A faster checkout can be an improvement, unless people are moving quickly because they have missed an important charge. More time in an application can indicate useful work, or difficulty completing it. UX metrics quantify aspects of an experience, but their meaning depends on the goal and the behaviour behind the number.

They include observed outcomes, such as task success and errors, and reported judgements, such as perceived ease or satisfaction. These sources answer different questions. The right set gives the team enough information to judge the experience without turning every available event into a performance target.

Begin with the outcome people need

Define what the product should help someone achieve, then identify evidence of progress and a measure that represents it. Google’s HEART framework uses a goals–signals–metrics process alongside categories covering happiness, engagement, adoption, retention and task success. It is a way to reason about selection, not a requirement to fill every category.

For a fictional document-signing tool, a useful goal might be completing an agreement with confidence about what was signed. Completion rate alone misses comprehension; time alone may reward haste. The team might combine accurate completion, errors and a relevant post-task judgement, depending on the decision.

Choose measures the team can interpret and act on. A broad business outcome can matter while being too distant from a specific interaction to diagnose it. Connect product-level and task-level evidence without assuming they are interchangeable.

Write the definition precisely enough to repeat

Task success needs a scoring rule. Time on task needs a start, an end and a decision about pauses or abandonment. Retention needs a qualifying action, population and period. A change in any of those definitions can move the metric without changing the experience.

For reported measures, preserve the instrument and context where comparison requires them. The System Usability Scale measures perceived usability, while Customer Effort Score focuses on perceived effort or ease. Neither establishes what happened in a task merely because its score is numerical.

Test instrumentation and scoring before relying on a trend. A duplicate event, changed survey trigger or new eligibility rule can create a persuasive but misleading improvement.

Interpret change with uncertainty and context

Report the relevant sample and confidence interval or other appropriate uncertainty measure. For comparisons, account for repeated observations, clustering and the sampling design. Do not assume that every movement between two averages is meaningful.

Review qualitative evidence where it can explain the result. In the signing example, participants may complete more quickly because a clearer summary helps them, or because a redesign hides detail they should inspect. The same metric movement can support very different conclusions.

Guard against optimising a proxy at the expense of the goal. Pair a speed target with the outcomes that should not deteriorate, and periodically examine whether the measure still represents value to the person using the product. A useful metric set is small enough to understand and broad enough to reveal an important trade-off.