Welcome to the shop floor, with an AI sidekick
Hi, and welcome to your first level! Before we start writing prompts, let's get a clear picture of what an AI assistant actually is. What's it good at, and where does it trip up?
You've probably heard the names: ChatGPT, Gemini, Claude, Microsoft 365 Copilot. They're made by different companies, but at heart they all do the same kind of job. You type a request in everyday words, and you get back a reply in everyday words.
So how does that work? Here's a simple way to picture it.
Think of a very well-read new starter
Imagine a new colleague who has read millions of web pages, manuals and books. They're quick and polite, and they never get tired. Ask them to draft a shelf sign and you'll have it in seconds.
There's a catch, though. This new starter has never set foot in your shop. They don't know your stock, your prices or your returns policy unless you tell them. And they really don't like saying "I don't know".
That's close to the truth. An AI assistant writes its answer by predicting which words are likely to come next, based on patterns it learned from huge amounts of text. That's why it's brilliant with language. It's also why it can sound completely sure of itself and still be wrong.
OpenAI's own help page on ChatGPT's accuracy says this plainly. The model can produce incorrect or misleading answers, it can sound confident even when it's wrong, and it can make up facts, quotes or sources. People call this a "hallucination". OpenAI's advice is to verify important information using reliable sources.
Where AI tends to go wrong
What does that look like in a shop? Here are the usual suspects:
- Made-up details. It might invent a fabric blend, a battery life or a warranty length that sounds believable.
- Out-of-date or wrong-context facts. It doesn't automatically know today's price or this week's promotion.
- Confident policy answers. It might tell a customer they're entitled to something your store doesn't offer.
That last one matters more than you might think. Back in February 2024, a Canadian tribunal ruled against Air Canada after its website chatbot gave a customer wrong information about bereavement fares. The airline argued the chatbot was responsible for its own words, and the tribunal rejected that. The company was responsible for everything on its website, chatbot answers included.
The lesson for us is simple. If AI gets it wrong, the business still owns the mistake. And on the shop floor, you're the business.
So what's it actually good at?
Plenty! AI shines at jobs built on words and structure:
- First drafts of product descriptions, signs and social posts
- Summarising a long supplier document into a few key points
- Turning rough notes into a tidy rota or task list
- Rewording a reply to a customer so it sounds calmer and clearer
- Brainstorming display themes or questions to ask a supplier
Notice the pattern? AI gives you a head start, and then you add the facts and the judgement.
Sorting your tasks: three buckets
Here's a handy habit for the whole course. When a task lands on your desk, ask yourself which bucket it belongs in.
Bucket 1: AI can help. These are low-risk, creative or formatting jobs, like ideas for a window display theme or a friendlier way to word a staff notice. If the draft is a bit off, nobody gets hurt.
Bucket 2: AI with careful checking. These are jobs where facts matter, like a product listing, a customer email about a delivery, or a summary of sales figures. AI can draft them, but you check every fact against the real source: the label, the system or the policy.
Bucket 3: Must stay human. These are decisions about people, money, the law or safety. Agreeing a refund, deciding whether to sell an age-restricted item, handling an upset customer face to face, or disciplining a colleague. AI might help you prepare, but a person decides and a person is accountable.
Try one now. Where does "telling a customer whether a jacket is waterproof" go? Bucket 2, at least. AI could word the answer nicely, but the fact itself has to come from the label or the product data, not a guess.
How are retailers really using it?
Let's look at some real, dated examples. That's not just me showing off! Figures in this field go out of date fast, so the date always matters.
Walmart, 24 June 2025. Walmart announced a suite of AI tools for 1.5 million store associates in its associate app. It said its AI task management tool had cut shift planning time from 90 minutes to 30, and its translation tool covered 44 languages. Walmart also stressed that AI's full potential comes when it's paired with the strengths of its people.
Amazon, 30 October 2025. On Amazon's quarterly earnings call, CEO Andy Jassy said 250 million shoppers had used its Rufus AI shopping assistant in 2025. He said Rufus was expected to generate over $10 billion in annual incremental sales.
UK retailers, June 2026. A survey of more than 100 senior UK retail executives, reported by IT Brief UK on 8 June 2026, found that 58% of UK retailers were investing in generative AI tools for store workers. Interestingly, 49% were also investing in hiring more store staff, so the tools were arriving alongside people, not instead of them.
Whenever you quote a figure like these, say who published it and when. "Walmart said in June 2025" is far more trustworthy than "apparently Walmart..."
Key takeaways
- AI assistants predict likely words. That makes them great with language, but they can be confidently wrong ("hallucinations").
- They don't know your shop's stock, prices or policies unless you give them that information.
- Sort every task into three buckets: AI can help, AI with careful checking, or must stay human.
- If AI gives a customer wrong information, the business is still responsible, as the February 2024 Air Canada ruling showed.
- Always quote AI statistics with their source and date, because they go out of date fast.