When Cross-Border E-Commerce Meets AI, a Lack of Fundamentals Is the Real Problem

A while back, an old friend in foreign trade came over for a meal. Halfway through, he set his chopsticks down and let out a sigh.

"I've bought just about every AI tool on the market that goes by a name. Tens of thousands of yuan — gone in a flash. Yet this past month, business is exactly where it was. Be straight with me: was that money just thrown away?"

I didn't rush to answer.

Because first, I want to walk you through a set of accounts.

A Closer Look at How Huge Cross-Border E-Commerce Really Is

According to the Ministry of Commerce, in 2023 China's total imports and exports of goods reached 41.76 trillion yuan, up 0.2% year over year.

Do you see what that number really means?

That year, traditional foreign trade was essentially treading water.

But in the same year, cross-border e-commerce imports and exports hit 2.38 trillion yuan, up 15.6% — with exports at 1.83 trillion yuan (up 19.6%) and imports at 548.3 billion yuan (up 3.9%).

Foreign trade's big engine slammed the brakes, while cross-border e-commerce hit the gas.

These are two entirely different worlds.

Look back another five years, and cross-border e-commerce's trade scale had expanded more than tenfold.

By 2024, that arena had taken on this shape: more than 120,000 cross-border e-commerce players, more than 200,000 independent sites, more than a thousand cross-border industrial parks, and more than 2,500 overseas warehouses — of which over 1,800 serve cross-border e-commerce alone, covering more than 22 million square meters of storage. Nationwide, cross-border e-commerce comprehensive pilot zones now number 165.

In short: shelves of Chinese goods now sit right at the doorstep of consumers around the world.

Is that striking? Striking.

But ask yourself: the bigger the ship, the more people crowd aboard.

The Way You Sell Has Changed, Too

In the old days of foreign trade, a factory just needed to hold on to its big customers. Large batches, big orders, one steady production line — that was comfort.

Now you have to open stores on Amazon, eBay, AliExpress, Lazada, and Shopee at the same time. Your customers are individual consumers scattered across the globe. Orders are smaller, they come more often, and they want them fast.

So three things are shifting at once.

Independent websites have taken off. Anker, China's first independently listed cross-border e-commerce company, saw 71.75% revenue growth from its own sites in 2022. By 2024 the independent-site market had reached 3.4 trillion yuan, roughly 35% of the cross-border e-commerce B2C market.

Livestream selling has gone global. Selling through TikTok has turned cross-border e-commerce from "buying from pictures" into "buying from people." Sellers no longer fly across continents to trade fairs; they close deals while chatting in a livestream.

The platforms are arming themselves with AI too. On the Alibaba international marketplace, more than 100 million product listings are now AI-generated, covering more than 40 e-commerce scenarios. And market researchers estimate that over 40% of Amazon's sellers worldwide come from China.

Channels, platforms — all covered. The biggest variable still to consider is AI itself.

What Is AI, Really?

In plain terms: hand it a pile of raw material, and it hands back text, images, audio, code. It also summarizes, sorts, answers questions, and creates new things.

Within cross-border e-commerce, AI is the engine that wipes out the repetitive, labor-heavy work.

Is the trend big? Big.

As of May 2024, users of AIGC-style apps in China had already topped 73 million — up 8× from a year earlier. Individual users are adopting it faster than anyone.

Policy is catching up. On August 15, 2023, the "Measures for the Management of Generative AI Services" formally took effect; by April 2024, 117 large models had been registered nationwide. Some institutions still project that generative AI will add close to 90 trillion yuan in global value by 2035 — with China's share possibly exceeding 30 trillion.

But the enterprise side is noticeably half a beat behind. Adoption of generative AI in manufacturing, healthcare, telecom, and retail still sits below the industry average.

Individuals went crazy first; companies are only warming up.

Which raises the question: if the trend is this big and the dividend is right there, why do small and mid-sized cross-border sellers still treat AI like a hot potato?

Four Obstacles

Between March and April 2023, a research team interviewed managers at 20 Chinese cross-border e-commerce SMEs and collected every complaint about using AI. Stripped down, there are four obstacles.

Obstacle one: homogenized content.

Spread the same product across Amazon, Lazada, Shopee, and TikTok — use the same AI tool and the same dataset — and will the copy it writes, the listing pages, or the livestream scripts differ?

