Redesigns are exciting and dangerous. New look, new layout, new everything. Then three weeks later conversions are down and nobody can say which change did it.
It does not have to go that way. The trick is to stop treating a redesign as a single big reveal.
Why redesigns so often lose conversions
- Everything changes at once. When layout, copy, navigation, and CTA all move together, you cannot tell what helped and what hurt. A win and a loss can cancel out into a flat, confusing result.
- It is designed for the portfolio, not the visitor. The new site looks award-worthy and quietly loses checkouts.
- The scrappy bits that worked get cleaned away. The old page often has an ugly element carrying real conversions. A tidy redesign removes it without knowing it mattered.
Treat a redesign as a hypothesis, not a verdict
A big-bang relaunch is one giant bet placed all at once. Break it into smaller, testable bets.
- Inventory what converts now. Before touching anything, know your current rates by page and by segment. That is your baseline. A quick CRO audit gives you one.
- Stage the rollout. Ship the new design to a slice of traffic first and measure it against the current version on live visitors.
- Keep the winners. Find the elements doing real work, a specific proof point, a certain headline angle, a short form, and carry them into the new design on purpose.
Redesign in slices, not in one drop
Change one system at a time: the hero, then social proof, then pricing, then navigation. Each slice gets measured against what it replaced. It feels slower, but you learn what actually moved the number instead of guessing after a flat relaunch.
What to measure
- Conversion rate by segment, not just overall, so a gain in one audience does not mask a loss in another. (See Simpson's paradox for why the aggregate can lie.)
- Micro-conversions like scroll depth and form starts, to localize where a drop happens.
- Speed. A prettier page that loads slower can lose more than the design gains.
The shortcut
Most redesigns are really a search for the best version of each section. That is exactly what continuous experimentation does, without the all-or-nothing risk: generate variants of each section, serve each visitor the one that works for them, and let the redesign settle on evidence instead of taste.
