Eight jobs a small shop can copy this month: writing product descriptions, building reorder lists from sales, listing slow movers for markdown, replying to customer questions, planning promotions, cleaning up product photos, reading supplier paperwork and posting to a Google profile. Each runs on ChatGPT, tools already in your till or shop platform, or a free plan.
None needs more than about $30 a month in new subscriptions, and several need nothing. These are deliberately small examples. Demand forecasting across thousands of stores or cameras scanning shelves depend on data and budgets one shop doesn't have. Each of these runs on a spreadsheet export, a phone and one afternoon, and each has a check you must make before trusting the output.
1. Product descriptions written from your stock sheet
The cheapest first job is also the safest, because you read every word before it goes live. A shop selling furniture from a small workshop might have 120 products and a spreadsheet of facts: materials, dimensions, finish, delivery type and price. The routine is to give ChatGPT, Claude or Gemini a batch of rows, a house style and one hard rule: use only what's in the row.
You write product pages for a furniture shop. For each row below, write a title of up to 70 characters, a description of 50 to 70 words and three bullet points. Use only facts that appear in the row. If a fact isn't in the row, leave it out. Do not use "handcrafted", "sustainable" or "premium" unless the row says so. Plain British English, no exclamation marks.
Row: Oak side table | solid oak top and legs | 45 x 45 x 55 cm | oiled finish | flat-pack, about 10 minutes to assemble | $149A good result, illustrated here, reads: "Oak Side Table, 45 cm Square, Oiled Finish. A solid oak side table, 45 x 45 cm and 55 cm high, finished in oil. It arrives flat-packed and goes together in about ten minutes. Small enough to sit beside a chair or bed." What I'd fix in typical output is invented detail: "dovetail joints", "sustainably sourced oak" or "ships next day", none of which the row contained. Check the numbers against the row, because models sometimes swap width and height.
The platform matters for length. eBay titles run to 80 characters, and Amazon capped product titles at 75 characters from 27 July 2026 in every category except media, so write to the shorter limit if you sell on both. If your shop runs on Shopify, Sidekick, the assistant built into the admin, writes product descriptions from the generate-text icon on the product page. It's included with Shopify plans, and its help page says it presents changes for your review before applying them.
A second pass catches what a skim misses. Paste the finished description back and ask: "List every claim in this description and the cell in the row it came from. Mark any claim with no source." Anything marked is invented, and you delete it. Save the first prompt, with your house style, as a note you reuse, so the 200th description reads like the first.
As a sum: written by hand at six minutes a product, 120 products take twelve hours. At two minutes of checking each, the same 120 take four. Start with the ten products whose current text is worst, and keep the rest of the catalogue until you trust the checks.
2. A reorder list from 12 weeks of sales and your lead times
Reordering is arithmetic, though it is often done from memory and instinct. The method is simple: target stock is weekly sales multiplied by supplier lead time plus a week of safety, and the order is the shortfall rounded up to the supplier's case size. What takes time is gathering the figures for every line, and that is where an assistant helps. Here is a gift and homeware shop working from the last 12 weeks of till sales (illustrative figures).
