Seven in Ten Carts Leave Before They Buy, and a Small Store Is the Least Likely to See Where

The average is 70 percent, and an average is not a diagnosis

The 70.22 percent figure is a real, well-documented aggregate: the Baymard Institute pools fifty separate cart-abandonment studies to reach it, and it has held near that level across the 2026 update of its running research. It is worth treating with two caveats before it is used for anything. It is a central tendency across very different stores and audiences, not a benchmark any single business should expect to hit exactly. And it conflates browsers who never intended to buy in that session with buyers who intended to complete a purchase and were stopped by the funnel itself. Only the second group is recoverable, and the average alone does not tell you which group you actually have or where in your funnel they left.

The compounding leak most small stores never see: mobile

Industry benchmark aggregations widely report that conversion rates run roughly 57 percent higher on pages that load under two seconds compared to three to four seconds, and that about 53 percent of mobile visits abandon a page that takes longer than three seconds to load. Those specific percentages are secondary-sourced industry benchmarks, not one disclosed primary study, and should be read as directional rather than exact.

What is not directional is the underlying standard they are measured against. Core Web Vitals, Google's own published performance metric, sets the bar most engines now use to judge a page's loading, responsiveness and visual stability. And the gap between where shoppers are and where sites currently perform against that bar is real and measurable: mobile traffic now accounts for roughly 62 percent of e-commerce sessions, while only about 42 percent of mobile e-commerce sites currently pass all three thresholds, compared with roughly 63 percent on desktop. A store optimized on the desktop the owner personally tests on, while most of its actual traffic arrives on a mid-range phone over a mobile connection, is being graded on the wrong device.

Why the small store is the least likely to see either leak

This matters more here than the statistics alone suggest, because of who actually runs the typical online store. The median Shopify or WooCommerce store is a small operation, not the large retailer the checkout research is usually framed around, and it is the store least likely to have anyone on staff whose job is to watch a Core Web Vitals dashboard or read a checkout funnel report weekly. A large retailer that loses a fraction of a percent to a slow checkout notices in a quarterly review. A small store loses the same fraction of a much smaller number of total buyers, and the loss simply reads as "traffic that did not convert," with nothing pointing at the specific screen or device responsible.

What a decade of checkout research says is fixable

The constructive half of the same research corpus is worth stating plainly: Baymard's Checkout Usability program, built on more than 300,000 hours of testing since 2011, finds the average large e-commerce site can gain roughly a 35.26 percent increase in completed orders from checkout-usability improvements alone. The two most-cited structural causes it names, cost disclosed too late and a forced account requirement, are both mechanical problems with mechanical fixes, not a pricing problem or a persuasion failure. None of it requires guessing. It requires walking the real path a buyer takes, on the device they actually use, and finding the specific point it breaks.

From a percentage to a ranked list of your actual leaks

A benchmark tells you where the industry loses buyers on average. It does not tell you which of those leaks is active on your store, on your checkout, on the phone your buyers are actually holding, or how much revenue each one is costing you this month. That is the difference between reading a statistic and getting a diagnosis, and for a store without a data team, the diagnosis is the part that actually changes anything.