Farmers Market Stock Planning: How Much to Pack Each Week With AI

Pack each product from what it actually sold on comparable past market days, not from what you brought. Take the average of the last four to six similar stall days, add 10 to 20 per cent if you sold out, scale by your own wet-day or hot-day ratio from the forecast, and raise or lower it by how costly a leftover is.

The reason this goes wrong for most stalls is that they track what they took home, not what they could have sold. A stall that sells out by 11 a.m. records a modest sales figure that hides the customers who turned away at noon, and next week it packs the same amount again. AI is useful here because it builds the log and the formulas in minutes and checks your numbers for patterns you stopped seeing. It does not know your pitch, your regulars or your weather. The sheet is arithmetic you can trust, and you supply the judgement.

Follow me on Instagram@sagnikteaches

Start with a stall-day log, not a forecast

A forecast needs history, and most traders hold it in their head. One row per product per market day, filled in when the stall closes, on the van, is all it takes. Columns that earn their place:

Connect on LinkedInSagnik Bhattacharya

Fill it in on a phone or in a notebook photographed later. Either way, keep the product names identical. "Eggs dozen", "Dozen eggs" and "Eggs" become three products to a spreadsheet. If you already have an old stack of tally sheets or till receipts, ask AI to turn them into rows (see what breaks when ChatGPT reads photos and spreadsheets), then read the result against the paper, because a misread digit is easy to miss.

Subscribe on YouTube@sagnikteaches

Why sold out days hide demand

A sales figure is demand only if you had stock left at closing time. If you sold your last loaf at 10.45 and the market runs to 1 p.m., you do not know what the last two hours would have brought. The fix is simple and slightly arbitrary: for products that sold out, add an uplift to the average. Start at 10 per cent for a sell-out in the last hour, 20 per cent for a sell-out midway, and tune it by watching what happens when you pack more. Treat these as starting settings, not facts. After eight weeks of leftover counts you will have your own figure.

A worked sheet: eggs and sourdough across six market days

Illustrative numbers for a small stall selling eggs by the dozen and sourdough loaves. First, the eggs.

Dry days sold 36, 40, 40 and 44, an average of 40. Three of the four ran out, so a modest uplift is justified: 40 plus 10 per cent is 44. Wet days sold 30 and 32, an average of 31, and neither sold out, so there is no uplift. The wet-day ratio is 31 divided by 40, about 0.78. For next Saturday, a dry forecast gives 44 dozen; a wet one gives 40 × 0.78 = about 31, so pack 32 to allow a small margin. If you had packed 40 on a wet day, as on day 6, you would have carried 8 dozen home.

The sourdough shows why different products need different rules. Say loaves sell for $6.00, cost $1.80 to make, and a leftover loaf is worth nothing because day-old bread goes to the staff or the bin. Each loaf you pack and do not sell costs you $1.80. Each loaf you did not pack and could have sold loses you $4.20 of profit. The stocking rule from this sum is simple: pack one more loaf as long as the chance you will sell it is above overage cost divided by total cost, which is 1.80 ÷ (1.80 + 4.20) = 0.30. In plain words, pack more loaves until you think there is less than a three in ten chance of selling the last one. For eggs, which keep well beyond a single market day and cost far less to carry, the threshold is lower still, so a cautious number would be wrong. Notice that this is arithmetic, not forecasting: you set it once per product and the sheet applies it.

Ask AI to build the sheet, not to do the maths

Chatbots are good at writing formulas and bad at arithmetic across long tables, so ask for the first and do the second in the spreadsheet. Paste a few rows of your log, without any names of customers (there should be none), and ask:

I run a farmers market stall. Below are 12 rows of my log with columns:
Date, Product, Packed, Sold, Left over, Sold out at, Weather.
Build me a Google Sheets setup with:
1) a Log tab (these columns);
2) a Summary tab with one row per product showing: average sold on dry days,
   average sold on wet days, wet-day ratio, how many of the last 6 days sold out;
3) a Pack tab where I choose Product and tomorrow's weather (Dry/Wet) and it
   returns a suggested quantity = average for that weather x (1 + uplift),
   where uplift is a cell I set per product (default 10%).
Give me the exact formulas for row 2 of each tab and say where to paste them.
Do not calculate any of the numbers yourself.

