Playbook: launch an A/B test that gives a trustworthy result

Playbook: launch an A/B test that gives a trustworthy result

Most "winners" are noise stopped too early. Discipline:
— Calculate sample size before launch; under ~1,000 conversions per arm, don't bother
— Test one change with a real hypothesis, not five tweaks you can't attribute
— Run full weeks only — weekday and weekend traffic convert differently
— Don't peek and stop at "95%"; let it reach the pre-set sample or you inflate false positives

VWO
👍 visual editor, heatmaps, decent stats engine
👎 quote-based pricing climbs fast, flicker on slow sites

GrowthBook (open source)
👍 free self-host, proper Bayesian stats, feature flags
👎 needs a dev to wire up, no WYSIWYG editor

Verdict: VWO if marketers run tests solo, GrowthBook if you have a dev and hate per-seat pricing.

Best for: teams that keep declaring winners that don't hold.

No affiliate link — we just use this stuff.