: segmenting by product, geography, channel, or user cohort to find where the change is concentrated, then checking for confounding events, a deployment, a pricing change, a seasonal pattern, a tracking bug, that coincide with the shift. This differs from purely predictive work in that the goal is an explanation with a defensible causal story, not just a correlated signal.

RCA matters because fixing a symptom without finding its cause tends to let the same problem recur in a different form. The most common pitfall is stopping at the first explanation that fits the data, especially one that confirms an existing assumption, rather than ruling out competing explanations, and treating a correlated factor uncovered by segmentation as a proven cause without further investigation.