measure

Use the Goal-Signal-Metric framework: state the user outcome as a goal, identify observable behaviors as signals, then define quantifiable metrics with thresholds and data sources. Select 2-3 HEART dimensions most relevant to the feature's intent rather than measuring everything.

HEART is Google's framework covering Happiness, Engagement, Adoption, Retention, and Task Success. Apply it per feature rather than globally, and pair every optimized metric with a counter-metric that could suffer, such as satisfaction declining while engagement rises.

Sample size depends on baseline conversion rate, minimum detectable effect, statistical power (typically 80%), and significance level (typically 95%). Lower baselines and smaller effects require dramatically more traffic; if required sample exceeds monthly traffic, use qualitative research instead.

Triangulate whenever numbers show what happened but not why, such as funnel drop-offs, satisfaction declines after redesigns, or inconclusive A/B tests. Start with quantitative to locate the problem, use qualitative to understand causes, then verify fixes quantitatively.

High engagement can signal addiction rather than satisfaction, and metrics become targets that invite gaming per Goodhart's Law. Pair engagement with satisfaction and task completion metrics, and check whether improvements could have been achieved through dark patterns.