AI for Convenience Stores: A Weekly Routine for Reorders and Rotas

Each week an AI assistant can turn the till's sales export into a reorder list, compare 20 of your shelf prices with two rivals, and draft the next rota, with you approving each step. Plan on about 90 minutes once the files are set up. Nothing reaches a supplier, a shelf label or a staff member without your sign-off.

The routine is deliberately dull: the same three exports, the same three prompts and the same three checks. Its value comes from running it every week and noticing when an answer looks wrong. Do it once the weekend's takings are in and before your wholesaler's order cut-off.

Follow me on Instagram@sagnikteaches

The weekly run-sheet, with times

Here is the whole routine on one page. The times are for you, after the set-up, and they leave out the walk round two rival shops, which a staff member can do in a quiet hour.

Connect on LinkedInSagnik Bhattacharya

Every step ends with you. The AI drafts and calculates; you approve, change or discard. That one rule is what makes this safe enough for a shop where one mistake can mean an empty chiller on a Saturday or a rota that breaks a rule.

Subscribe on YouTube@sagnikteaches

Step 1: from till export to a reorder list you can check

Start with the report your till calls "sales by item", "product mix" or something similar: one row per product, with units sold, for the last seven days and the last 28. Restrict it to your top 60 lines by units. Those lines drive the week's ordering, and slow lines can be ordered by eye each month. Add the units on the shelf and in the stockroom for the same lines, which a 10-minute count gives you, and a small reference sheet with each line's supplier, case size, case price and delivery days.

That reference sheet is a one-off job of about an hour for 60 lines, and it removes most of the guessing from every later run. Three rows show how little it holds.

The rule the assistant applies is plain arithmetic, and you should be able to say it aloud. A weekly wholesaler order placed on Monday for Wednesday delivery has to cover two days until the delivery lands, seven days until the next one and a day of safety margin, so the target stock is 10 days of sales. The suggested order is that target minus what is on hand, rounded up to whole cases, and skipped when the need is less than a quarter of a case.

You are helping a convenience store owner prepare the weekly wholesaler order. Use only the three files I upload: sales by item (7 days and 28 days), stock on hand, and the case-size sheet.
Rules:
1. Weekly demand = the 28-day average per week. If last week is more than 25% above or below that average, flag the line and show both figures.
2. Target stock = weekly demand / 7 x 10 days.
3. Suggested order = target - stock on hand, rounded up to whole cases. If the need is less than a quarter of a case, write SKIP.
4. Show your working for every line: weekly demand, target, need, cases.
5. Do not order any line that is not in the case-size sheet. Do not invent prices.
Output a table, then a list of flags.

Here is an illustrative extract for eight lines of a 60-line order.

The cola flag is the approval step doing its job. On the 28-day average the line needs five cases. Last week sold 150, which is 36% above that average, and the forecast says another hot week, so you overrule the average and use last week's figure: a target of 214, a need of 166 and seven cases. The AI could not know about the weather; you did. At the illustrative case prices ($7.20 for water, $11.50 for cola, $14.40 for crisps and for cat food, $23.40 for chocolate, $18.00 for batteries and $19.20 for energy drinks), those eight lines come to $369.70 with seven cola cases.

Three checks before you send anything. Redo three lines on a calculator, because assistants occasionally slip on a sum. Compare the whole order's total with your last four orders, and if it is more than about 15% away, find out why. Read every flag. Fresh and chilled lines sit outside this routine: they need a day or two of cover, not ten, and the short-dated markdowns tutorial covers what to do with what does not sell. If you want the reorder-point arithmetic explained line by line, the stock forecasting tutorial and the reorder points tutorial go further than there is room for here.

Five ways a till export misleads the AI

The arithmetic is easy. The data is where it goes wrong, and the assistant cannot see any of these problems unless you tell it.

