Files
2025-11-30 08:38:26 +08:00

246 lines
11 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Prototyping & Pretotyping: Advanced Methodologies
## Table of Contents
1. [Pretotyping Techniques](#1-pretotyping-techniques)
2. [Fidelity Selection Framework](#2-fidelity-selection-framework)
3. [Experiment Design Principles](#3-experiment-design-principles)
4. [Measurement and Validation](#4-measurement-and-validation)
5. [Common Failure Patterns](#5-common-failure-patterns)
## 1. Pretotyping Techniques
### Fake Door Test
**What**: Feature appears in UI but doesn't exist yet
**Setup**: Add button/link "New Feature X", tracks clicks, shows "Coming Soon"
**Measures**: Click-through rate (interest), wait list sign-ups (intent)
**Example**: Amazon tested new category by showing link, measuring clicks before building inventory
**When**: Test demand for new feature/product before building
### Concierge MVP
**What**: Manually deliver service that will eventually be automated
**Setup**: Humans do work (curation, matching, analysis) as if algorithm did it
**Measures**: Customer satisfaction, willingness to pay, time/cost to deliver manually
**Example**: Food delivery app founders manually taking orders/delivering before building platform
**When**: Learn what "good" looks like before automating, validate service value proposition
### Wizard of Oz
**What**: System appears automated but humans power it behind scenes
**Setup**: Build UI, users interact thinking it's automated, humans respond in real-time
**Measures**: User acceptance of automated experience, performance expectations, edge cases
**Example**: IBM speech recognition - person typing what user said, appeared like AI transcription
**When**: Test if users accept automated interface before building complex AI/automation
### Painted Door
**What**: Feature shown in UI as "Beta" or "Early Access" but not built yet
**Setup**: Badge/flag on fake feature, measure attempts to access
**Measures**: Click rate, request rate for access
**Example**: Slack showed "Calls" feature as "Beta", measured requests before building voice infrastructure
**When**: Test interest in feature when UI space is limited (avoiding clutter)
### Single-Feature MVP
**What**: Build one feature extremely well, ignore everything else
**Setup**: Identify core value hypothesis, build only that feature
**Measures**: Retention (do users come back?), engagement (how often used?), WTP (will they pay?)
**Example**: Twitter v1 - just 140-char posts, no replies/retweets/hashtags/DMs
**When**: Test if core value alone is enough before adding features
### Pre-Order / Crowdfunding
**What**: Collect money before building product
**Setup**: Landing page with product description, pre-order button, collect payments
**Measures**: Conversion rate (visitors → buyers), funding amount vs target
**Example**: Pebble smartwatch raised $10M on Kickstarter before manufacturing
**When**: Test willingness to pay and validate demand with financial commitment
### Explainer Video
**What**: Video showing product in use before building it
**Setup**: 2-3 min video demonstrating value prop, post to landing page, measure sign-ups
**Measures**: View-to-signup conversion, qualitative feedback in comments
**Example**: Dropbox video (3min) drove 70K→75K beta sign-ups overnight (10% conversion)
**When**: Complex product hard to explain in text, want viral sharing
### Manual-First Approach
**What**: Do work manually before building tools/automation
**Setup**: Spreadsheets, email, manual processes instead of software
**Measures**: Feasibility (can we do manually?), bottlenecks (what takes time?), quality (output good enough?)
**Example**: Zapier founders manually connecting APIs for first customers before building platform
**When**: Learn workflow requirements before automation, validate service value before tooling
## 2. Fidelity Selection Framework
### Decision Matrix
| Question | Recommended Fidelity | Timeline | Cost |
|----------|---------------------|----------|------|
| Do people want this? | Pretotype (Fake Door) | Hours-Days | $0-100 |
| Will they pay $X? | Pretotype (Pricing on landing page) | Days | $0-500 |
| Is workflow intuitive? | Paper Prototype | Hours-Days | $0-50 |
| Do interactions feel right? | Clickable Prototype | Days-Week | $100-500 |
| Can we build technically? | Coded Prototype | Weeks | $1K-10K |
| Will they retain/engage? | MVP | Months | $10K-100K+ |
### Fidelity Ladder Climber
Start low fidelity, climb only if validated:
1. **Pretotype** (Fake Door): 5% conversion → demand validated → climb to prototype
2. **Paper Prototype**: 8/10 users complete workflow → UX validated → climb to clickable
3. **Clickable Prototype**: 15% task completion <2 min → flow validated → climb to coded
4. **Coded Prototype**: <500ms latency at 100 req/sec → technical validated → build MVP
5. **MVP**: 40% week-1 retention → value validated → build full product
**Don't skip steps**: Each step de-risks before higher investment
### Cost-Benefit Analysis
**Example - Should we code prototype or stick with clickable?**
Clickable prototype cost: $500 (1 week designer)
Coded prototype cost: $8K (1 month engineer)
**Delta**: $7.5K, 3 weeks
Information gained from coded vs clickable:
- Performance data (real latency, not estimated)
- Integration complexity (real API issues, not mocked)
- Scalability constraints (actual database limits)
**Is $7.5K worth it?**
- If performance/integration unknown and high risk: Yes (de-risking worth cost)
- If performance/integration well-understood: No (clickable sufficient)
