Why 19% of E-commerce Is Invisible - And How We Learned to See It
Most e-commerce reports miss 1.75 million stores running custom platforms. We built a system that finds them - not by detecting technology signatures, but by reading what a website actually does.
The Problem Nobody Talks About
Every e-commerce report you've ever read has a blind spot.
They count Shopify stores. WooCommerce stores. Magento, PrestaShop, Wix. They look for platform signatures in the HTML - a meta tag here, a JavaScript file there - and call it market research.
But what about the stores that don't use any of these platforms?
In our latest scan of 300 million domains, we identified 9.1 million active e-commerce stores. Of those, 1.75 million - 19.2% of the market - run on custom or unidentifiable platforms. They have products, shopping carts, checkout flows, and real customers. They just don't leave a recognizable fingerprint.
Every other industry report effectively ignores them.
Why Platform Fingerprinting Isn't Enough
The traditional approach to e-commerce detection is straightforward: scan a website's HTML, look for known signatures. Shopify stores load scripts from cdn.shopify.com. WooCommerce adds woocommerce CSS classes. Magento has telltale directory structures.
This works well for the platforms you know about. But it has three fundamental problems:
1. It misses everything custom. A Korean fashion brand running proprietary software looks like "just a website" to a fingerprint scanner. So does a Brazilian marketplace built in-house, or a Japanese retailer using a platform with 200 users. These stores are real, generating real revenue, but they're invisible.
2. It produces false positives. A domain with Shopify code in the HTML isn't necessarily a store. It might show "Coming soon." It might be a parked page. It might have been a store last month but isn't anymore. Platform detection tells you what software is installed, not whether anyone is actually selling.
3. It can't answer the real question. The question isn't "is this website running Shopify?" The question is "is this website an active e-commerce store?" These are fundamentally different questions, and they require fundamentally different approaches.
8 Years of Getting It Wrong
We started ShopRank with platform fingerprinting. It was the obvious approach, and for a while, it was good enough.
Then our clients started asking harder questions. "How many stores are there in the Czech Republic?" We could tell them how many Shopify stores, how many WooCommerce stores - but we couldn't tell them how many stores. The answer was always incomplete.
Our first attempt at a broader solution was rules-based. We wrote patterns: if a page has a shopping cart AND product prices AND an "Add to Cart" button, it's probably a store. This worked for English-language stores with standard layouts. It fell apart everywhere else.
A Japanese e-commerce page doesn't look like an American one. A Brazilian marketplace doesn't structure its checkout like a European boutique. An Arabic store reads right-to-left and uses completely different payment terminology. Rules that worked for one market were useless in another.
We tried expanding the rules. More patterns, more languages, more edge cases. The rule set grew from dozens to hundreds to thousands. Maintenance became a nightmare. Every new market we added broke something in an existing one.
The fundamental problem was clear: we were trying to encode human intuition about what "looks like a store" into rigid rules. But e-commerce doesn't look one way - it looks a thousand different ways across cultures, languages, and platforms.
We needed a system that could learn what a store looks like, not be told.
The Breakthrough
In 2025, we rebuilt our classification system from the ground up.
Instead of telling our classifier what a store looks like, we showed it. The classifier doesn't look for platform signatures. It reads the actual page and answers a different question: "is this website selling physical products you can have shipped to you?" - not "what software is it running?"
A store with a checkout and courier delivery counts. A SaaS site, a restaurant, a car dealership - they don't, even if they have e-commerce code.
It works across dozens of languages and markets - a Korean fashion store, a Brazilian marketplace, and a Czech electronics shop are all recognized the same way. Not because we wrote rules for each language, but because commerce looks like commerce everywhere once you know what to look for.
What This Means for Our Data
When we say we track 9.1 million e-commerce stores, that number includes 1.75 million stores on custom-built platforms - stores that don't show up in any platform detection report. They account for nearly 1 in 5 online stores worldwide.
Custom is the third-largest segment - bigger than PrestaShop, Magento, Wix, and Squarespace combined. And invisible to every platform detection tool on the market.
If you're a payment provider evaluating market size - you're undercounting by 19%. If you're an investor comparing platforms - you're missing a fifth of the competitive landscape. If you're a logistics company planning coverage - you have a blind spot larger than any single platform except Shopify and WooCommerce.
What the Non-Stores Actually Are
When our classifier says a domain with Shopify or WooCommerce code is "not a store," what is it looking at?
We dug into this. Among all domains with detected e-commerce software that we classified as "not a store," 95 -98% are real websites with substantial content - pages with titles, descriptions, and 200+ words of text. Less than 1% are "coming soon" pages or placeholders.
So what are they? Businesses, blogs, portfolios, restaurants, schools - real organizations with real websites that happen to have e-commerce software installed somewhere in their stack. A church running WordPress with a dormant WooCommerce plugin. A design agency whose developer added Squarespace Commerce during setup and never configured it. A corporate intranet running Magento because it's good enterprise software, not because anyone's buying shoes.
The distinction matters: having shopping software isn't the same as selling things.
WooCommerce is the clearest case: more than half of all classifiable WooCommerce domains aren't stores. We published a deep dive into that analysis.
The Journey Continues
E-commerce keeps changing - social commerce, headless storefronts, AI-generated product pages. Our classification system evolves with it.
But the core insight from 8 years of building detection systems holds: you can't understand e-commerce by counting platforms. You have to understand what selling online actually looks like - in every language, on every platform, and on no platform at all.
That's what we do.
Data and analysis by ShopRank. We scan 300M+ domains and track 9.1M e-commerce stores across 340+ platforms, updated monthly.