Industry Research · Ecommerce & AI Commerce
The checkout page is becoming optional. The commerce stack behind it isn’t.
Cart abandonment still hovers near 70%, but the next battle isn’t fewer form fields. Wallets, tokenized credentials, and AI shopping agents are moving discovery, cart, and payment beyond the merchant’s checkout page, and only some software can follow.
The bottom line
- The 70% abandonment number hides two different problems. Baymard puts the average at 70.22%, but about 43% of it is people just browsing, not a checkout you can fix. The rest is fixable friction. Two problems, not one ↓
- The biggest conversion gains moved below the visual layer. Wallets and tokenized payment credentials now lift conversion and authorization without redesigning the storefront. Below the page ↓
- Your next customer may never open your storefront. AI-referred shoppers already convert better than humans, and Google, OpenAI and Visa are building the rails for agents to buy directly. Programmable checkout infrastructure is best positioned to sell through them. The agent checkout ↓
For years, the checkout advice was the same: cut the form fields, add a progress bar, and abandonment will fall. The forms got shorter. Abandonment didn’t move. It has sat near 70% for more than a decade, which is a strong hint the real problem was never the number of fields.
Two forces are finally shifting it, and neither is a shorter form. The first is that identity and payment credentials, not layout, increasingly decide whether a checkout converts. The second is stranger: a fast-growing share of shoppers now arrive through AI, and new agentic-commerce protocols are starting to let them buy without ever opening your checkout page.
The 70% abandonment number hides two different problems
Baymard has tracked cart abandonment for over a decade, and it barely moves. The mistake is treating all of it as sales you could win back.
The Baymard Institute puts the average documented cart abandonment rate at 70.22%, a meta-analysis of more than 50 studies that has held near that level for years. That figure gets quoted as if 70% of your revenue is walking out the door. It isn’t. Baymard’s own survey work finds that about 43% of shoppers abandon simply because they were browsing or not ready to buy, behavior no checkout change was ever going to capture.
Strip those out and what remains is genuinely fixable, and it is mostly about cost and effort, not button placement.
| Reason | Share who cite it |
|---|---|
| Just browsing / not ready to buy | 43% |
| Extra costs too high (shipping, tax, fees) | 39% |
| Site wanted them to create an account | 19% |
| Checkout too long or complicated | 18% |
Source: Baymard Institute. “Just browsing” is a single main reason; the rest are multi-select among shoppers who abandoned for a reason other than browsing, so they don’t sum to 100%.
Baymard estimates a typical large ecommerce site could lift conversion by roughly 35% through checkout improvements alone, once you aim at the fixable half instead of the browsing half.
Don’t optimize against “70% abandonment.” Separate the shopping behavior you can’t fix from the checkout friction you can, then judge checkout platforms on how well they kill the fixable half: costs shown early, guest checkout, trusted payment, fewer steps.
The biggest gains have moved below the checkout page
The highest-leverage improvements aren’t visual anymore. They live in identity and payment credentials, underneath the form.
Start with identity. Shopify says its Shop Pay wallet can convert up to 50% higher than guest checkout, and that simply offering Shop Pay drove about a 5% lower-funnel lift even for shoppers who didn’t use it. Those are vendor-published figures from a Big Three consulting study Shopify commissioned in 2023, so treat them as Shopify’s numbers, not a law of nature. The more useful part for buyers: Shopify now offers Shop Pay on other commerce platforms, so a store can bolt on a better checkout component without replatforming everything. That is composability in a form you can actually picture.
Then the payment credential itself. Visa reports that Click to Pay’s tokenized card-not-present transactions carry a 4.3% average authorization-rate lift versus raw card numbers, because the network pre-validates the token before it reaches your bank. At scale that compounds: Visa says its first 10 billion tokens generated more than $40 billion in incremental ecommerce revenue globally and saved about $650 million in fraud in a single year. Those are Visa’s own measurements, but they’re a different class of evidence than a checkout-tips blog.
That matches what I keep seeing when we test these tools for our checkout-platform rankings: the ones that pull ahead aren’t fixing the checkout page, they’re fixing what happens behind it. Airwallex is the one that stuck with me. I pushed a test order through from overseas, and it just charged the buyer in their own money and paid me out in mine, no plugin, no doing exchange-rate math in my head. It takes something like 160 local ways to pay, so most shoppers see an option they already trust. They never notice any of it, and that’s the whole point.
When you compare cart and checkout platforms, don’t stop at templates and button placement. Ask about accelerated wallets, network tokenization, authorization optimization, and whether you can swap payment components without a rebuild.
Your next customer may never open your storefront
This is the part that dates the whole conversation to 2026. A fast-growing share of shoppers arrive through AI, and the rails for agents to buy on their own now exist.
Adobe, which tracks more than a trillion visits to US retail sites, found that AI-referred traffic grew 393% year over year in the first quarter of 2026. The bigger surprise is quality. In March 2025 that AI traffic converted 38% worse than everyone else. By March 2026 it converted 42% better, with revenue per visit running 37% above non-AI traffic. That is a full reversal in a single year. Adobe also found the obvious catch: it scored the average retail homepage at 75% machine-readable but product pages at just 66%, meaning roughly a third of a typical product page can’t be read cleanly by the very agents now sending their best traffic.
The rails are being built fast, and by the biggest names. Google introduced a Universal Commerce Protocol with Shopify, Etsy, Wayfair, Target and Walmart to let agents work across discovery and purchase. OpenAI shipped an Agentic Commerce Protocol so an agent can complete a purchase while the merchant stays merchant-of-record on its existing payment stack. And Visa built a Trusted Agent Protocol so a merchant can tell an authorized shopping agent apart from a malicious bot.
