CRO that starts with research, not split tests
Most A/B test programmes flatline because they test the wrong things.
If you can't articulate why a test should win, you're guessing. Research-first CRO has higher hit rates and better-quality learnings. The teams running 30 tests a quarter with a 12% win rate are usually beaten by teams running 8 tests a quarter with a 50% win rate and explainable mechanisms.
The research stack
- Session replays for friction — Hotjar, FullStory or Microsoft Clarity.
- Exit-intent surveys for objections — one question, free text.
- Customer interviews for language — 8 per quarter, recorded and tagged.
- Support ticket analysis for unmet expectations.
The hypothesis template
Because [research finding], we believe [change] will lift [metric] for [audience]. Anything that doesn't fit isn't ready to test. The template forces specificity and rejects pet ideas that aren't grounded in evidence.
Test prioritisation
Score on potential impact (reach × effect size), confidence (research evidence) and ease (engineering hours). PIE, ICE, RICE — pick one and use it consistently. The framework matters less than the discipline of using it.
What kills programmes
Calling winners early, running tests during major external events (sales, launches, outages), and changing the hypothesis after seeing the data. Document the decision rule before launch, follow it after.