Generative AI in E-Commerce: Let Me Lay Out Four Accounts for You

Recently, a friend of mine who runs an e-commerce business texted me in the middle of the night: five hundred-plus products, five hundred-plus detail pages, and a copywriting team running on fumes until their eyes were bloodshot. So, he asked — should we give AI a shot at writing this?

I told him: you should have tried it ages ago. Nobody told you to carry the whole load yourself.

He pressed on: but is anything AI writes actually any good?

I didn't answer him directly. Instead, I threw two numbers at him, then told him a few stories.

Let's start with the numbers.

Back in 2023, a survey of marketing and advertising professionals in the US found that 37% were already putting AI to work. That was three years ago.

Research firm Gartner predicted around the same time that by 2025, 10% of all new data generated worldwide would come from generative AI — back in 2021, that figure was under 1%. Tenfold.

And another forecast: by 2030, the generative AI market alone will be worth more than $110 billion.

Now hear me out — this isn't a laboratory toy anymore. It's a fully grown industry.

So what exactly is generative AI?

Let's get one thing straight up front: you don't need to memorize a single technical term.

Put simply, you feed a machine a pile of existing material, it learns from that material, and before long it can generate brand-new things on its own — text, images, short videos, it can produce all of them.

Neural networks, variational autoencoders, generative adversarial networks — go ahead and scroll right past those. All you need to remember is one line:

AI is a pen that has learned to imitate. You set the prompt — it writes for you.

In e-commerce, that pen can do four big jobs for you — four accounts, if you will. Let me walk through them one by one.

Account No. 1: Writing

The first burden e-commerce folks can't escape is product copy.

Every product needs its features, its selling points, its specs, and its use cases laid out. One SKU, one complete set. Launch several hundred or a thousand products a year, and this is a workload nobody ever dares to count too closely.

So let me count it for you: a decent description, written carefully, takes 15 minutes — that's not asking too much. A thousand products means fifteen thousand minutes, or more than two hundred and fifty hours. That's a full-time editor working for a month or two.

And nowadays, a machine can do that job.

There's a technology in e-commerce called natural language generation. You feed it a product's features and specs, and it spits out a description on its own. Feed it a pair of noise-cancelling headphones, for instance, and it gives you something like this:

This noise-cancelling headset uses advanced noise-cancelling technology that filters out ambient noise, giving you a clear, immersive listening experience.

Read that out loud — wouldn't you swear the editorial team wrote it?

Go further: email subject lines, app push notifications — it writes those too. A company called Phrasee does exactly this for a living, and Domino's and eBay both use it.

So what does it actually save? In the end — manpower.

Ask the ops people at any e-commerce shop: every new product launch season, they're hunched over product spreadsheets all night, editing descriptions until they're cross-eyed. AI eats exactly that kind of soul-sapping, repetitive drudgery — which frees up the people who actually matter to do the work that really needs a human brain.

Account No. 2: Images

Copy done, there's still one more hurdle: product photography.

Studio rental, models, lighting techs, photo retouchers — run the whole process for one product and you're looking at several hundred to a thousand, at minimum. And with new launches coming in wave after wave, the visual team's budget is squeezed tight all year round.

AI's job on the image side is simple: give it a pile of existing product photos, and it generates new ones modeled on them.

The underlying technique is called a generative adversarial network. Picture it as two AIs dueling: one makes the images, the other picks them apart. After a few thousand rounds of going back and forth, whichever fake the critic can no longer catch gets released. The result: images that are almost impossible to tell apart from real photos.

And big brands are already playing this game.

Heinz had AI generate a photo of a ketchup bottle with a label nearly identical to its own, then proudly put it out there: "Look — this is what AI thinks ketchup looks like." Why so similar? Because Heinz bottles made up about half the training data.

Nestlé went further, using AI to rework a seventeenth-century Vermeer masterpiece to promote its yogurt brand. Mattel, meanwhile, just had AI draft the design concepts and marketing visuals for its new toys.

Add it all up, and the budget for photography and retouching can be squeezed down by a genuinely significant chunk.

Account No. 3: Selling

The first two accounts help you sell your products better — they save you money. This one is aimed squarely at making you more money.

The big one is recommendations.

Open a platform and it pulls up everything you've viewed and bought before, guesses what you might want next, and quietly nudges that item in front of you. The move looks subtle, but there's an entire algorithm working behind the scenes.

The most interesting player here is Stitch Fix, a clothing company in San Francisco with a very unusual business model: a personal stylist, delivered online. When you sign up, you enter your height, weight, and style preferences; the company digs through fashion trends, your body measurements, and your past rating feedback, then preps a slate of candidates for the stylist. The stylist picks out a few outfits, ships them to your door, you keep one, and send the rest back.

It doesn't make its money on the box — it makes it on a model that predicts what you want better than you can.

Amazon has pushed recommendations to the extreme. Forbes reported in 2021 that 35% of what consumers buy on Amazon is driven by recommendations.

Run that 35% against your own store and think about it: that's an entire extra block of sales appearing out of thin air.

Account No. 4: Creating

The first three accounts only touched the products you already have. This last one cuts deeper: it creates brand-new products for you.

It's called generative design.

You don't have to draft blueprints first. Give it a rough form, and it "grows" a range of possible shapes on its own — the rest is up to you to pick from.

New Balance is already on it. A Boston design-software company called Nervous System provides the tools, and New Balance uses them to generate the geometric structure of a sole, then fine-tunes it to each person's foot shape and support preferences. Wherever your foot needs a little extra support, the sole gets a little more material right there.

This isn't a prettier shell. This is AI driving straight into the deepest R&D layer of the business.

---

Back to the friend from the start of this story.

When I finished laying out the four accounts, he was silent on the other end of the call for a few seconds. Then he said: okay, let's start by trying it on the copy, then?

Of course, I said. Starting by trimming the smallest cost is the best move you can make.

AI won't run your business for you. It just does the things you want done — faster and cheaper.

And in an e-commerce race where fast fish eat slow fish, being "faster" is nearly the moat itself.

And to wrap up — here's to every ops person still patching up product copy in the dead of night: hand the pen to the machine soon, and finally get yourself a proper night's sleep.

Good night.