| Item | Sold a week | In stock | Lead time (weeks) | Target stock | Shortfall | Case size | Order |
|---|---|---|---|---|---|---|---|
| Scented candle | 14 | 22 | 2 | 42 | 20 | 6 | 24 (4 cases) |
| Ceramic mug | 9 | 30 | 3 | 36 | 6 | 4 | 8 (2 cases) |
| A5 notebook | 11 | 18 | 1 | 22 | 4 | 10 | 10 (1 pack) |
| Tea towel | 5 | 40 | 2 | 15 | none (25 over) | 6 | 0 |
| Card box | 8 | 5 | 2 | 24 | 19 | 12 | 24 (2 boxes) |
| Tote bag | 3 | 12 | 4 | 15 | 3 | 5 | 5 (1 case) |
Target stock is sold a week multiplied by lead time plus one, so for candles it is 14 x (2 + 1) = 42. The tea towels show why the list earns its keep: 40 in stock against a target of 15 means no order this week and a note that they are overstocked. Give the assistant the export and the table's columns:
Here is 12 weeks of sales by product, current stock, supplier lead time in weeks and case size. For each product calculate: average sold per week (exclude returns), target stock = weekly sales x (lead time + 1), shortfall = target - stock, and an order quantity rounded up to whole cases. Show your formulas. Flag any product with more than 20 weeks of cover. Return a table.Three checks stop the usual errors. Recalculate three lines by hand, because models slip on arithmetic. Make sure returns weren't counted as sales and that a week of stockouts didn't drag the average down, since sales history understates demand whenever you ran out. And look for seasonality: twelve weeks that include a December rush will overstate January. For a shop reordering 60 lines a week at two minutes each by hand, the routine of exporting, prompting and checking ten rows takes around 40 minutes, which saves roughly 80.
The tools vary by platform. Shopify's Stocky app stopped working on 31 August 2026, and purchase orders now live in the Shopify admin, where you save a draft and mark it Ordered. Sidekick's help page says it can write and run its own code to forecast, compare or summarise store data, though in some stores it can't run code. Lightspeed's pricing page lists "Insights" (forecasting and order recommendations) as a feature row, so check which plan carries it. ChatGPT can analyse an uploaded CSV, though the free plan limits both uploads (three a day) and data analysis. A fuller walk-through is in how small shops can use AI to forecast stock and reorder.
Two habits keep the list honest. Run it on the same day each week, so the sales window is comparable, and keep last week's table beside this week's: a line whose order jumps from 10 to 40 deserves a question before you place it. For a new product with no history, don't let the assistant extrapolate from nothing. Enter a manual estimate from a similar item and mark it as a guess, then replace it after four weeks of real sales.
3. A slow-mover list with a markdown ladder
Slow stock ties up cash quietly. Take an online clothing store with 300 live variants where, say, 70 haven't sold in 12 weeks. A linen shirt in XL costs $18, sells for $60, and 14 are sitting in the stockroom, which is $252 of cash that isn't earning. The job is to list the slow lines and decide a sequence of markdowns, with a floor so that no step falls below cost plus a small margin.
| Step | Timing | Price off $60 | Price | Margin on $18 cost |
|---|---|---|---|---|
| Start | Weeks 1 to 3 | 20% | $48 | $30 |
| Second | Weeks 4 to 6 | 30% | $42 | $24 |
| Third | Weeks 7 to 9 | 40% | $36 | $18 |
| Last | Week 10 onward | 50% | $30 | $12 |
Ask the assistant to list the candidates and attach a step, with a rule that stops it going too deep:
From this export (SKU, size, colour, stock, units sold in the last 12 weeks, cost, price), list every variant with no sales in 12 weeks or more than 26 weeks of cover. For each, propose a starting markdown and a ladder, never below cost plus 5%. Group by product so I can bundle sizes. Show the cash tied up (stock x cost).The output worth acting on is the grouping. If XL, XXL and a dozen other sizes in the same shirt are slow, a bundle or a clear-out on that style beats nibbling at each variant. The warning is that "slow" can be a symptom of a stockout: if the XL was missing for five of those 12 weeks, the sales history reflects your gap, not the customer's lack of interest. Check the dates before you cut a price. Finding dead stock and slow movers with AI goes further into the thresholds, and Shopify's Sidekick or Lightspeed's reports can pull the same list without an export.
Decide where each markdown is announced before you apply it. A first step of 20% can go quietly onto the site, while the 40% step suits an email to past buyers of the style, who are the likeliest to want another. Write the ladder dates into the sheet so that the markdowns happen on schedule instead of whenever someone remembers. If a line is still unsold after the last step, the choices are a bundle, a donation or a bin, and cash recovered at cost is better than stock that depreciates on a rail.
