AI for ecommerce

AI for ecommerce is a phrase covering everything from writing product descriptions to forecasting demand, and most of it is worth nothing to a shop doing a few hundred orders a month. Five places pay reliably, they pay in a particular order, and the first of them is not exciting at all: the numbers your shop reports about itself have to stop disagreeing.

The five, in the order they pay

One stock figure that is true. The warehouse, the shop and the marketplace hold three numbers for the same item. Everything downstream inherits that gap — overselling, cancelled orders, a ranking penalty you never see. Fix this first or the rest is decoration.
The listings, written and kept. Titles, attributes, translations, the same product across four channels. Machines are genuinely good at this and it is the least risky place to start being ambitious.
The messages. Where is my order, can I change the size, why has it not shipped. Repetitive, answerable from your own systems, and the volume grows exactly when you have least time.
The returns. Every claim judged by the same written rules whether it lands at nine in the morning or six at night, with a batch to sign rather than a day to lose.
The daily read. What slipped yesterday, which product stopped selling, which campaign is burning money. A short list each morning beats a dashboard nobody opens.

What almost never pays in a shop this size

Demand forecasting on a few hundred orders a month. There is not enough history for a model to beat a person who knows the range, and the confident number is worse than no number.
Dynamic pricing without written limits. It will find the price that sells and it will not care what it does to your margin or your brand.
A recommendation engine before the stock figure is honest. It will happily push products you cannot ship.
Anything generating listings straight to a marketplace with no person approving. One bad batch and the account, not the software, takes the penalty.

Marketplace or your own shop — the difference matters

On your own shop you own the data and the rules, and the work is mostly connection: the store, the warehouse, the accounting, the messages. On a marketplace you rent both. The rules change without notice, the API decides what you may automate, and a mistake is punished by somebody else's algorithm rather than by your customers.

That is not an argument against marketplaces — most of the systems behind this site run on one. It is an argument for building the parts you keep: the stock figure, the record of what happened, and the rules, in your own store rather than inside somebody's platform.

Too early for you if

You are doing a few orders a day and you can see the whole business from one screen. Automation earns on repetition, and the repetition is not there yet.
Your catalogue lives in one platform that already does the job. Use it; the work starts when two systems disagree.
Nobody has decided which system is right when the numbers differ. That is a decision between people, and no software settles it for you.

Which of the five these are, running in a real shop

Recommendation shelves — the second and third of the five, on top of a stock figure already trusted: 968 products tagged by their real attributes, with the cross-sell shelves rebuilding themselves as the range moves — which is the order this page argues for.
Promotion guard — products kept out of forced marketplace discounts nobody agreed to, on a schedule, with the opt-out still on the owner's word.
New-arrivals radar — the category scanned on a schedule and the movers ranked, so a competitor's launch is not a surprise three weeks later.
Photo factory — product listings produced for Amazon and Shopify — the part of a shop where a machine is genuinely better than a tired person.

Honest answers

What does AI actually do for an online store?

Five things reliably: keeps one true stock figure, writes and maintains listings across channels, answers the repetitive messages, judges returns against written rules, and produces a short daily read of what moved. Almost everything else sold under this phrase is worth less than it costs.

Where should a shop start?

With the stock figure, however boring that sounds. Every other system reads it, so every other system inherits its errors, and fixing it is usually smaller than it looks.

Is this only for large stores?

No, but it is for repetitive ones. The test is not turnover: it is whether the same task happens dozens of times a week. A shop with fifty orders a day and four channels has more to automate than one with two hundred orders on a single platform.

Will it write my product descriptions?

Yes, and this is the safest place to be ambitious — with a person approving before anything reaches a marketplace, because the penalty for a bad batch lands on your account rather than on the software.

What about demand forecasting?

Rarely worth it below a few thousand orders a month. There is not enough history for a model to beat somebody who knows the range, and a confident wrong number costs more than no number at all.

Tell me where your stock number comes from

Which system holds it, which ones copy it, and how often they disagree. That single answer tells me which of the five is your first project and what it is worth. The first conversation is an hour and it is free.

Book that hour →