The Thesis · How we get to +50%
You don't lift CVR with one big swing. You stack surgical, evidence-backed wins — and kill the bad bets before they ship.
Our own test history is unambiguous: full-page redesigns almost always lose (the two biggest destroyers in our archive were rebuilds, −$135K and −$132K/mo). The wins come from component-level changes grounded in real customer objections and proven levers. This engine turns that discipline into a ranked, build-ready queue. Every idea below cites the data that triggered it and the prior test that calibrates its confidence.
Sitewide CVR (90d)
2.76%
2.16% session · 84.8% mobile
The pinch point
3.29%
Add-to-cart — not checkout
Mobile gap
−17%
2.68% mob vs 3.21% desk
Target
4.14%
top-decile · +50% lift
The stacked-win model — illustrative path to 4.14%
Honest framing: this is a model, not a promise. Most individual tests lose or go flat — that's normal and expected. The +50% comes from shipping enough high-RICE bets that the winners compound while losers are caught cheap. Confidence on each idea below reflects how our own history rates that type of change. Treat anything under 55% confidence as a swing, not a sure thing.
Backlog
Priority Board
•
Experiment
Confidence
Evidence base
RICE ↓
P0 · ship next
P1 · this quarter
P2 · backlog
RICE = (Reach × Impact × Confidence) ÷ Effort