4. Customer replies that quote your policy and stop at the edge
This is the job where an error is visible to customers, so I put it last in the order I'd adopt. Take a specialist food shop that answers the same questions daily: dispatch cut-off, return of perishables, delivery to a particular address and, with luck, nothing about ingredients. The method is a facts block that the assistant must quote and a handover list it must obey.
You draft replies for a food shop. Use only the facts below. If the answer isn't there, write: "Thanks for asking. I'll check this and come back to you today." Never answer questions about allergens or ingredients; always use the same line. Under 80 words, friendly, first name only.
FACTS
Orders placed before 14:00 on weekdays go out the same day.
Perishable items can't be returned unless faulty. Report faults within 48 hours with a photo.
Delivery is $5.50 under $60 and free over.
Open Tuesday to Saturday, 10:00 to 17:30.
Customer: Hi, I ordered a jar of anchovies and it arrived cracked. Can I send it back?A good draft apologises, asks for a photo within 48 hours as the facts say, and promises a refund or replacement once the photo arrives. Watch for the model promising a refund, which is a decision, or quoting a number the facts don't contain. A second message shows the edge. A customer asks, "Have you got the 500 ml olive oil in stock?" The facts block holds no stock levels, so the correct draft says someone will check and reply today. Without that rule, a model will happily say yes. Allergens are a hard line. A model answering from memory about nuts in a sauce can hurt someone, so the rule is that a person reads the label and replies.
Tools differ in how directly they sit in your workflow. You can paste messages into ChatGPT. Google says Gmail's "Help me write" can match the tone of your past emails on Business Starter, Standard and Plus plans and on Google AI plans, since May 2026. Shopify's help page says Shopify Inbox gives AI-generated instant-answer suggestions and suggested replies, which you accept or change. Inbox also has an opt-in Inbox agent that answers customers on its own from your catalogue, policies and knowledge base. Leave it switched off until it has passed your own test questions on returns and delivery and has shown that it hands allergen questions to a person instead of answering them. Until then, keep a human on the send button. As a sum, 15 messages a day at 2.5 minutes each is 37.5 minutes, and at one minute a draft it's 15, saving about 2.25 hours across a six-day week.
5. Promotions where the margin sum comes before the wording
AI writes persuasive promotion copy easily, which makes it easy to run a promotion that loses money. Do the sum first and give the assistant a floor. Take a gift shop item that costs $22 and sells for $40, a $18 margin per unit, 45%. The table shows how many extra units each discount must sell to leave total gross profit unchanged.
| Discount | New price | Margin per unit | Units needed to match old profit | Extra volume |
|---|---|---|---|---|
| 10% | $36 | $14 | 1.29 x | +29% |
| 15% | $34 | $12 | 1.50 x | +50% |
| 20% | $32 | $10 | 1.80 x | +80% |
| 25% | $30 | $8 | 2.25 x | +125% |
A 20% discount has to sell 80% more units just to stand still, so ask yourself whether your shop can realistically do that before offering it. That is why the prompt should ask for options that protect margin:
Item: scented candle, cost $22, price $40. Give me three promotions for a weekend email that keep gross margin at or above 35%. At least one must not be a price cut (for example, a free gift wrap that costs me $1.20). For each, give an email subject line under 50 characters and a two-sentence body. Show the margin per unit for each offer.The non-price option is often the best. Free gift wrap at $1.20 reduces the $18 margin by under 7%, where 15% off removes a third of it ($6). The assistant also proposes subject lines faster than you will, but you set the floor and check the maths. Rules about "was" prices vary by place, so keep a record of the price an item really sold at before any markdown. On cost, Mailchimp's free plan covers 250 contacts and 500 sends a month but no scheduling or automations. MailerLite's free plan allows 250 subscribers, 2,500 emails and three automations of up to five steps. Setting prices and promotions with AI covers the pricing side in more depth.
