Split Conversion Rate by New vs Returning Before You Touch the Site
A clear order confirmation email answers the questions buyers ask support most often: what was bought, what it cost, when it ships, and how to track it. Getting those details right cuts ticket volume in the busiest week of the order cycle.
A clear order confirmation email answers the questions buyers ask support most often: what was bought, what it cost, when it ships, and how to track it. Getting those details right cuts ticket volume in the busiest week of the order cycle.
This guide walks through order confirmation emails in plain steps, with the trade offs stated honestly and a short list of checks you can run yourself. Nothing here needs special equipment or a professional call out to get started, and each section ends with the decision that actually matters.
Address validation checks a typed address against postal data as the shopper types, then suggests the corrected version. Used well it cuts failed deliveries and reshipping costs. Used badly it rejects valid rural, military, and new build addresses, so always let the shopper keep their original entry.
A good checkout error names the exact field, says what went wrong in plain words, and tells the shopper the one action that fixes it. Errors should appear next to the field, keep everything the shopper already typed, and never blame the shopper for a formatting difference the store could handle itself.
Neither format wins on its own. Drawers suit small baskets and single product stores because they keep the shopper on the page they were browsing. Full cart pages suit larger baskets, discount codes, and gift options, where shoppers want room to review before paying. Pick based on how your shoppers actually edit a basket, not on what looks modern.
Export the last ninety days of returns and group them by reason code first, product second. A sizing problem on one product, a quality complaint spread across a category, and a wave of arrived late codes are three different fixes with three different owners. Do it now, while there is still time to change a size chart, a product photo, or a carrier promise before peak volume multiplies whatever is already wrong.
Before any redesign meeting, pull conversion rate separately for new and returning visitors for the last ninety days and compare each line to the same period last year. If returning conversion is flat and new visitor conversion is flat, but the blend fell, nothing on the site broke. The mix shifted, usually because paid or social traffic grew as a share. That is a media problem wearing a website costume.
The blend is an average of two different stores
New and returning visitors do not convert at remotely similar rates, and everyone in ecommerce knows it, yet the number in the weekly report is still the blend. Returning visitors arrive knowing sizes, prices, and the returns policy. New visitors are still deciding whether the store is legitimate. When the share of new traffic rises, the blend falls even if both groups convert exactly as well as before.
Do the arithmetic once, out loud, in the meeting. A store converting returning visitors at three times the new visitor rate only needs a modest mix shift toward new traffic to post a scary looking decline. No page changed. No test explains it. The chart just punishes whoever owns the website that week.
What to pull, and in what order
Four lines: new visitor sessions and conversion, returning sessions and conversion, each against the same weeks last year. Then split new visitors once more by channel, because a surge of low intent social traffic and a surge of branded search traffic are both new visitors and they behave nothing alike. That second split is where the actual story usually sits.
Watch the session definition while you are in there. A returning visitor on a new phone is a new visitor to your analytics, and cookieless or consent rejected sessions inflate the new bucket in ways that have grown in the last two years. The split is still worth more than the blend. Just do not treat the boundary as clean.
When the split says the site really is the problem
Sometimes it does. If returning conversion drops on its own, that is your best customers failing, and it points at the site or the offer: a checkout change, a payment option that vanished, a loyalty discount that stopped stacking. Returning visitors are the control group you did not have to pay for, because they already trust you. Their decline is the loudest signal in the account.
New visitor conversion dropping while returning holds usually means the landing experience and the traffic promise stopped matching: an ad promising one product, a page leading with another, a price in the ad that the page corrects. Fix the match before you fix the page design.
Make the split the default report
Change the weekly report so the blend never appears alone. Two lines and a mix percentage. It takes one reporting change and it ends a recurring argument, because the question shifts from who broke the site to what changed in traffic, offer, or experience, in that order. Stores that report this way stop redesigning in a panic every time a paid campaign scales.
Keep a short log of mix shifts next to the site change log. When conversion wobbles in a quarter, you want both records side by side before anyone proposes work. Half the time the log answers the question and the meeting ends early, which is the real return on this whole exercise.
Keep reading: How Edge Personalization Works Without Slowing the Storefront Down, Edge Observability: Seeing Storefront Errors Where They Actually Happen, and Image and Font Delivery at the Edge: The Quiet Speed Win cover the nearby cases in this series.
Frequently asked questions
Why does blended conversion rate fall when a store grows paid traffic?
Paid and social traffic are mostly new visitors, who convert at a much lower rate than returning visitors. Growing their share lowers the average even when each group converts exactly as well as before.
Which drop is more worrying, new or returning visitor conversion?
Returning. They already trust the store, so their decline points at the site, the checkout, or the offer. A new visitor decline more often means traffic and landing page stopped matching.
How far back should I compare the split?
Ninety days against the same period last year is a workable default. It smooths weekly noise and controls for seasonality better than comparing to last month.
Can analytics tell new from returning visitors perfectly?
No. A returning customer on a new device or browser reads as new, and consent rejection inflates the new bucket. The split is still far more informative than the blend, as long as you treat the boundary as fuzzy.