동적 QR Code A/B 테스팅
Split testing with dynamic QR codes: traffic distribution, landing page variants, statistical significance, and decision frameworks.
Dynamic QR Code A/B Testing
Dynamic QR codes enable server-side A/B testing by splitting redirect traffic between multiple destination URLs from a single printed QR code.
How It Works
The dynamic QR code encodes one short URL. The redirect server splits incoming requests:
- 50% of scans → Destination A (e.g., landing page with video)
- 50% of scans → Destination B (e.g., landing page with image gallery)
The split is typically random or round-robin, and the user assignment is sticky (same device always sees the same variant).
Test Design
1. Hypothesis: Define what you are testing and the expected outcome. - "Landing page with product video will have 20% higher conversion than image gallery"
2. Variants: Create the two (or more) destination URLs with controlled differences.
3. Sample size: Determine how many scans you need for statistical significance. For a 5% conversion rate with 95% confidence, you need approximately 1,500 scans per variant.
4. Duration: Run the test long enough to capture day-of-week and time-of-day patterns (minimum 2 weeks).
What to Measure
| Metric | How to Track |
|---|---|
| Scan rate | QR platform analytics (same for both variants) |
| Page load time | Web analytics per variant page |
| Bounce rate | Web analytics per variant page |
| Engagement (scroll, clicks) | Web analytics events |
| Conversion rate | Goal tracking per variant |
Statistical Analysis
With QR code campaigns, sample sizes are often smaller than digital ad campaigns. Use appropriate statistical methods:
- Chi-squared test: For comparing conversion rates between two variants
- Bayesian analysis: Better for smaller sample sizes; gives probability of each variant winning
- Sequential testing: Check results periodically without inflating false positive rate
Multi-Variant Testing
Some platforms support more than two variants:
- 3-way split: 33/33/34% traffic distribution
- Weighted splits: 70/30 for conservative testing
- Multi-armed bandit: Automatically shift traffic to the winning variant
Key Takeaways
- One printed QR code can test multiple landing page variants server-side
- Need ~1,500 scans per variant for reliable results at 5% conversion
- Run tests for at least 2 weeks to capture temporal patterns
- Bayesian analysis works better than frequentist for small QR sample sizes
- Sticky assignment ensures consistent user experience within a session