Almost never.

Worse: the competition is using the same generation of AI and the same training data. Buyers then can't tell you apart from the shop next door. When they can't tell you apart, they start comparing prices; comparison eats your margin; and brand loyalty gets ground away, post by identical post.

Obstacle two: not enough people, and nobody knows how to use it.

AI tools come in endless varieties and update fast, and the cross-border industry was always short on talent. At many SMEs, management's grasp of AI is stuck at "I've heard ChatGPT is popular."

ChatGPT only launched in 2022, and the Alibaba international marketplace's AI assistant did not arrive until 2023. The whole industry is learning, adopting, and worrying at once.

Net effect: they don't know which tool to pick, how to connect AI to their business, or whether their own staff is even using it correctly.

Obstacle three: you can't verify the suggestions AI gives you.

AI makes things up with great confidence. It makes no promise of accuracy, and it takes no responsibility for a wrong answer.

For an SME, the real chokepoint is here: how do you check whether an AI suggestion holds up? Frontline operations and junior marketers are the ones using AI most — yet they're exactly the people with the least to compare it against. Confirming it against market feedback takes money, time, and effort — and for cross-border sellers, who already spend a large share of revenue on selling, that extra expense is a hard pill to swallow.

Obstacle four: data security.

Would you dare to feed AI your customers' purchase records, a new product's specs, or your pricing logic?

Reality: many sellers already do. And worse, one person alone can be running four or five AI tools at once — each one a different leak path. Once a competitor gets your product design and marketing playbook, the damage stops being a few thousand yuan.

All right, the four obstacles are laid out.

So what can be done?

It's Not About Buying More AI; It's About Mastering the Basics

My answer sounds a bit contrarian: before buying more AI, stabilize the basics first.

That same research also surfaced five actions to take. I'll go through them with you.

First: invest in training and resources — don't be cheap about it.

Small companies have limited resources, so start by wringing out what the platforms give away free. Amazon and Alibaba both offer AI training courses and free trials — use every one. When employees buy a paid tool with their own money, reimburse it fairly to encourage their self-study. Take part in the industry's AI skill competitions and salon events, and you can even team up with universities to set up "marketing sandboxes" where staff can try and fail freely in a cheap, risk-free, simulated space.

Second: build verification into your process.

Don't push AI output straight into the real battlefield. Run it first through a sandbox, confirm the expected outcome, and only then ship. The data side becomes a real advantage: big data digs insight, AI generates content based on it, and big data loops back to check whether the AI's output is right. And because cross-border transactions bring you face-to-face with overseas buyers, ask them directly how they feel about the AI-generated content — take the honest feedback instead of guessing for yourself.

Third: huddle together.

One person using AI is a singleton; a group using it together is an ecosystem. Co-build application ecosystems with platforms, technology vendors, and peers, sharing both the wins and the missteps along the way. On the public training-data platform the government is standing up, contribute your own non-sensitive data. A rising tide lifts all boats.

Fourth: give AI a role.

AI won't replace people, but it does replace a lot of repetitive toil. Manage it like a "registered virtual employee": define its scope — what it handles, and where a human must double-check — in writing. Watch carefully against over-delegation: handing an AI matters beyond its ability leads to churn and a mess. As its abilities grow, redraw its boundaries regularly, fold it into the talent pipeline, and adjust it dynamically.

Fifth: draw a clear bottom line.

Before you use an AI tool, confirm the provider has finished registration. The Measures that took effect on August 15, 2023 are the bottom line. Classify customer data by tier, deciding clearly what may enter an AI and what may not. Companies with the means should bring in legal counsel and run compliance audits periodically; those without should at least implement internal data supervision up front. The most expensive asset is customer data — and the biggest risk often sits close, with your own people.

---

Let me return to that friend at the beginning.

You asked me, once more, whether those tens of thousands were worth it.

I want to say this: the money you spend is where the answer gets generated from.

This AI wave has handed every seller a weapon. The dividend will not fall automatically to whoever buys the most weapons — it goes to whoever has laid the most solid foundations.

Fundamentals are dull to practice, and not showy. But on the day the storm comes down, they'll hand you one more umbrella than the next person.

So: were those tens of thousands worth it or not?

Once your fundamentals are complete, you'll know yourself.

May your next big order rise out of the second session of the livestream.