Checkout is where the money is closest, so it attracts the most opinions. It is also where most changes are tested on too little traffic to mean anything, so the opinions survive.
Every term used above is defined in our experimentation glossary, and three experiments taken apart step by step sit in the case studies. Longer pieces with the workings attached are in the field notes.
Documented average cart abandonment sits above 70%, but a large share of it is people who were never going to buy on that visit. Treating all of it as recoverable produces targets nobody can hit and changes nobody needs.
Do this: Measure abandonment from the step where payment details begin, not from add to basket. The two numbers describe different people.
Delivery, tax and fees revealed at the final step is the most common documented reason for leaving. It is also the cheapest to fix, because it is a disclosure problem rather than a design one.
Do this: Show the full total, delivery included, before the payment step. If that number loses you orders, they were being lost anyway, later and more expensively.
Send traffic and baseline rate. We reply with the smallest effect your test can actually detect. One email, no call.
Requiring registration before purchase adds a step whose only beneficiary is your database. Guest checkout with an optional account after the order keeps both.
Do this: Offer guest checkout. Ask about the account on the confirmation page, when the person has already succeeded.
At 40,000 orders a year, a month of traffic detects roughly a 12% relative change, not the 3% you hope for. Seeing 3% would take about sixteen months, which is not a test, it is a season.
Do this: Compute this before you build. If the test cannot see your expected effect, make a bigger change or ship on judgement.
This is teaching material and our own reading of standard practice, not advice for your specific site. Check anything important with your own specialist before you act on it.
Traffic, baseline rate and what you changed. We reply with the minimum detectable effect and whether the result meant anything. No cost, no call.