1. Department keys. A sale rung up on a generic "grocery" key records money but no product, so it never reaches the reorder list. Check one day's till roll for generic keys and scan items instead.
2. Promotion weeks. A multibuy week inflates last week's sales. Mark promotion weeks in a column so the average can absorb or exclude them.
3. Stock-outs hide demand. A line that was empty for three days looks like a slow seller. Note which lines ran out and tell the assistant how many days each was missing.
4. Deliveries not booked in. If a delivery was never entered, the till's stock figure is wrong. For the fast lines, trust the shelf count over the system.
5. Split rows. A new barcode or a renamed product splits one line's sales across two rows. Merge them by hand once, and the problem stays fixed.

The exceptions note: what the AI cannot see

The files hold the past, and the week ahead lives in your head. Keep a short note, updated on Fridays, and paste it into every prompt. It should hold anything that changes demand, availability or staffing: weather, school holidays, local events, promotions, supplier changes and leave.

EXCEPTIONS NOTE, week of the 12th
- Weather: hot weekend forecast (Sat, Sun). Treat cola and water as +30% (my guess).
- Supplier: crisps multipack promotion starts Wednesday; last week's crisps sales are not a base.
- Delivery: wholesaler moves from Wednesday to Thursday this week, so use 11 days of cover, not 10, and move the delivery-hour cover to Thursday.
- Staff: B on holiday all of next week (25 contracted hours to cover). F has exams on Tuesday and Wednesday afternoons.
- Local: roadworks close the main road outside the store on Thursday and Friday.

Without that note, the hot weekend looks like a trend and the rota looks fine. With it, the assistant can recalculate the target for a delivery that arrives a day late (three days to cover before it lands, plus seven and a day's margin makes 11 days), leave last week's promotion sales out of the base, and show you that B's holiday leaves 25 contracted hours to cover, with F's exam afternoons as a second constraint. The note takes five minutes and prevents most of the surprises that make owners distrust a forecast.

Step 2: twenty shelf prices against two rivals

Choose 20 lines: the ten you sell the most of and ten that shoppers remember the price of, such as milk, bread, a can of cola and a bottle of water. Pick two rivals that actually take your customers; for most shops that is the nearest supermarket-format store and the nearest other independent. A staff member photographs each shelf label, including the unit price printed on it, or types the prices straight into a sheet. Typing is slower and more accurate. Assistants can read labels from photos but now and then misread a digit or a pack size, so check three against the photos.

Where a rival publishes its shelf prices online, you can copy them from the public page. Check the site's terms before setting any tool to collect prices automatically, and see the price tracking tutorial for monitors that stay within bounds.

Compare my shelf prices with two rivals. I will paste: line, my price, my pack size, my cost price, rival A price and pack size, rival B price and pack size, date seen.
1. Convert every price to a unit price (per item, per 100 g, per 100 bags, whichever fits) before comparing.
2. Show my gap to the cheaper rival as a percentage.
3. Flag any line more than 10% above the cheaper rival, and any line where pack sizes differ.
4. For flagged lines, show my margin at my price and at the cheaper rival's price. Never recommend a price below my cost.
5. Do not suggest agreeing or coordinating prices with anyone.

The tea bags show why unit prices matter. Per 100 bags we charge $3.50, rival A $2.75 and rival B $3.74, so we are 27% above A's larger pack and 6% below B, and the headline prices said nothing of the kind. On the rest of the table I would change a price only when three things are true: the line is more than 10% above the cheaper rival, it is one shoppers compare, and the margin after the change stays above your floor. That leaves bread, water and chocolate to review, and the assistant's job ends at showing the margin at each price. The assistant flagged batteries too, correctly by its rule, but they are a distress purchase that shoppers rarely compare, so keeping the premium may be sensible.

One slip is worth designing for, because it hides a problem instead of creating one. In another week a blurred label turns rival A's $1.29 loaf into $1.79 in the assistant's reading. With rival B at $1.70, your $1.85 loaf then looks 8.8% above the cheaper rival and the verdict is "leave", when at the true price it is 43% above. That is why you check three prices in person, and why at least one of the three should be a line the assistant marked "leave".

