Six changes tested over a quarter on a store doing 12k orders a month, ranked by impact.
We ran these one at a time on a store doing 12,000 orders a month, each for two weeks against a holdout. Ranked by what actually moved completion rate, not by what sounded clever.
What worked
Removing the account-creation step was worth more than the other five changes combined. Second was showing shipping cost on the cart page rather than at step three of checkout — people do not abandon because shipping is expensive, they abandon because it appeared late. Third was making the payment step render without waiting on the gateway’s script.
What did nothing
A trust-badge row under the pay button: no measurable change. A progress indicator across the top of checkout: no measurable change, and it added a render-blocking stylesheet. Both are on every “increase conversion” listicle.
The pattern across all six: changes that removed a step or moved information earlier worked; changes that added reassurance to an unchanged flow did not.