The model will usually return AVERAGEIFS formulas (average sold where product and weather match) and a lookup for the Pack tab. Paste them into a copy of your sheet, test with three rows you know the answer to, and fix anything that returns an error before trusting it. If you prefer Excel, the same formulas work with minor changes; prompts for Google Sheets covers how to ask for them. The reason to insist that the model does not calculate is explained in why AI is bad at maths and what to check.

A sanity check you can run in two minutes

Once the sheet gives you a pack list, paste it with last week's actuals into the chatbot and ask: "Which products on this list differ from last week's packed quantities by more than 25 per cent, and what in my notes could explain it?" The model is good at that kind of comparison. A big swing with no note beside it is a sign to look again at the data (a mistyped number or a missing row), not at the forecast.

Turning weather into a number you can use

Do not ask AI what weather "usually does" to farmers market sales. There is no general rule that is right for your pitch, and invented percentages in a chatbot reply are only plausible. Take your own records instead.

1. Choose four categories: dry, wet, hot (a temperature you set, such as 28°C or 82°F), cold (below 5°C or 41°F, say).
2. For each product, compute the average sold per category once you have three or more days in each. The ratio of a category's average to the dry average is its factor.
3. Where you have fewer than three days, borrow the factor from a similar product (cold-drink factor for ice cream, for example) and flag it as borrowed.
4. Check the forecast twice: the evening before you pack and the morning of the market. Pack the main quantity on the evening forecast and carry a flexible reserve for the morning change.

Expect surprises. Wet days often cut footfall but raise the share of regulars who come anyway, so your best-known lines may drop by less than the rest. Hot days push up cold drinks and salad and push down bread. Your own factors will show this; a generic one will not.

Pack in two layers: the core load and the reserve

The forecast gives one number, but weather and footfall are uncertain, and the cost of being wrong is not equal across products. Pack in two layers. The core is the forecast. The reserve is a small extra amount you pack only for items that keep or have a second life.

• Keeps well (eggs, honey, preserves, dried goods): carry a reserve of 10 to 20 per cent; a leftover costs you only the carry.
• Short life, can be discounted (bread, pastries, soft fruit): pack the core and one box of reserve you can sell at half price late in the day or take to another outlet.
• Very short life (cut herbs, salad bags, fresh meat on ice): pack close to the core and plan the leftover (a staff meal, a discount bundle, a freezer batch) before you leave home.

Put the reserve in a labelled box at the back of the van. If the day goes well, you open it. If it does not, it goes home intact and does not have to be unpacked, reduced and relabelled.

Four example days and how to handle each

A festival weekend. The log does not know about a large local event. Add a note column entry ("festival, expect double footfall") and use a manual multiplier. The first time you will guess; after two events you will have a ratio. Until then, pack the core at 1.5 times the usual and double the reserve of items that keep.

A new neighbouring stall. A baker sets up two pitches along and your bread sales drop 20 per cent. The sheet will take three weeks to show it, because the average includes older days. Use "last four days" rather than "all days" for any product with a competitor change, or add a note and exclude the earlier rows from the average.

A price change. Raising the price of honey from $8 to $9 and watching the weekly sales will show whether demand dropped. Keep rows from before and after the change apart (a "price" column helps), and do not average across both.

A product trial. For a new line, there is no history. Pack a small amount (half of what your gut says), note the sell-out time, and treat the first four days as a test that builds the log.

A weekly rhythm that takes twenty minutes

1. Market evening, five minutes: enter packed, sold, left over and sell-out times while it is fresh.
2. Midweek, ten minutes: look at the Summary tab; check any product with two or more sell-outs or more than 20 per cent leftover for two weeks running.
3. Evening before, five minutes: check the forecast, pick Dry or Wet, read the Pack tab, adjust for known events, write the pack list.

If you also sell online or take orders, the numbers fit together: pre-orders are demand you already know, and the market forecast only has to cover the walk-up customers. AI for farmers market vendors covers how to take those pre-orders. If you run the stock from a farm shop or boxes as well, weekly boxes and labels with AI shows how to plan both from one record.

Recording the day faster on the van

The log fails when it takes more than five minutes, so make it light. Three ways that work for traders with cold hands and a queue:

• Tally as you go. A laminated card with one line per product and a pen on a string. At close, count the ticks and the leftovers and type one number per product.
• Voice note at close. Say it in a fixed order: "Eggs, packed forty-four, sold forty-four, sold out at half past twelve. Sourdough, packed thirty, sold twenty-six." Ask the chatbot or your phone's transcription to turn the recording into rows with the same columns. Read them against what you said before pasting into the sheet, since numbers are the part of a transcript most likely to go wrong, and a "forty" that became "fourteen" would skew every average that follows.
• Photo of the tally card. Take it as the stall closes, and have the model read it into a table. Check each line against the card.

