AI helps a small supermarket most with three weekly jobs: checking shelf prices against supplier costs, turning till sales into a reorder list, and drafting offer posters. Each takes under an hour a week with a spreadsheet export and a chat assistant, and each result needs a manager's check before it reaches a shelf, a supplier order or the window.
The chains behind most of the advice online run forecasting software fed by years of transaction data from every store. One store cannot copy that and does not need to. The same three questions, asked every week of your own numbers, catch the usual leaks: a cost that rose unnoticed, a shelf that ran empty on Saturday, an offer nobody could read from the door. The assistant sorts and drafts, the spreadsheet does the sums, and a person signs off.
What a chain's system does, and the one-store version
Big grocers split the work across separate systems. One forecasts demand for each item in each store, one holds the pricing rules and pushes changes to every till and shelf edge, and one assembles the weekly promotion plan. A single store has one site, one buyer and, usually, one person who knows why a certain brand of rice sells out on Thursdays. In your first months with AI that is an advantage, because you can run the whole loop on a short list you check by hand instead of trusting a model with every line.
| The chain's system | The one-store version | Weekly time once set up |
|---|---|---|
| Demand forecast for each item and store | Eight weeks of till sales by item in a sheet, a reorder rule, and an assistant to build and audit the formulas | 30-40 minutes |
| Central pricing rules pushed to every shelf | A margin target per category, a short list of lines shoppers watch, and a weekly check of cost against shelf price | 30-45 minutes |
| Promotion planning and printed material | An offers register built from overstock and supplier deals, wording drafted by AI, one person checking the printout | 40-60 minutes |
| Markdowns on short-dated stock | A date list from the morning walk and a fixed reduction ladder, covered in the markdown tutorial linked below | 10-15 minutes a day |
Those times are my estimates for a store with a few thousand lines, not measurements. Time your own first run.
What you cannot copy is scale. A chain can try a price in some stores and leave the rest alone; you have one shelf, so change a handful of prices at a time and write each one down. The tutorial on testing price changes covers how to read the result a fortnight later. None of the three jobs below needs customer data. Item-level sales, costs and shelf prices are business figures, so keep loyalty-card exports out of the assistant altogether.
Job one: a weekly shelf-price check against supplier costs
Start with 40 lines: your fastest sellers plus any item a supplier sent a price-rise notice for. For each, put five facts in a sheet: cost last month, cost now, shelf price, weekly units and the category's target margin. Margin here means shelf price minus cost, divided by shelf price, both before tax. Add a column marked yes for known-value items, the dozen or so lines shoppers remember (milk, bread, eggs and similar). Those move last and least.
Say the owner's targets are 15% on known-value items, 24% on ambient grocery, 28% on frozen and 30% on household. They are the owner's own choices, not industry figures; set yours from your own accounts. Costs come from the supplier price list or the latest invoices, and the tutorial on supplier invoices and shelf prices shows how to pull them into the sheet without retyping.
Supplier notices usually arrive as a PDF or an email with the new prices buried in a long table. Paste the text into the assistant and ask it to return only the lines you stock, as old cost, new cost and percentage change, so the sheet stays current without retyping. A line from an illustrative notice might read "Cooking oil 1 L x 12, was 25.80 a case, now 31.20 a case". The assistant should hand back cooking oil 1 L, $2.15 to $2.60 a unit, up 20.9%. Check the pack count every time. Treat that case as 10 instead of 12 and the cost jumps to $3.12, so the margin appears as 2.2% instead of 18.5%. A margin that looks absurd usually means a wrong pack count, not a wrong shelf price.
Eight lines from an illustrative Monday
| Line | Cost last month | Cost now | Shelf price | Margin now | Target | Verdict |
|---|---|---|---|---|---|---|
| Whole milk 2 L (known value) | $1.55 | $1.71 | $1.99 | 14.1% | 15% | Below target, cost up 10.3% |
| Sliced white bread 800 g (known value) | $0.88 | $0.94 | $1.35 | 30.4% | 15% | Fine |
| Eggs, dozen (known value) | $3.10 | $3.42 | $4.29 | 20.3% | 15% | Fine, but cost up 10.3% |
| Cooking oil 1 L | $2.15 | $2.60 | $3.19 | 18.5% | 24% | Below target, cost up 20.9% |
| Dishwasher tablets, 40 | $4.40 | $4.40 | $5.99 | 26.5% | 30% | Below target, cost unchanged |
| Frozen peas 1 kg | $1.38 | $1.38 | $2.19 | 37.0% | 28% | Fine |
| Tinned tomatoes 400 g | $0.41 | $0.47 | $0.69 | 31.9% | 24% | Fine, but cost up 14.6% |
| Instant coffee 200 g | $3.95 | $4.45 | $5.49 | 18.9% | 24% | Below target, cost up 12.7% |
The profit sums further down use weekly units of 160 for milk, 45 for cooking oil, 18 for dishwasher tablets and 30 for instant coffee, so you can check them. Eight lines is a teaching size; the real sheet has 40.