Here is where the old “should my checkout be headless?” debate finally pays off. An agent clicking through a human registration form is brittle; it works far more reliably against a checkout it can call programmatically. That is the same modular architecture vendors have pitched for years, except the reason has changed: not shipping features faster, but staying sellable through interfaces that don’t look like your storefront. The MACH Alliance’s enterprise research, from an association that advocates composable architecture, reports that organizations further along in composable adoption are far more likely to say they can deploy AI successfully.
It is worth remembering how we got here. Back in 2020, Gartner predicted that companies adopting a composable approach would outpace competitors by 80% in the speed of new feature implementation by 2023. No published follow-up ever measured it, and vendors have recycled the line since. The prediction came true in a way nobody framed at the time: the competitor you can’t keep up with may not be another retailer at all, but an agent that transacts through APIs you either expose or you don’t.
You can see the same tradeoff in the platforms themselves. WooCommerce runs something like 38.7% of all online stores, and people love it because you own everything: your checkout, your customer list, the code itself. That’s also the kind of open, plug-into-anything setup an AI agent can actually work with. But our checkout-platform research is blunt about the flip side, and so am I. I ran a store on it for a while, and the freedom was real, but so was the night I realized that if my checkout sprang a security hole, nobody but me was going to patch it.
As checkout logic moves into the browser and across services, the payment page itself becomes the target. The PCI Security Standards Council’s 2025 guidance focuses on authorizing every script on a payment page, checking its integrity, and monitoring it for tampering, because browser-side scripts are the main e-skimming risk.
Judge a platform on two futures at once: how well it removes human checkout friction today, and whether its catalog, cart, and checkout are exposed as clean APIs that an AI agent, and a payment page you can lock down, can actually use.
Who this matters to
Best use: lifting conversion without a full replatform. Must have: accelerated wallets, tokenization, guest checkout, API access. Watch out: confusing total abandonment with the fixable kind.
Best for ecommerce brands →Best use: authorization and routing at scale. Must have: multiple processors, network tokens, payment analytics, resilient APIs. Watch out: one point of auth-rate difference is real money at volume.
Best for ecommerce businesses →Best use: stacks whose parts swap without client rebuilds. Must have: webhooks, modular components, clean docs. Watch out: integration sprawl can erase the theoretical benefit of composability.
Best for marketing agencies →Best use: being discoverable and buyable through agents. Must have: machine-readable product data, structured pricing, agent-compatible checkout. Watch out: pages a human can browse but a machine can’t parse.
All checkout platforms →Where to take this next
The evidence changes what “best checkout platform” means. A strong platform still has to remove human checkout friction, but increasingly it also needs portable identity, tokenized payments, clean product data, and programmatic access to cart and checkout. Our rankings judge platforms on that fuller bar.
How we measured this
This report synthesizes publicly available research rather than presenting original transaction data. We prioritized independent ecommerce usability research (Baymard), direct commerce telemetry (Adobe, based on more than a trillion US retail visits), official payment-network data (Visa), technical standards (PCI SSC), and current platform documentation and protocol announcements from Google, OpenAI and Shopify.
Performance figures published by Shopify and Visa are identified as vendor-reported or vendor-sponsored, not independent results. The MACH Alliance figures are industry-association research from a body that advocates composable architecture, and are enterprise-wide rather than ecommerce-only. The 80% feature-speed figure is a 2020 Gartner prediction, not a measured outcome. Product and protocol capabilities were checked against current documentation in August 2026.
Sources & references
Every figure links to its primary source. Baymard and Adobe are independent research; Shopify and Visa performance figures are vendor-reported and labeled as such; MACH Alliance is industry-association research; the 80% is a 2020 Gartner prediction, not a measurement.
- Baymard Institute. Cart abandonment statistics. Average documented rate of 70.22% across 50+ studies; about 43% of abandonment is just browsing. Independent research.↗
- Baymard Institute. Checkout usability research. Reasons for abandonment (extra costs 39%, forced account 19%, checkout too long 18%) and the ~35% conversion uplift available from checkout improvements. Independent research.↗
- Adobe. AI traffic analysis (1T+ US retail visits). AI-referred traffic up 393% YoY in Q1 2026, converting 42% better in March 2026 (from 38% worse a year earlier); much retail content still isn’t machine-readable.↗
- Shopify. Shop Pay. Vendor-published Big Three study (2023): converts up to 50% higher than guest checkout, ~5% lower-funnel lift; Shop Pay is now available on other platforms. Vendor-reported.↗
- Visa. Click to Pay for merchants. Tokenized card-not-present transactions show a 4.3% average authorization-rate lift versus raw card numbers. Vendor-reported.↗
- Visa. 10 billionth token announcement (2024). Visa tokens generated $40B+ in incremental ecommerce revenue globally and saved ~$650M in fraud in a year. Vendor-reported.↗
- Google. Universal Commerce Protocol. Co-developed with Shopify, Etsy, Wayfair, Target and Walmart to let agents work across discovery and purchase.↗
- OpenAI. Agentic Commerce Protocol / Instant Checkout. Agents complete purchases while the merchant stays merchant-of-record on existing payment infrastructure.↗
- Visa. Trusted Agent Protocol. Lets merchants distinguish authorized shopping agents from malicious automated traffic.↗
- MACH Alliance. Enterprise Technology Report. Composable-mature organizations are far likelier to report deploying AI successfully. Industry-association research (pro-composable); enterprise-wide, not ecommerce-only.↗
- PCI Security Standards Council. Payment-page security guidance (2025). Authorize, integrity-check and monitor payment-page scripts to prevent e-skimming.↗
- Gartner. Composable business research (2020). Prediction: composable adopters would outpace competitors by 80% in feature-implementation speed by 2023. Analyst forecast, widely recycled by vendors; no published follow-up measurement.↗