Two housekeeping rules. Change the shelf label and the till price on the same day, because a label that disagrees with the till is the quickest route to an argument at the counter and, in many places, a pricing complaint. And keep to observing: noting prices on public display is ordinary shop practice, but agreeing prices with a rival, or even sharing your plans, is something competition rules in many places forbid. Ask a qualified adviser if you are unsure where the line is.

Step 3: a rota draft that lists its own gaps

A rota is a set of constraints, and checking constraints is what an assistant does well. Give it the opening hours, the minimum cover you insist on, the extra cover that busy periods need (the hourly sales shape from the till tells you when), who can work when, and each person's contracted hours. Use first names only.

The extra-cover hours come from the till's hourly sales for the last 28 days, averaged by weekday hour. Here is an illustrative weekday, with one simple rule: two people on whenever average sales pass about $150 an hour.

Lunchtime at $140 an hour sits just under the line, so one person covers it. If the average creeps past $150, the rule adds a second person and the assistant should tell you. Now the totals for the whole week.

Take a store that opens from 6.30 am to 9.30 pm every day, which is 105 hours of single cover a week. Add one extra person on weekday mornings from 7 to 9 (10 hours), on weekday afternoons from 4 to 7 (15 hours), on Friday and Saturday evenings from 7 to 9.30 (5 hours) and for the Wednesday wholesaler delivery hour (1 hour). That makes 136 hours. Six part-timers on contracts of 30, 25, 20, 20, 16 and 12 hours give 123, and the owner's 13 hours make 136, so on paper the rota balances exactly. In practice availability never lines up with demand, which is why the useful output is the list of gaps.

Draft next week's rota for a convenience store (Monday to Sunday).
Opening hours: 06:30-21:30 every day.
Minimum cover: one person at all times; two people Mon-Fri 07:00-09:00 and 16:00-19:00, Fri and Sat 19:00-21:30, and Wednesday 14:00-15:00 (wholesaler delivery).
My rules: no one works more than 5 days in a row; at least 11 hours between shifts; no shift over 9 hours; staff F is under 18, so flag any shift that might break limits for young workers and I will check the rules.
Staff (first names), contracted hours, availability: [paste]
Output: a table by day and shift, then a numbered list of every gap or rule conflict with two options for each. Do not fill a gap by breaking a rule.

GAPS IN THE DRAFT (illustrative)
1. Sunday 17:00-21:30: nobody assigned. Only the owner is free, and the owner closes on Saturday.
   Options: (a) owner covers 17:00-21:30 (4.5 h); (b) D extends Sunday from 13:00-17:00 to 13:00-21:30, an 8.5-hour shift (D would reach 24.5 h against 20 contracted).
2. Wednesday 14:00-15:00 (delivery): one person only. Option: move E's start from 15:00 to 14:00; or the owner covers the hour.
3. Friday 19:00-21:30: one person short for the extra cover. Option: C stays on to 21:30 (+2.5 h); or D swaps with Thursday.
4. A is down for six days in a row. Your rule says five.
5. F's Friday shift ends at 21:30. Check the limits for young workers before approving.

The numbers in those rules (five days, 11 hours' rest, nine-hour shifts) are examples, not legal advice. Put in the limits that apply where you trade, and confirm them yourself, because an assistant will treat any figure you give it as gospel and may not know the real one. Before you publish, check contracted hours against the draft, rest and break rules, the rules for young workers, your lone-working policy, holiday and swaps, and your own hours, which are the first to be squeezed. The till-sales rota tutorial shows how to turn hourly sales into heads per hour in detail.

Which assistant, and what it can take

Any mainstream assistant that accepts spreadsheets will do. Claude accepts CSV and XLSX uploads. ChatGPT's Free plan allows three file uploads a day, which is tight for the weekly run, while paid plans allow up to 80 files every three hours, and its data analysis, the part that does the sums properly, is limited on Free. Gemini's apps take up to 10 files per prompt. Individual paid plans list at about $20 a month, and a combined sheet with one tab per step keeps you inside the upload limits. Limits change, so check the current help page for your plan.

