What AI actually looks like in a small shop

Independent retailers are told constantly that AI will save them, usually by people who have never run a shop. The advice tends to be either abstract or priced for a chain of two hundred stores.

Here is the practical version, based on what we actually build.

Product descriptions

If you list products online, writing descriptions is probably the single largest recurring content cost you have. A hundred new lines a season, each needing a title, a description, specifications and search-friendly copy.

This is a genuinely solved problem. Photograph the item, add a few facts, and get a draft in your own house style in seconds. You still check it, because the AI does not know that this particular supplier's sizing runs small. But you are editing rather than writing, and that is roughly a tenfold difference in effort.

Answering the same eleven questions

Every retailer has the same handful of questions consuming most of their support time. Do you deliver to my postcode. Is this in stock in blue. What is your returns window. Can I collect today.

An assistant connected to your actual stock and delivery data handles all of these instantly, at any hour. The gain is not only labour. It is that a customer at 10pm gets an answer rather than leaving.

Buying better

Most independent stock decisions are made from memory and instinct. That works until it does not, and the cost shows up as either dead stock on the shelf or an empty peg during your best week.

Forecasting from your own sales history, adjusted for seasonality and local events, is unglamorous and consistently valuable. It will not be perfect. It only has to be better than the guess.

Knowing what is really happening

Most retail systems produce reports that answer questions nobody asked. Being able to type "which lines have not sold in eight weeks" and get an immediate answer changes how often you ask.

The value is in the frequency of the asking. Questions that take forty minutes to answer get asked quarterly. Questions that take ten seconds get asked daily, and that is where the decisions improve.

What to ignore for now

AI-generated imagery for products you actually sell. Customers spot it and it erodes trust.

Fully automated pricing. The reputational risk of an algorithmic mistake on price is considerably larger than the margin gain.

Anything sold to you as "AI-powered" without a clear explanation of what it does. That phrase is now doing an enormous amount of marketing work and very little technical work.

A sensible order

Start with product content, because the saving is immediate and the risk is nil. Add customer support next, because it buys back evenings. Then move to stock forecasting, which needs a year of decent data to be worth much.

Three projects, spread across a year, each paying for the next.