The prompt that does the margin sums
I run a single-store supermarket. The attached CSV has one row per product:
name, category, known_value (yes/no), cost_last_month, cost_now,
shelf_price, weekly_units. All prices exclude tax.
Target gross margin: known-value items 15%, ambient 24%, frozen 28%,
household 30%.
1. Add a column for current margin, calculated as
(shelf_price - cost_now) / shelf_price. Show me the formula.
2. List every row whose margin is below its target.
3. For each of those rows, suggest a new shelf price ending in 5 or 9
that reaches the target, and work out the change in weekly gross
profit if weekly_units stay the same.
4. For known_value = yes rows, only suggest a change if cost rose by
more than 5%, and mark it "check a rival's price first".
5. Separately list rows where cost rose by more than 10% even though
margin is still above target.
Do not invent any figure that is not in the file. If a row has missing
data, list it at the end instead of guessing.Asking for the formula matters. An assistant can slip on arithmetic across a long list, and a visible formula lets you recalculate three rows yourself before believing the rest.
What the assistant sent back, and what I would change
| Line | Margin now | Suggested price | Weekly profit change if units hold |
|---|---|---|---|
| Cooking oil 1 L | 18.5% | $3.45 | +$11.70 |
| Instant coffee 200 g | 18.9% | $5.89 | +$12.00 |
| Dishwasher tablets, 40 | 26.5% | $6.29 | +$5.40 |
| Whole milk 2 L (known value) | 14.1% | $2.05 | +$9.60, check a rival's price first |
Its watch list named eggs (cost up 10.3%) and tinned tomatoes (up 14.6%). That output is illustrative, but the shape is realistic, and four judgements are still yours.
- Milk. The assistant applied the rule correctly, yet the rule only says the cost rose. Moving the line shoppers compare most is a decision about your reputation, so I would hold milk this week and let the other lines carry the correction.
- Dishwasher tablets. The cost did not change; the line was simply always under target. Whether household goods must hit 30% is a policy question, so decide it once instead of letting the sheet nag you weekly.
- Volumes. The sums assume shoppers buy the same quantity at the new price. Rerun them with each repriced line selling 10% fewer units. For oil, coffee and tablets that turns the gain from $29.10 a week into about $17.50, still positive, which is a better thing to know than the optimistic figure.
- Price endings. It chose endings of 45, 89, 29 and 05 cents because the prompt asked for 5 or 9. If your shelf uses 49 and 99, say so in the prompt.
Change the till and the shelf edge on the same morning and print the list. A shelf that disagrees with the till is the complaint that costs most, and it is where automatic repricing would hurt you first. The tutorial on short-date markdowns handles the separate question of reductions on food nearing its date.
Job two: a reorder list from eight weeks of till sales
Export units sold by item for the last eight weeks, one column per week. Add stock on hand (from a count or the system), the supplier's delivery days and order cut-off, the case size and the shelf life. The rule is plain enough to write on the back of a delivery note: order daily sales times (days until the next delivery plus days between deliveries), add a safety stock, subtract what is on the shelf and what is already on order, then round up to whole cases.
Take natural yoghurt, 500 g, sold in cases of 12. It sells 42 a week, so 6 a day. The next delivery lands in 2 days, then every 3 days after that, and the owner keeps one day's sales (6) as safety stock. Needed: 6 × (2 + 3) + 6 = 36. There are 14 on the shelf and in the chiller and none on order, so 22 more are needed, which rounds up to 2 cases (24 units). The 14-day shelf life leaves plenty of room. A long-life line with the same sales would carry five days of safety stock, and a fresh line one day or less; those are choices you make per category.