## 3. Experiment Design Principles
### Minimum Viable Data
**Qualitative**: n=5-10 for pattern identification (Nielsen Norman Group: 5 users find 85% of usability issues)
**Quantitative**: n=100+ for statistical confidence (conversions, A/B tests)
**Don't over-collect**: More users = more time/cost. Stop when pattern clear.
### Success Criteria Template
**Good criteria** (set before testing):
- Specific: "10% landing page conversion"
- Measurable: Can be tracked with analytics
- Actionable: Tells you to pivot or persevere
- Realistic: Based on industry benchmarks
- Time-bound: "In 2 weeks"
**Decision thresholds**:
- **Persevere**: ≥10% conversion → validated, build it
- **Pivot**: <5% conversion → assumption wrong, change direction
- **Iterate**: 5-10% conversion → unclear, refine and re-test
### Bias Mitigation
**Confirmation bias**: Seeing what we want to see
- Fix: Set success criteria before testing, blind analysis (analyst doesn't know hypothesis)
**Sampling bias**: Testing wrong users
- Fix: Screen participants (e.g., "Do you currently use X?"), recruit from target segment
**Social desirability bias**: Users say what's polite
- Fix: Observe behavior (clicks, time), don't just ask opinions
**Leading questions**: "Wouldn't you love feature X?"
- Fix: Neutral framing: "How would you solve problem Y?"
## 4. Measurement and Validation
### Behavioral Metrics (Reliable)
**Pre-commitment signals** (ranked by strength):
1. **Paid**: Actual payment (strongest signal)
2. **Pre-ordered**: Credit card on file, will be charged later
3. **Waitlist with phone/email**: Provided contact info
4. **Clicked "Buy"**: Navigated to checkout (even if abandoned)
5. **Clicked feature**: Showed interest by interaction
**Engagement metrics**:
- Task completion rate: % who finished workflow
- Time on task: How long (too long = confusing)
- Error rate: Mis-clicks, form errors
- Return visits: Came back without prompt
- Referrals: Told others (strongest retention signal)
### Opinion Metrics (Less Reliable)
**Survey responses**: "Would you pay $X?" (70% say yes, 10% actually pay → 7× overestimate)
**Net Promoter Score**: "Would you recommend?" (aspirational, not predictive)
**Satisfaction ratings**: "How satisfied?" (grade inflation, social desirability)
**Use opinions for context, not decisions**: "Why did you abandon checkout?" (explains behavior) not "Would you buy this?" (unreliable prediction)
### Statistical Confidence
**Sample size for conversions**:
- Baseline conversion: 10%
- Want to detect: 2% change (10% → 12%)
- Confidence: 95%
- **Required sample**: ~1,000 per variant (use online calculators)
**Too small sample**: False confidence (random noise looks like signal)
**Too large sample**: Wasted time/money (pattern already clear at n=200)
### Qualitative Analysis
**Thematic coding**:
1. Collect observations/quotes (n=10 interviews)
2. Identify recurring themes (e.g., "confused by pricing", "wanted export feature")
3. Count frequency (7/10 mentioned pricing confusion)
4. Prioritize by frequency + severity
**Think-aloud protocol**:
- Users narrate thoughts while completing task
- Reveals mental model mismatches: "I expected X here but saw Y"
- Uncovers unspoken assumptions: "I assume this button does..."
## 5. Common Failure Patterns
### Overbuilding
**Symptom**: Coded prototype for question answerable with landing page
**Root cause**: Excitement to build, uncomfortable with "fakery", underestimating learning from cheap tests
**Fix**: Force fidelity ladder (start low, justify climbing), set "maximum time to first test" (e.g., 1 week)
### No Success Criteria
**Symptom**: Ran test, got data, unclear what it means
**Root cause**: Didn't define success before testing, moving goalposts
**Fix**: Write success criteria document before building prototype, get stakeholder sign-off
### Testing with Wrong Users
**Symptom**: Positive feedback from test, market launch flops
**Root cause**: Tested with friends/family (not target), convenience sample (not representative)
**Fix**: Screen participants (qualifying questions), recruit from target segment (ads, outreach)
### Opinion over Behavior
**Symptom**: "Users loved it in interviews" but no one uses product
**Root cause**: Relying on what users say, not what they do (social desirability, hypothetical bias)
**Fix**: Measure behavior (clicks, payments, retention) as primary, opinions as secondary context
### Single Test Overconfidence
**Symptom**: One test shows X, assume validated forever
**Root cause**: Confirmation bias, small sample, didn't test alternatives
**Fix**: Multiple tests, test variations, update beliefs with new evidence
### Prototype Becomes Product
**Symptom**: Shipped prototype code, now have technical debt/security issues
**Root cause**: Pressure to ship fast, reluctance to "throw away" working code
**Fix**: Treat prototypes as disposable (document learnings, rebuild properly for production)
### Analysis Paralysis
**Symptom**: Months refining prototype before testing
**Root cause**: Perfectionism, fear of negative feedback, unclear scope
**Fix**: Time-box prototype building (e.g., 1 week max), test with "good enough" version
### Ignoring Negative Results
**Symptom**: Test shows assumption wrong, but team proceeds anyway (sunk cost fallacy)
**Root cause**: Ego, sunk cost, optimism bias ("this time will be different")
**Fix**: Pre-commit to decision rule ("if conversion <5%, we pivot"), make pivoting psychologically safe