Prototyping & Testing

Prototyping & Testing

The most expensive way to test a design idea is to build it fully, ship it, and measure the results. The cheapest way is to sketch it on paper and show it to five users before writing a line of code. Prototyping is the discipline of testing ideas at the lowest cost relative to the confidence you need.

The Fidelity Spectrum

Prototypes exist on a spectrum from concept to product:

Paper prototyping — hand-drawn screens, sometimes with physical cut-outs that simulate interaction. Takes hours to create. Perfect for testing information architecture, navigation flows, and major layout concepts. Users readily provide honest feedback on rough sketches because they feel the design is not precious — it clearly invites change.

Wireframes — digital greyscale layouts showing structure, content hierarchy, and layout without visual design. Tools: Balsamiq (deliberately rough), Figma/Sketch (precise). Wireframes separate structural decisions from visual design decisions, preventing premature fixation on aesthetics.

Interactive prototypes — clickable wireframes or high-fidelity mockups that simulate navigation flows. Tools: Figma prototyping, InVision, ProtoPie. Test the actual interaction sequence — do users know where to click? Can they complete the task? — without requiring functional code.

High-fidelity prototypes — pixel-perfect designs that closely resemble the final product. Useful for testing fine-grained interactions, visual hierarchy, and microinteractions. Risk: users focus on visual details ("I don't like that font") rather than structural and conceptual issues.

Coded prototypes — working code, potentially with mocked data. Required for testing performance-sensitive interactions, specific animations, and technical feasibility.

The rule: use the lowest fidelity that can answer your current question. Testing whether users can navigate between sections? Paper. Testing whether a specific animation communicates state change? Coded prototype.

Usability Testing Protocol

A basic usability test:

1. Define tasks — tasks must be realistic ("Find the most recent invoice") not instructions ("Click on Billing"). Tasks that reveal how users approach the interface rather than testing whether they can follow directions.

1. Select participants — 5 participants reveal 85% of usability problems in each round of testing. Match participants to the target user profile; generic "regular internet users" miss domain-specific mental models.

1. Facilitate without leading — the facilitator's job is to keep users thinking aloud, not to guide them to success. When users get stuck, wait. Ask "What are you thinking right now?" not "Can you see the button on the right?"

1. Capture observations — video (screen + audio + face), co-facilitator notes, or dedicated observation rooms. Record the number of users who failed each task, the failure point, and what they said when they failed.

1. Analyse and prioritise — affinity diagram failures, count frequency, assess severity. Severity = (frequency × impact). Fix high-severity problems first; minor polish issues can wait.

Remote and Unmoderated Testing

Remote testing (participants in their own environments) broadens the geographic and demographic range and reduces the Hawthorne effect (behaviour change caused by being observed). Tools: Maze, UserTesting.com, Lookback. Unmoderated testing (participants complete tasks without a facilitator) scales to hundreds of participants but cannot probe unexpected behaviours.

A/B Testing

A/B testing compares two variants (or more, in multivariate tests) on real users in production. Key requirements:

• Sufficient traffic — small samples produce statistically insignificant results. Rule of thumb: 1,000 users per variant to detect a 5% lift with 95% confidence.
• One variable at a time — changing multiple elements simultaneously makes attribution impossible.
• Statistical significance — do not call a winner early. Running tests until a result becomes significant (p-hacking) produces false positives.
• Business metrics, not just UX metrics — measuring click rates on a button is not enough if that button's clicks don't convert. Tie test metrics to revenue, activation, or retention.

Standardised Measurement

System Usability Scale (SUS) — 10 alternating positive/negative statements rated 1–5. Scoring produces a 0–100 value. Above 68 is above average; below 51 is considered failing. SUS enables cross-system comparison and tracking improvement over time.

Net Promoter Score (NPS) — "How likely are you to recommend this product to a friend?" on a 0–10 scale. Promoters (9–10) minus Detractors (0–6) = NPS. Measures loyalty and word-of-mouth potential rather than usability directly.

Task Completion Rate and Time on Task — the most direct usability metrics. What percentage of users completed the task? How long did it take? These should improve with each design iteration.