A dry-goods line shows the rounding judgement. Take 1 kg rice, in cases of 10, selling 30 a week (about 4.3 a day). The weekly dry-goods delivery arrives in 4 days and then every 7, and the owner holds 5 days of safety stock. Needed: 4.29 × (4 + 7) + 4.29 × 5 comes to about 68.6 units. With 38 on hand and nothing on order, that leaves 30.6 to order, which is just over 3 cases, so the sheet says 4 cases (40 units). With a long date, 4 is a comfortable answer. If the same maths landed on a line with a three-week date, I would order 3 cases and accept a small risk of running short, because spoiled stock costs more than a late reorder. The sheet gives the number; you apply the rounding that suits the date.
Where the assistant earns its place is in building and auditing the sheet, not in guessing the order. Give it your columns and ask for formulas, then ask it to explain each formula in one sentence and to list three ways the sheet could mislead you. For formula help in general, the Google Sheets prompts guide is a good companion.
I run a single-store supermarket. Build a reorder sheet from these
columns: item, weekly units for weeks 1-8, stock on hand, units on
order, case size, delivery in (days), days between deliveries,
safety stock (days of sales).
Calculate average daily sales from weeks 1-8, ignoring any week I mark
as "stock-out". Then calculate the order quantity in units and in
whole cases. Flag any item whose weekly sales vary by more than 40%
from its average. Write the formulas so they fill down, explain each in
one sentence, and list three ways this sheet could mislead me.In a run like this, the line that trips people up is the zero week. Suppose a chilled dessert shows no sales in week 5 because the supplier missed a delivery. If the sheet averages that week in, daily demand comes out about an eighth too low (seven weeks of real demand averaged over eight), and the line keeps running out. The "stock-out" marker in the prompt exists for exactly this. The tutorial on forecasting mistakes lists the other common traps, such as treating a promotion week as normal demand.
To check it worked, count the empty facings on your top 50 lines at the same time each Saturday morning for four weeks, and keep a tally of what you throw out. If empties fall and the waste tally does not rise, the rule is doing its job. If both rise, your safety stock is wrong in opposite directions for different lines, so split the list by shelf life and set each group separately.
Job three: offer posters and tickets that match the till
Choose offers from the stock sheet rather than from whatever feels festive: lines you are overstocked on, lines with a supplier deal, and lines near their date. Put each offer in an offers register with the item, the barcode, the till price and the end date. Only then ask the assistant for wording, because it will happily write a persuasive poster for an offer that loses you money.
Check the offer before the poster. Take pasta sauce that normally sells at $1.79 and costs $1.12, a 37.4% margin. A "3 for $5" offer saves the shopper 37 cents against three singles ($5.37) and cuts your margin to 32.8%. Per unit you earn $0.55 instead of $0.67, so you need roughly 23% more units sold to make the same profit. If the shelf usually sells 40 a week, that means about 49 in the offer week. If the offer cannot plausibly lift sales by that much, it is a giveaway, however good the poster looks.
A filled-in row of the register shows how little it needs. For the sauce: item "Pasta sauce 400 g, any flavour"; till price $1.79; offer "3 for $5"; till promotion code entered by the person programming the till; runs Monday to Sunday; poster wording approved by the manager. Shelf-edge tickets add one more calculation. Where a price per kilo or litre must appear on the ticket, let the assistant work it out from the sheet (instant coffee at $5.49 for 200 g is $27.45 a kilo) and have a person spot-check five tickets, since a unit price built on the wrong pack size is worse than none.
Before and after: the same offer as a poster
The first draft the owner wrote on the back of a till roll, and the version after the assistant tightened it against the register:
Before: SUPER SAVER!!! Pasta sauce 3 for $5, best prices in town, hurry
After: Pasta sauce, 3 for $5 (usually $1.79 each). Any flavour, 400 g jars. Until Sunday, while stocks last.
The "after" version adds the usual price, the pack size and an end date, and drops "best prices in town", a comparison you would have to prove. Any "usually" or "was" price should be a price you actually charged for a reasonable period; rules on reference prices differ by place, so ask a qualified adviser if you are unsure. Canva's Free plan lists a monthly allowance of up to 20 Standard or Premium AI uses, so check how far that stretches before you plan a weekly run. The full design routine, templates and print sizes are in the tutorial on supermarket posters in Canva.
Build a Monday check into the routine. The person who programmes the till signs the offers register, and a second person walks the first five aisles before opening with the register in hand, comparing each poster and shelf ticket with it. It takes ten minutes and catches the case where the poster says 2 for $3 and the till still holds last week's 2 for $3.50.
