Experience cannot be measured directly through task performance the way usability can; it is only inferred

What it is

Usability has a direct measurement path: assign a task, watch completion, count errors, record time—the performance is the verdict. Experience has no such path: feeling is an internal state, and task performance is only a weak correlate. Experience measurement is inherently inferential—proxy signals such as self-reports, behavioral traces, and physiological readings estimate a quantity that cannot be read directly, out of noise.

Why it happens

The impossibility of direct reading comes from experience's multi-source synthesis: feeling is generated by performance, expectation, social context, and memory reconstruction together, and any single signal covers one source. Self-report scales measure "the feeling one is willing to report," shaped by willingness to express and by setting; behavioral traces (reuse, recommendation, detours) are outcomes of experience, not experience itself; physiological signals measure arousal, not evaluative polarity. Every proxy has systematic bias, so inference quality depends on how well the proxy portfolio covers the sources and how far the biases cancel—which is exactly why experience research costs more and moves slower than usability testing.

Studying it

The norm for inferential measurement is multi-proxy triangulation: self-report (scales, diaries) plus behavior (retention, recommendation, feature revisits) plus context records (occasions, accompanying events), reporting agreement and divergence across proxies. Divergence is itself informative—"says satisfied but never returns" points at social desirability or context-binding. Design the inference chain explicitly: state the proxy-to-construct assumption (why recommendation stands for experience) and calibrate proxy validity against products of known standing (widely good or bad).

Where it stops holding

Indirect does not mean unmeasurable; it means "measurement is modeling," and wrong model assumptions are more dangerous than metric noise. Proxy signals react fast to sharp short experiences (catastrophic failures) and slowly to drift (habituation, fatigue), which long-window data must cover. Across cultures, self-report baselines differ, and proxy models do not transfer unchanged. Any "experience score" must ship with its model description and error bounds; citing it as a bare point estimate is misuse.

Applying it

• Fix a multi-proxy portfolio for experience dashboards: self-report scores, retention/return behavior, and key-scenario observation, each annotated with its bias direction.
• When reporting experience conclusions, write the inference chain—which proxies, what assumptions, what error—never a single "experience score."
• Calibrate against products of known reputation periodically to check that proxy signals still align with the construct.

Related

• Same group: B5.11.1 Usability asks whether the task is achieved; experience covers all feelings before, during, and after use · B5.11.2 Experience includes expectation, brand, and social context, most of which the interface does not control · B5.11.3 Under some experience goals, deliberate inefficiency is correct design, in direct conflict with usability goals
• Nearby: Q4 Measurement and Reliability · B5.05 Experience versus usability
• Search terms: experience measurement · proxy measures · triangulation