To judge whether a promotion worked, compare gross profit per trading day with the same weekdays over the previous four weeks, not revenue. Revenue can rise while profit falls. Note any stock the promotion pulled forward from the following week, because a weekend spike followed by a quiet Monday may be the same sales moved earlier. Keep the result in one row of a sheet per promotion and the next set of options gets better.
6. Product photos tidied without a studio
Photos do much of the selling online, and a phone picture taken against a plain sheet by a window can be turned into a clean, consistent product image in seconds. An online clothing store with 40 garments and five angles each has 200 images; at two minutes each by hand, that's nearly seven hours, against perhaps an hour of batch processing and checking.
Shoot consistently first, because no tool rescues a bad original. Use daylight from a window, a plain sheet pulled taut, the phone at chest height, and the same five angles for every garment: front, back, side, a fabric close-up and a label shot. Consistent inputs make batch processing predictable.
The before: a shirt on a hanger with a creased cloth behind it, uneven light, a cluttered edge. The after: the same shirt on a plain white or pale grey background, edges clean, a soft shadow, sized for the store. Tools: Photoroom's Pro plan was listed at $12.99 a month, or $7.50 a month billed yearly, with 1,000 exports a month, when I read the page on 29 October 2026. Shopify's file editor can replace an image background with a colour, and Shopify's help pages describe Tinker, a free mobile app with more than 100 AI tools for product photography and marketing videos.
The rule is that the image must show the product you actually ship. Don't let a tool "fix" a sleeve length, a hem, a texture or, above all, a colour. A shirt that looks navy on screen and arrives midnight blue becomes a return. Compare the processed image to the garment in daylight, and ask someone else to compare them too. If you list products in Google Merchant Center, Google's image requirements say images created using generative AI must carry metadata indicating that, such as the IPTC DigitalSourceType tag, and that you shouldn't strip that metadata. The wording covers images created using generative AI. Whether a plain background clean-up counts depends on the tool, so check what metadata your tool writes and don't strip it. Google Merchant Center image requirements for AI-generated product photos sets out the details. Generating an imaginary model wearing your garment is a different and riskier job.
7. Delivery notes and invoices checked against the order
Receiving stock is where small shops lose money without noticing: short deliveries invoiced in full, wrong lines, best-before dates too close. A specialist food shop with six deliveries a week might spend about 12 minutes matching each against the purchase order (an assumption to replace with your own). The AI version is to photograph or scan the delivery note, paste it with the order into ChatGPT or Claude, and ask for a three-way table of ordered, delivered and invoiced.
| Line | Ordered | Delivered | Invoiced | Action |
|---|---|---|---|---|
| Olive oil, 500 ml | 24 | 24 | 24 | None |
| Anchovies, jar | 24 | 18 | 24 | Ask for a credit for 6 jars |
| Pasta, 500 g | 48 | 48 | 48 | None |
| Capers, jar | 12 | 12 | 10 | Invoice is low; tell supplier |
The shortage is worth money: at $4.10 a jar, six missing jars is $24.60 you'd otherwise pay for goods you never got. Ask the assistant to draft the supplier email with the line, the quantities and the amount, and read it before sending. The line invoiced low matters too: a supplier who spots the error later may bill the difference, so tell them now. If the notes show best-before dates, add a rule that flags any line with less than 90 days left, but only where the dates are legible; a model can't read what the photo doesn't show.
Whether ChatGPT can read an image-only scanned PDF isn't spelled out in OpenAI's file-upload FAQ, so test yours, and type the six figures in if it can't. ChatGPT limits uploads on the free plan to three a day. Xero's built-in Smart Document Capture, launched in July 2026, reads receipts and bills for bookkeeping and is listed as free on all plans, which covers the accounting side. For the stock-update side, can AI read delivery notes and update stock automatically explains how far to take it.
