Every cart abandonment guide tells you the same things. Shorten the form. Hide the shipping cost until later. Add Apple Pay. Send three recovery emails over seven days. The advice has barely shifted in a decade, and neither has the result.

Baymard Institute has been tracking global cart abandonment since 2006. Their figure for 2024 is 70.19%. Their figure for 2023 was 70.19%. So was 2022, give or take a tenth of a point. Two-thirds of shoppers still walk away from a full cart, year after year, while the industry congratulates itself for shaving fields off checkout forms.

There's a reason for this, and it's uncomfortable.

The audit nobody runs

Walk into almost any conversion rate consultancy and you'll get the same playbook: a Hotjar replay, a checkout heatmap, a form-field analysis, maybe a Crazy Egg scroll map if the consultant is feeling thorough. The deliverable lands on your desk as a PDF with twelve recommendations, all of them about layout, copy, or trust badges.

Useful. Sometimes. But it stops at the browser.

What happens after the user clicks "Place Order" is where most online stores quietly bleed conversions, and it's almost never on the audit. The browser sends the request. The server has to validate the cart, check stock against your product database, recalculate taxes and shipping, run the payment gateway handshake, and write the order to disk. On a shared host with a noisy neighbor and a database that hasn't been tuned since launch, that single click can take four, five, six seconds. Sometimes it just times out.

The shopper doesn't see "your server failed." They see a frozen screen. So they close the tab and Baymard records another abandoned cart, filed politely under "checkout was too long or complicated."

It wasn't the form. It was the server.

What the performance numbers actually say

Google's recommendation for Time to First Byte sits at under 200 milliseconds for cached pages and under 600 milliseconds for dynamic ones. Checkout pages, by definition, can't be cached: they're rendered fresh for every user, every session, every cart. Most shared-hosted ecommerce stores measured in 2025 come in at 1,200 to 2,000ms on checkout endpoints, sometimes worse during evening traffic.

Cloudflare's internal performance research found that pulling TTFB from 1,200ms down to 400ms lifts conversion by 7 to 12 percent. Deloitte's 2020 study with Google, working across 30+ retail brands and millions of sessions, found that a 0.1-second improvement in mobile site speed raised retail conversions by 8.4% and average order value by 9.2%. Amazon's famous internal calculation, every 100ms of latency costs them 1% of sales, is from 2007. It still gets quoted because nobody has published a number that contradicts it.

So why does almost every conversion guide skip past this?

Because backend performance isn't a marketing problem. It's a hosting problem, a database-tuning problem, sometimes a server-sizing problem, and conversion-rate consultants don't sell those services. The PDF gets longer when the recommendation is "redesign the checkout." It gets shorter and more awkward when the recommendation is "your server is the bottleneck."

The traps that look like UX failures

Every ecommerce platform (Shopify, WooCommerce, Magento, BigCommerce, and the self-hosted carts in between) ships with assumptions that turn ugly under load. Default caches that weren't built for traffic spikes. Database queries that scan instead of indexing once your catalog passes a few thousand SKUs. Session tables that grow without anyone pruning them. Payment plugins that block the request thread for half a second waiting on a gateway timeout.

By the time a store hits any kind of real traffic, say the modest spike of a paid Meta campaign on a Thursday evening, the server is doing more work than the hosting plan was sized for. Nothing visibly breaks. The site loads. The cart works. It's just slow enough that a meaningful fraction of users give up.

I've seen this pattern in stores doing €15,000 a month. The owner blames the theme, then the extensions, then their checkout flow. They install three different upsell apps. They redesign the cart page. Nothing moves. Then they migrate to a host that can actually handle the load, and conversions jump 18% in the first week.

Not because anything visible changed. Because the server stopped making people wait.

Where to start when you suspect the problem is upstream

If you've already optimized the obvious things (short forms, guest checkout, transparent shipping, Apple Pay enabled) and your conversion rate still sits at the category average, run two tests before you touch the design again.

First, open WebPageTest from three different locations and check your TTFB on the checkout endpoint, not the homepage. Homepages cache aggressively. Checkouts don't. If you're north of 800ms, the server is your problem.

Second, enable your platform's slow query log for 48 hours during your peak window. If you see database queries against product or session tables taking more than 200ms, you have an indexing or hosting issue, not a UX one. The fix isn't another A/B test on button color.

Once you've confirmed the bottleneck is infrastructure, the next decision is whether to optimize the host you have or change hosts entirely. That depends mostly on what kind of plan you're on and how patient your provider's support team is when you ask them to enable PHP OPcache and persistent database connections. For merchants who've concluded the answer is "change," Gizmodo's editorial team published a comparison earlier this year that walks through current web hosting options with the kind of practical detail that's useful when you're picking infrastructure for a real store, not a portfolio site. It's one of the few comparisons that doesn't treat shared hosting as adequate for ecommerce by default, which, given what the conversion data says, it isn't.

The thing the abandonment statistics don't measure

Baymard's 70.19% figure has held flat for thirteen years not because users have stopped caring about checkout design, but because the part of checkout design they actually experience as friction has been mostly solved. The forms are short now. The buttons are big. The trust badges are everywhere. Apple Pay and Shop Pay have removed entire categories of typing.

What hasn't been solved is the half-second pause between clicking the button and seeing the next screen. That pause is invisible in heatmaps. It doesn't show up in form analytics. It looks identical, on a session replay, to a user changing their mind. And every conversion-rate consultant in the industry has a financial incentive to keep blaming the form instead of the server, because the form is what they sell solutions for.

If you've redesigned your checkout twice and the needle hasn't moved, the redesign isn't the issue. Run the TTFB test on Tuesday evening at 8 PM, when your audience is actually shopping, and see what your stack does under real conditions. That's the audit almost nobody runs, and it's the one most likely to tell you why two-thirds of your shoppers leave.