Root Cause vs Symptom
Decompose a metric change before choosing a fix. · Product Judgment · Lesson 08 · 3 min
Root Cause vs Symptom · 3 min
Situation
“Conversion is down.”
A store's checkout conversion drops from 40% to 32%. The team proposes a redesigned payment page.
The metric tells you something changed. It does not tell you where, for whom, or why. A redesign is an expensive diagnosis if the underlying problem is a broken payment provider or a new traffic mix.
Mental model
Move from observation to mechanism.
First verify the measurement. Then break the journey into steps and segments. Finally investigate the mechanism behind the affected step.
Separate fewer eligible users reaching checkout from more checkout users failing to pay. Use the same denominator and time window when comparing rates.
Example
Follow the narrowing evidence.
Suppose desktop conversion is stable and mobile conversion fell. On mobile, address entry is unchanged but payment submissions fail more often.
Logs show a validation error after an address-field update. Support reports the same symptom. That pattern points toward the payment contract or submitted data, rather than the visual design of the whole checkout.
Failure case
Stopping at the first plausible story.
“The release happened yesterday” is useful timing, not proof. A campaign, outage, tracking change, or shift in country mix could coincide with it.
Look for evidence that would distinguish explanations. Did unexposed users change too? Do server records show fewer completed orders, or only analytics events?
PM decision
Stabilize while you investigate.
If an obvious regression is causing material harm, a targeted rollback may be justified before full causal certainty. State what you know and what remains uncertain.
After recovery, confirm the failure mechanism and add prevention. “The metric recovered” is not sufficient explanation if multiple things changed simultaneously.
Remember this
A symptom names what happened; a cause explains how.
Decompose until the explanation points to a test or intervention. Keep competing explanations alive long enough to check them.