Once a month, ask the assistant to total the mismatches by supplier from your log: short deliveries, wrong lines, late deliveries and the value of credits recovered. A supplier who is wrong on one delivery in five is a conversation, and the log is the evidence. Keep the log as a plain sheet with the date, supplier, line, quantity and amount.
8. Google profile posts and review replies in ten minutes a week
For a bookshop or any shop that depends on local searches, the Google Business Profile is the shop window, and it needs a regular post and prompt replies. The routine is one sitting a week. Gather three facts: new arrivals, an event, a change in hours. Ask for three posts, then schedule them. Google Business Profile supports scheduling natively through "Schedule this post" and "Repeats", so you don't need another tool.
Write three Google Business Profile posts for a small independent bookshop. Facts: new in this week are five staff-pick novels and a children's picture book; author evening on [day] at 18:30, free entry, 40 seats; closing at 16:00 on [day]. Each post under 120 words, no hashtags, one clear action. Do not invent prices, authors or dates.Two details save embarrassment. Events need a start and end date, and Google archives posts older than six months unless they are dated. On reviews, a typical case is a three-star comment: "Lovely shop but couldn't find the book I wanted." A draft might thank the reviewer, say you can order almost any title within a few days, and invite them back. The fixes are the usual ones: delete defensive wording, don't mention the reviewer's personal details, and never offer a reward in return for a changed rating. Google's review rules forbid incentives and selectively asking only happy customers for reviews. Replies can take around ten minutes to appear and occasionally much longer.
Google has reportedly been testing AI-suggested review replies with some accounts, though its help pages don't describe the feature yet; if it appears in your profile, treat any suggestion as a draft. To see whether the routine helps, look at the profile's own performance figures each month, such as calls and direction requests, and compare with the month before. Google's report doesn't offer a weekly view, and search terms refresh monthly, so judge over a month, not a week. For more on profiles, local SEO for a retail shop: profile, stock posts and reviews with AI has a longer routine.
Which job to copy first, and what all eight cost
Order the jobs by what it costs when they go wrong, not by how impressive they sound. Text you read before publishing comes first, jobs with money at stake come next, and anything a customer sees without you comes last.
| Job | Risk if wrong | Main check | New cost a month (list prices) |
|---|---|---|---|
| 1. Product text | Low | Facts match the row | $0 (free plan) or ChatGPT Plus at about $20 |
| 8. Google posts and replies | Low | Dates, no incentives | $0 |
| 7. Delivery notes | Low to medium | Figures match the paper | $0 on existing tools |
| 2. Reorder list | Medium | Three lines by hand | $0 |
| 3. Slow movers | Medium | Stockout dates | $0 |
| 5. Promotions | Medium | Margin floor | $0 on free email plans |
| 6. Photos | Medium | Real colour, real item | $7.50 (Photoroom Pro billed yearly) or $0 with Shopify's editor |
| 4. Customer replies | High | Human on send; allergens always manual | $0 on existing tools |
All eight together need at most about $27.50 a month in new spend, ChatGPT Plus at roughly $20 and Photoroom Pro at $7.50 billed yearly, and a shop using free tiers can do most of them for nothing. Confirm every list price before you pay.
A four-week order that respects the risk column: product text and Google posts in week one, delivery notes in week two, reorder list and slow movers in week three, then promotions and photos, and last the customer replies once you trust your own checks. Whenever a job has a check, write the check down as a line in the prompt, not a thing you promise to remember.
What the big-retailer examples don't shrink to fit one shop
Three popular examples don't scale down. Dynamic pricing engines react to competitors' prices every hour, which one shop with a printed shelf label can't use. Demand forecasting that blends weather, local events and thousands of products needs years of data per line, and a single shop often has little more than a season. Shelf cameras need hardware, installation and a privacy review. If a vendor pitches any of these to a one-shop owner, ask what they'd do on your 400 products and your three staff, and what they'd cost compared with the eight jobs above. For most shops the honest answer is that the eight jobs earn more, sooner, for less.