Some tills are adding assistants of their own. If yours is, use this routine as the check: run both for a month and see whose suggestions you end up trusting. The AI agents in retail tutorial covers what a one-till shop can actually run today.

Who does what once the routine is running

The routine should not live in one head. After a month, hand the data work to someone on shift and keep the approvals yourself, because the approvals are the part that needs your judgement of the shop.

On those figures staff take 45 minutes of the week and you keep 80, of which only 75 minutes is the three approvals. The one task never to delegate is the sign-off on the order, because it is the only step that spends money. If a shift lead is to run the exports, give them the five ways an export misleads as a one-page list, and ask them to flag anything odd instead of fixing it quietly.

Adding the steps one week at a time

Do not start all three together. In week one run the reorder step beside your usual ordering and compare the two. In week two let the list drive the top 20 lines only. In week three add the price check, and in week four the rota. By then the files are clean, the case-size sheet is complete, and you know which of the AI's habits you correct every time. That last list is worth writing down and pasting into the prompts as extra rules.

Four weeks of numbers: empty shelves, edits and rota gaps

Four numbers tell you whether the routine earns its 90 minutes. This is how they might move, with illustrative figures.

The pattern to look for is the last row falling towards 90 while the first row falls too. If the gaps on the shelf do not shrink, the problem is usually in the data (a missing delivery, a generic key) rather than in the AI, so go back to the five ways an export misleads before changing the prompt.

Questions owners ask before running the routine

Can AI place the order with my wholesaler for me?

Not safely, in my view. This routine produces a draft that you place yourself. Some tills are adding ordering assistants: Lightspeed, for example, has a tool that reads supplier invoices to draft purchase orders, and since October 2026 an opt-in beta of agent features for selected customers. They are worth testing alongside this routine for a month. Keep the final click yours until their suggestions have matched your own for several weeks.

How do I stop one hot week skewing the order?

Use the 28-day average as the base, and have the prompt flag any line where last week is more than 25% above or below it. Then decide using what the AI cannot see: the forecast, a school holiday, a local event, a promotion. Mark promotion weeks in the export so the average absorbs them, and note any line that ran out, because empty shelves hide demand.

Is it all right to check rivals' shelf prices?

Noting prices that are on public display is ordinary shop practice, and copying them from a public page is usually fine too, though you should check a site's terms before setting a tool to collect prices automatically. Agreeing or coordinating prices with a rival is a different matter, and competition rules in many places forbid it. Keep to observing, and ask a qualified adviser if you are unsure.

What if my till export is a mess?

Clean it once by hand rather than fighting it every week: merge the split rows for renamed products, replace generic department keys with scanned items where you can, and keep a short sheet of case sizes and suppliers. A tidy file of your 60 fastest lines is enough to start. If the till cannot export sales by item at all, ask the vendor which report can before you change anything else.

Further reads

• AI Tools for a Grocery Store: Supplier Invoices and Shelf Prices — Read supplier invoices and set shelf prices that hold margin.
• Can AI Help a Convenience Store Drive Sales? Where to Start — Where to start if the aim is more sales, not less admin.
• AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — When a rota spreadsheet stops being enough for the team.
• Restaurant Labour Cost Percentage: Test Rota Changes With AI First — Test a rota change against takings before you publish it.
• Convenience Store AI Cameras: Cost and the Privacy Checks First — Privacy checks to make before any in-store AI camera.
• Examples of AI in Retail: Eight Jobs a Small Shop Can Copy — Eight other small-shop jobs you can copy.
• How Can AI Help a Small Supermarket? Pricing, Stock and Posters — A one-store version of what the chains do: a weekly price check, a reorder list and offer posters, with the prompts, sums and checks to run each.
• Supermarket Pricing Strategy: Can AI Help Test Price Changes? — How a small supermarket tests price changes with a spreadsheet and AI: item roles, a shelf survey, a margin check, a noise check and a decision rule.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT, Claude and Gemini help pages for file-upload limits; Lightspeed announcements on its retail AI tools; the tutorial's figures are illustrative.