A/B testing (QR codes)

The redirect behind a dynamic code assigns each scan to a variant by weight, for example 50/50 or 80/20, and records which one it chose. Results are then compared on what happens after the scan: sign-ups, orders or redemptions, tracked with a different utm_content value per variant or with conversion tracking on each page.

Splits can be decided per scan, so the same person may see both versions on two scans, or made “sticky” with a cookie or a hash so a returning visitor keeps their variant. Testing two printed designs is a different test: that needs two codes, one on each design, compared as separate placements.

Testing with one printed code keeps everything else equal: same placement, same audience, same days. That makes it a fair comparison of what happens after the scan, which no amount of guessing provides.

The usual statistical cautions apply. Judge on conversions, not scans, because every variant gets scans by design. Decide how long the test runs before it starts, do not stop it the moment one variant pulls ahead, and expect small samples to mislead. Run variants at the same time rather than one after another, so time of day and season affect both equally.

A code on 2,000 flyers gets 2,000 scans, split 50/50. Page A turns 60 of its 1,000 scanners into sign-ups (6.0%) and page B 44 (4.4%). That looks like a clear win for A, but a standard two-proportion test gives p ≈ 0.11: a gap this size would turn up by chance about one time in nine, short of the usual 5% threshold. Keep the test running.