AI for online retailers: the use cases that move revenue

If you run a small online shop, AI earns its keep in three places: the product copy, the customer questions, and what each shopper sees. Most UK businesses have not started. Pick one of those levers, change the work around it, and measure what it does to revenue.

That last part is where the money is. Buying a tool is easy and changes nothing on its own. Rebuilding one routine around it, so the tool does the first draft and a person keeps the judgement, is what eventually shows up in the numbers.

The rest of this is about which lever to pick, what a realistic gain looks like, and the trap that swallows most of the effort.

Most shops have not started, and that is the opportunity

Adoption is real, but thinner than the noise suggests. The Office for National Statistics found that a quarter of UK businesses were using some form of AI in late December 2025, up 15 percentage points since the question was first asked in late September 2023. Among businesses with 250 or more staff the figure was 44%. Around one in seven firms said they planned to adopt AI within three months.

Read that as a small retailer and it says something useful. The larger competitor is close to twice as likely to be using AI as the average business. What the big firm has is not better software. The tools are the same ones you can open this afternoon. What it has is someone whose job it is to change the process around the tool.

That gap, between owning a tool and changing how the shop runs, is where the margin sits.

Meanwhile, shoppers are already using AI to find you

The demand side has moved faster than the supply side. Adobe reported in June 2026 that traffic to US retail sites from AI sources rose 1,324% between October 2024 and May 2026. For scale, Adobe's analysis of the 2025 holiday season, covering more than a trillion visits to US retail sites, put online spending at $257.8 billion, up 6.8% on the year, with shoppers embracing generative AI tools to find what they wanted.

Those are US figures and the UK picture will differ. The direction is not really in doubt, though. A growing share of the people landing on a product page were sent there by a machine that read the page first.

The practical consequence is dull and reassuring. Product copy that is clear, specific and accurate does better with an AI tool for the same reasons it does better with a person: it says what the thing is, who it suits, and what it does not do. The two audiences have not split.

The four use cases that move revenue

Product copy is the first, and usually the best place to start. A shop with hundreds of items almost always has thin, duplicated or missing descriptions, because writing them properly is a job nobody has time for. AI drafts a first version from your own attributes, dimensions, materials, fit, use, and a person edits for accuracy and tone. The gain is not literary. It is coverage: every product finally has copy worth reading.

Customer support is the second. A large share of what lands in the inbox is the same dozen questions about delivery, returns, sizing and stock. A well-fed assistant, pointed at your real policies, answers those quickly and hands the rest to a person. Handled well, it shortens replies and stops the queue eating the day. Handled badly, it invents a returns policy you do not have.

Personalisation is the third, and the one most often oversold. For a small store this rarely means a recommendation engine. It means sensible, specific groupings: what to show a repeat buyer versus a first-time visitor, which email goes to someone who browsed and left, what sits in the "you may also like" slot. AI helps you draft and vary that copy at a volume a small team could not otherwise manage.

Merchandising and buying is the fourth, and the least glamorous. Summarising reviews to find the recurring complaint, spotting which lines quietly stopped selling, drafting the questions to ask a supplier. This is pattern work on data you already hold, and it tends to pay back in decisions rather than clicks.

In all four, the tool produces raw material and a person keeps the judgement. That line is what separates a shop getting real value from a shop quietly taking on risk. One practical rule worth setting early: customer names, addresses and order details stay out of consumer chat tools that your business has no agreement with.

The bolt-on trap

The common failure is not choosing the wrong tool. It is adding a tool without changing anything else.

A shop buys an AI writing add-on, generates four hundred product descriptions in an afternoon, and publishes them unread. Some are wrong. A few contradict the spec table directly above them. Returns tick up, because the copy promised a fit the garment does not have. The tool did exactly what it was asked. Nobody changed the workflow around it, so nobody caught the errors, and the shop is now worse off than before.

Run that same job properly and it looks different. The same four hundred drafts get generated, then a person reviews them in batches against the spec sheet, starting with the fifty products that earn the most. The remaining three hundred and fifty stay in draft until they are checked. It takes a fortnight rather than an afternoon. It also works.

The tool is the cheap part. The review step, the person who owns it, and the decision about what "good" looks like are the actual work, and they are what the four-hundred-in-an-afternoon version skips.

Pick one lever, and measure it

Do not run AI across the whole shop at once. Pick a single revenue lever and give it a fixed run.

Choose the one where the constraint actually binds. If products sell once someone reads the page, but half the catalogue has no page worth reading, that is product copy. If support is the thing that stops you doing anything else, start there. Choosing that first job well matters more than choosing the tool, which is why our note on how to choose an AI use case is a useful next read.

Then measure the thing that pays, not the thing that is easy to count. Hours saved on drafting is not the number. Conversion on the pages you rewrote, first-response time and the rate of repeat contacts, revenue per email sent: those are the numbers. Time saved only becomes value when you can name what it was spent on instead, which is the distinction we draw between productivity and business value.

Give it six to eight weeks. Compare the products or tickets you touched against the ones you did not. That comparison, imperfect as it is, tells you more than any vendor case study.

A starter stack

The stack matters less than the discipline, and for most small retailers it is short. One business-grade general assistant for drafting copy, summarising reviews and answering internal questions. Whatever AI features your existing platform and helpdesk already include, which are usually enough to test the idea before you buy anything new. One person named as the owner, who reviews outputs and decides what ships.

That is it. No new platform, no migration. If someone is proposing a rebuild before you have proven a single lever, that is a signal, and our note on how to choose an AI consultant covers what to ask them.

What to do next

Open your product catalogue and sort it by revenue. Take the top twenty items and read their descriptions as a stranger would. Mark every one that is thin, duplicated, or contradicts the specification beside it. That list is your first AI job: draft new copy for those twenty, have one person check each against the spec, publish, and watch conversion on those pages for six weeks against the rest of the catalogue.

Twenty products, one owner, one number to watch. If it works, widen it. If it does not, you have lost a fortnight rather than rebuilt a shop around something that never paid.

If it would help to work out which lever is worth pulling in your shop, and what to measure, book a call and we will think it through with you.

Common questions

How should a small online retailer start with AI?

Start with one revenue lever, not a platform. For most small shops that is product copy, because thin or missing descriptions are common and the fix is measurable. Draft with AI, have a named person check every output against the specification, publish to your highest-earning products first, and compare conversion on those pages against the rest of the catalogue over six to eight weeks.

Where does AI actually move revenue in ecommerce?

In four places: product copy, customer support, personalisation of what each shopper sees, and merchandising decisions such as spotting recurring complaints in reviews. The gain comes from changing the routine around the tool, not from owning the tool. A shop that generates hundreds of descriptions and publishes them unread usually ends up worse off, with returns driven by copy that promised something the product does not do.

Does AI change how customers find an online shop?

It is starting to. Adobe reported traffic to US retail sites from AI sources rising 1,324% between October 2024 and May 2026. UK numbers will differ, but a growing share of shoppers now reach a product page through a tool that read the page first. The response is not a trick: clear, specific, accurate product copy serves both the machine and the person reading it.