run-usability-test

Run Usability Test

Conduct a structured usability test that reveals real user problems in a prototype or live product with as few as 5 participants.

Why This Is Best Practice

Adopted by: Nielsen Norman Group, Google UX, IDEO, Apple HIG team — usability testing is the core research method in every major UX practice Impact: Nielsen's landmark research shows 5 users discover ~85% of usability problems; each additional 5-user round yields diminishing returns, making small iterative rounds more cost-effective than large studies

Usability testing replaces opinion with observation. Watching a real user fail to complete a task is more persuasive to stakeholders than any heuristic review or UX opinion.

Steps

1. Define research questions — Write 3–5 specific questions you want answered: "Can users find the cancel subscription option without help?" not "Is the UI good?" Questions drive task design.
2. Write task scenarios — Create 4–8 realistic tasks that reflect actual user goals. Phrase tasks as goals, not instructions: "You want to invite a teammate to your workspace. Please do that." Never name UI elements in the task.
3. Recruit representative participants — Recruit 5 users per distinct audience segment. Screen for real usage context, not demographics. Use a screener survey.
4. Set up the test environment — Use screen recording + audio. For remote tests: Lookback, UserTesting, or Zoom with screen share. For in-person: a quiet room, observer in the back, facilitator at the keyboard.
5. Explain the think-aloud protocol — Brief the participant: "We're testing the product, not you. There are no wrong answers. Please say everything you're thinking as you go." Do not say "testing your skills."
6. Facilitate without leading — When a user gets stuck, ask "What are you looking for?" not "Did you try the menu?" Silence is a research tool — let users struggle enough to reveal the real problem.
7. Observe and note — Note every hesitation, misread label, wrong path, and verbal expression of confusion. Timestamp against the recording for review.
8. Synthesize findings into a severity matrix — After all sessions: list every problem, rate severity (1–4: cosmetic to critical), and count how many participants hit it. Fix in severity order.

Rules

• Tasks must be written as user goals, never as instructions that describe the UI.
• Facilitators do not help users complete tasks — that eliminates the data.
• Run a pilot session first with a colleague to debug the task script and tech setup.
• Synthesize findings within 48 hours of the last session while observations are fresh.
• Share a video clip of the most critical finding — a two-minute clip changes stakeholder minds faster than a 20-page report.

Examples

A checkout flow test with 5 users revealed that 4 of 5 participants assumed "Continue" meant "save and exit" rather than "proceed to payment" — a single word change in the button label resolved a 12% cart abandonment rate in the subsequent A/B test.

Common Mistakes

• Tasks that describe the UI — "Click the Settings icon in the top right" removes all the discovery behavior you were trying to observe.
• Facilitator over-helping — Jumping in when a user hesitates gives you false success data; real users won't have a facilitator.
• Testing too late — Running usability tests on shipped code instead of prototypes means findings require expensive rework; test early and often.

When NOT to Use

• Do not run a moderated usability test when the question is quantitative — if you need to know what percentage of users succeed or which of two designs performs better, an unmoderated study or A/B test provides statistically valid data that think-aloud sessions cannot.
• Do not use usability testing to validate business strategy or product-market fit — if users can complete tasks easily but do not want the product, that is a market research problem, not a usability problem.
• Do not recruit convenience participants (teammates, friends, or people in the office) when the target audience has specialized domain knowledge or context — testing a medical records interface with non-clinicians produces misleading results about real-world task failure.