What are UX Metrics?

UX metrics are quantitative measures of a user's experience with a product, spanning what people do (behavioural metrics) and what people report (attitudinal metrics). The behavioural family includes task success rate, time on task, error rate, adoption, engagement, and retention; the attitudinal family includes satisfaction ratings, ease scores, standardised instruments like the System Usability Scale, and loyalty measures like NPS. The two families answer different questions and fail in different ways, which is the first principle of the craft: a healthy metric set always mixes them, because behaviour without attitude misses resentful compliance, and attitude without behaviour misses cheerful abandonment.

Task-Level and Product-Level

Metrics operate at two altitudes. Task-level metrics come from studies: success rates, completion times, error counts, and post-task ease ratings gathered in usability tests and benchmarking rounds. They are diagnostic, comparable across design iterations, and collectable in a week; an unmoderated study in a platform like Ballpark yields success, timing, and a closing questionnaire from every participant in one pass. Product-level metrics come from live usage: adoption, engagement depth, retention, and survey programmes running against production traffic. They are the truth about scale, and they lag: a design change surfaces in task metrics in days and in retention curves in months. Mature measurement runs both altitudes and reconciles them.

Choosing Metrics: The HEART of It

The best-known selection framework is HEART, developed by Kerry Rodden and colleagues at Google: five categories (Happiness, Engagement, Adoption, Retention, Task success) crossed with a goals-signals-metrics process that forces the discipline most teams skip. You start from what the product is trying to achieve (goals), identify what user behaviour or attitude would indicate progress (signals), and only then define the number (metric). The framework's quiet wisdom is that not every category applies to every product, and that metrics chosen without stated goals are just numbers that happened to be loggable. Whatever framework you use, the selection tests are the same: does this metric connect to a decision someone will make, can it be moved by work the team controls, and is it paired with a counter-metric that catches the obvious gaming (speed paired with accuracy, engagement paired with satisfaction)?

Running a Credible Metrics Practice

1. Define each metric precisely, once.
"Task success" needs a written rubric (what counts as partial? what counts as gave up?), or every study measures something slightly different and the trend line is fiction.

2. Measure consistently, on a cadence.
Metrics earn their value longitudinally. Same tasks, same recruitment profile, same instruments, repeated on a schedule: that consistency is what makes quarter-over-quarter movement meaningful.

3. Report uncertainty and sample size.
Task metrics from a dozen participants carry wide confidence intervals, and pretending otherwise is how three-point wobbles become quarterly narratives. The rules of statistical significance apply to UX numbers exactly as they do to revenue numbers.

4. Keep the why attached.
Every metric movement is a question, and the answer lives in qualitative material: the session recordings, the verbatims, the interviews. A metrics readout without an explanation pipeline is a dashboard of mysteries.

5. Guard against Goodhart.
When a measure becomes a target, it stops measuring. Counter-metrics, measurement owned separately from the teams incentivised on it, and periodic audits of whether the number still tracks the experience are the standard defences.

The Benefits

A disciplined metric set turns experience quality from opinion into trend, gives UX a seat in resourcing conversations conducted in numbers, catches regressions before support tickets do, and focuses teams: a small set of well-chosen measures is a statement of what the product is actually for.

The Limitations

Metrics compress rich experience into thin numbers, and the compression always loses something; what is easy to measure is rarely what matters most, and the drift toward loggable proxies is constant. Attitudinal measures inherit every response bias, behavioural ones every ambiguity of intent (long time-on-task: engagement or entrapment?). Single metrics get gamed, worshipped, and weaponised. Measurement is a complement to understanding users, never a substitute for it.

The Takeaway

UX metrics done well are few, precisely defined, behaviour-and-attitude balanced, tracked consistently, reported with their uncertainty, and permanently accompanied by the qualitative work that explains them. Choose them from goals rather than from what the analytics tool happens to log, and remember the purpose: not a healthier dashboard, but a steady, honest answer to whether the experience is actually getting better.