Questions small shops ask before copying these jobs
Which of the eight jobs should a small shop copy first?
Start with the job that costs least when it goes wrong: product text or Google profile posts, because you read every word before it goes live. Do reorder lists next, checking three lines by hand. Keep customer replies, price changes and anything touching allergens or safety until your checks are working.
Do I need Shopify or a modern till to use these jobs?
No. Most of them run on a spreadsheet export and a chat assistant such as ChatGPT, Claude or Gemini. Shopify's Sidekick, Shopify Inbox and Lightspeed's forecasting rows save you the export step, but a shop with a plain till and a CSV can do the same work.
Is it safe to paste a sales export into ChatGPT?
Export by product, not by customer, and strip names, emails and addresses first. On individual plans, check the privacy switch that stops chats being used to improve the models. ChatGPT Business doesn't train on business content by default but needs at least two seats, which a one-person shop may not want.
Can AI reorder stock on its own?
Not safely. Tools can suggest quantities or draft purchase orders, as Shopify's Sidekick and Lightspeed's forecasting can, but a person should approve every order. The model can't see supplier stockouts, price rises or the promotion you plan next month, and sales history understates demand whenever an item was out of stock.
Further reads
- Taking a Local Shop Online With AI: Product Listings in a Weekend — A weekend plan for taking a local shop online with AI listings.
- How Boutiques Can Answer Customer DMs Within an Hour Using AI — Reply to social messages within an hour using AI drafts.
- What AI Costs a Small Retail Shop in Its First Year — What AI really costs a small shop in its first year.
- AI Agents in Retail: What an Independent Shop Can Actually Run — Which AI agents an independent shop can actually run.
- AI Chatbot for Retail Stores: Can It Answer Stock Questions? — Whether a shop chatbot can answer stock and opening-hours questions.
- How Will AI Affect Retail? The Jobs a Small Shop Hands Over First — The wider view of which shop jobs AI takes on first.
- Retail KPI Dashboard in Excel: A Weekly Sheet From Till Data With AI — Turn weekly till data into a simple Excel dashboard.
- AI Customer Service Kiosk for a Shop Counter: Worth the Cost? — A payback test for a counter kiosk, what it can and cannot answer, and the cheaper QR code chatbot most small shops should try first.
- Retail AI Consultant: Hire-or-Not Checklist for a Small Shop — A hire-or-not test for small shops: signs you need help, what to put in scope, fixed fee versus day rate, and what to ask a retail AI consultant.
- AI for Retail Store Design: Layout and Display Ideas — Plan a shop re-layout with AI, safely: measure, get three options, tape-test the best, check exits and access, then measure whether sales moved.
- Customer Loyalty Programmes for Retail Stores: Can AI Design One? — A small shop's loyalty programme drafted with AI: points, stamps and tiers costed line by line, plus the five corrections the draft needed.
- How AI Can Help in Sales: A Gift Shop's Bundle and Follow-Up Plan — A gift shop's AI sales plan built on its own till data: three priced bundles, gift-note wording and four follow-up messages for one holiday month.
- Retail Staff Scheduling: Build a Rota From Hourly Till Sales With AI — A single-shop method: export timestamped sales, average them by weekday hour, set heads per hour, draft the rota with AI and check it.
- AI for Convenience Stores: A Weekly Routine for Reorders and Rotas — One weekly run-sheet for a convenience store: a reorder list from the till export, shelf prices against two rivals and a rota draft, each step approved by you.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Shopify help pages on Sidekick, AI-powered tools, Shopify Inbox and purchase orders; Google Merchant Center image requirements; Photoroom, Lightspeed Retail and Square pricing pages read on 29 October 2026; Mailchimp, MailerLite, Xero and Google Business Profile details from vendor pages as checked in October 2026.