Conversion Rate Optimization (CRO) Guide 2026
A modern CRO guide — hypothesis, experiments, funnels, heatmaps, and the mistakes that waste your test bandwidth.
Map the Funnel First
Instrument every step from ad click / SEO landing → activation / purchase.
Find the worst-performing step (biggest drop-off relative to intent).
Fix biggest leaks first — 30% improvement on your worst step >> 3% on your best.
Hypothesis Discipline
Every test starts with: 'We believe [change] will cause [effect] because [insight].'
Insight from user research — not gut feeling.
Predict expected lift and audience segment.
Qualitative Research
Session recordings (Hotjar, FullStory, LogRocket, Clarity).
Heatmaps for above-the-fold engagement.
5-user interviews per major screen.
Exit surveys.
Where to test first
| Funnel step | Common leaks |
|---|---|
| Landing hero | Vague value prop, weak CTA |
| Pricing | Bad tier structure, hidden prices |
| Signup | Too many fields, no social auth |
| Onboarding | Too many steps, weak first outcome |
| Cart | Surprise costs, forced signup |
Running Tests
One variable per test unless clearly independent.
Sample size calculator first — don't launch under-powered tests.
Run for full weekly cycles (7, 14, 21 days) to capture day-of-week variance.
Don't peek.
Tools
Google Optimize is dead (RIP). Replacements: VWO, Optimizely, Kameleoon, GrowthBook (open-source).
PostHog for full-stack experimentation + analytics.
Convert.com, AB Tasty for mid-market.
High-Impact Tests
Headline + subhead on landing pages.
Primary CTA copy + placement.
Pricing page layout + anchoring.
Signup form length + fields.
Onboarding step order + count.
Common Mistakes
Testing tiny changes (button colors) that can't produce learning.
Under-powered sample sizes → false positives.
Peeking + stopping tests early.
Testing without a hypothesis.
Ignoring segment-level results.
Frequently Asked Questions
How much traffic do I need to A/B test?+
Rough rule: 1000+ conversions per variant to detect a 10% lift. Below that, focus on qualitative + iterative changes.
Can I test everything at once with a multivariate test?+
Only with huge traffic. For most sites, sequential A/B is safer.
Is 95% confidence enough?+
It's the industry default. For high-stakes decisions, use 99%. Pair with practical significance.
What about bandit tests?+
Great for optimizing over long periods (e.g., content headlines). Not a replacement for hypothesis-driven A/B.
When to stop testing?+
When you've hit sample size AND you've captured full weekly cycles. Never based on 'it looks good so far'.