Four ways this routine goes wrong in a supermarket
Stock the assistant cannot know. It may write "plenty in stock" or "delivery on Friday" because that is what such sentences usually say. Strip any statement about quantities or dates from drafts, and let a person add them.
Allergen and ingredient claims. In one illustrative case the assistant labelled an oat and honey bar "nut-free" because nothing in the product name suggested nuts, while the pack said "may contain nuts". Ingredient and allergen statements come from the pack or the supplier's specification sheet and nowhere else. Keep them off posters unless a person has copied them from the label.
Images that do not match the pack. If you generate pictures of products, shoppers will look for the one on the poster. Use the supplier's pack shots or your own photographs, and keep generated images for backgrounds and borders.
A sheet nobody updates. The commonest failure is a routine that quietly stops in week three because nobody owns it. Name one person for each job, put a 30-minute slot in the diary on the same morning every week, and let the sheet show a "last updated" date at the top so a stale one is obvious to everyone who opens it.
A four-week trial around a normal trading week
| Week | Job | What you produce | The decision at the end |
|---|---|---|---|
| 1 | Price check on 40 lines | Margin column, three to five proposed changes | Which changes go live on Monday |
| 2 | Reorder sheet for 25 chilled and fresh lines | A sheet that returns an order in whole cases | Does it match your gut on 20 of 25 lines? |
| 3 | Offers register and posters | Four offers, checked against costs and the till | Which offer sums were wrong before the poster? |
| 4 | Review | Minutes spent, empties counted, corrections logged | Keep, change or drop each job |
The free versions of ChatGPT and Claude are enough to try this. ChatGPT Free allows three file uploads a day, which suits a weekly routine; paid ChatGPT plans allow up to 80 files every three hours, and ChatGPT reads CSV and Excel files up to roughly 50 MB, so a store's weekly export fits easily. ChatGPT Plus and Claude Pro are each listed at $20 a month if you outgrow the free tiers. Treat the first month as an experiment with a stopwatch, a tally sheet and a log of every correction you had to make. A job that needs fewer corrections each week is earning its place; one that does not should go back to a person.
What a supermarket owner asks before trying this
Do I need software that connects to my tills?
No. Every job here starts from a spreadsheet export: item sales by week, stock on hand and a cost list. Most till and back-office systems can export a CSV file, so confirm yours can before anything else. If the export is awkward, solve that first, because every later job depends on it. Live connections between systems can wait until the routine has earned its keep.
Can AI change my shelf prices automatically?
It can, but I would not start there. A weekly review where the assistant proposes and a manager approves leaves a record of every change and the reason for it. Automatic repricing also risks a shelf label and a till disagreeing, which produces the complaints that cost you most. Keep a person between the suggestion and the shelf.
Is it safe to put supplier costs into a chat assistant?
Supplier cost lists are commercial information rather than personal data, so the main question is the tool's data use. Business plans such as ChatGPT Business and Claude Team do not train on your content by default; personal plans have a switch in the privacy settings. Check supplier agreements for confidentiality clauses, and keep loyalty-card data out entirely.
How long does the weekly routine take once it is set up?
In the illustrative store above, about two hours in total: 30 to 45 minutes on prices, 30 to 40 on the reorder sheet and 40 to 60 on offers. Those are my estimates, not measurements. Time your first two weeks with a stopwatch, and drop any job that costs more minutes than it saves.
Further reads
- How AI Helps Marketing in a Small Supermarket: 5 Jobs and the Limits — The marketing jobs a small supermarket can hand over, and where they stop.
- How Small Shops Can Use AI to Forecast Stock and Reorder — A wider look at forecasting and reordering for any small shop.
- AI for Convenience Stores: A Weekly Routine for Reorders and Rotas — A weekly routine for a smaller format with rotas included.
- Retail KPI Dashboard in Excel: A Weekly Sheet From Till Data With AI — Turn the same till export into a one-page weekly dashboard.
- How to Find Dead Stock and Slow Movers With AI — Find the lines that tie up cash before you reorder them.
- How to Use AI to Set Reorder Points and Prevent Stockouts — Set reorder points line by line once the weekly sheet runs.
- Can AI Help a Small Shop Set Prices and Promotions? — How a small shop sets prices and plans promotions with AI.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's help pages on file uploads, Canva's pricing page, ChatGPT and Claude plan pages, checked 29 October 2026. Prices, margins and volumes in the examples are illustrative.