After any of these, run a quick check by asking the model "which rows have sold greater than packed, or leftover that does not equal packed minus sold?" A well-kept log should have none, and the question catches typing slips before they reach the averages.

Several markets and one van

If you attend two or more markets, treat each as its own stall. A Saturday town market and a Wednesday evening street market may differ in customers, price tolerance and what sells. Keep one log with a Market column and let the Summary average by product and market, not by product alone. When you start a new market, borrow the factors from the most similar one for the first four visits and mark them as borrowed. The sheet will tell you when the borrowed numbers have been replaced by real ones.

A monthly review prompt

Once a month, paste the Summary tab (not the whole log) into a chatbot with a short brief. The aim is not a forecast but a second pair of eyes on patterns.

Here is my market summary for the last 8 weeks: product, average packed,
average sold, average left over, number of sell-outs, wet-day ratio.
Tell me: 1) which three products lose me the most money in leftovers
(use my cost per unit column), 2) which three sell out most often,
3) any product whose wet-day ratio is far from the others and what that
suggests, 4) three questions I should answer from my own knowledge before
changing any pack quantity. Do not recommend prices.

A typical reply points out, say, that the soft fruit line produces the largest leftover cost and that preserves sell out on cold days. The model is reading your numbers back to you, which is its strength. The decision to change a quantity stays with you, and it should wait for a second week of evidence.

When the forecast sheet misleads you

A trader tracks sales for eight weeks, builds the sheet, and sees the recommended pack for tomatoes fall every week. Each week they pack less, sell out earlier, record a smaller figure, and the sheet recommends packing less still. The sheet is a mirror: it learns from sales, and sales were limited by what was packed. The fix is the sell-out uplift and the "sold out at" column, plus a rule that says a product cannot fall below its last week's pack if it sold out in the last hour. The point is wider than tomatoes. Any system that learns from sales needs a way to learn from the sales you could not make.

Another failure is a sheet that is never updated. Weather factors drift, new competitors arrive, and a product that was a star last season is a laggard this one. Review each factor each season, and when in doubt, trust the most recent four days over the long-run average. For the broader approach to finding lines that no longer sell, find dead stock and slow movers with AI is a useful companion.

A filled-in pack list for next Saturday

A finished pack list for a dry forecast might look like the one below. All numbers are illustrative.

Print it, tick off what you load, and write down the real figures tonight. Next week's sheet starts from those numbers, not from this plan.

Questions market traders ask about forecasting what to pack

How many market days do I need before a forecast is useful?

Four to six days of the same stall in the same season give you a usable average for each product, and ten or more let you separate wet days from dry ones. Until then, pack from the average and keep a written note of every sell-out and every leftover so the sheet improves quickly.

Should I trust the AI's number or my own judgement?

Use the sheet for the starting quantity and your own judgement for what it cannot see: a road closure, a festival, a new neighbouring stall, a supplier problem. If you override the sheet, write down why, then check afterwards who was right. That record is what makes the next forecast better.

What if I sell out every week?

Then your sales figures understate demand, and the average will keep pulling you down. Record the time you sold out and raise the starting quantity by 10 to 20 per cent for those products, then watch whether the leftover grows. A stall that never has leftovers is probably leaving sales behind.

Does this work for produce that does not keep?

Yes, and it matters most there. Perishable items carry the highest cost of over-packing, so the sheet should use a lower pack-up point for them, and you should plan a markdown or a second outlet for what is left rather than carrying it home.

Further reads

• How to Use AI to Set Reorder Points and Prevent Stockouts — Reorder points for the stock you buy in for the stall.
• How Bakeries Can Use AI to Predict Demand and Cut Unsold Stock — The same demand-prediction idea applied to a bakery counter.
• How Food Trucks Can Use AI to Plan Stock, Pitches and Posts — Another mobile seller planning stock around location and weather.
• AI in Food Retail: Jobs a Deli or Farm Shop Can Hand Over First — Other food-retail jobs worth handing over early.
• How to Forecast Covers and Cut Food Waste With AI — A sister method for forecasting demand to cut food waste.
• Free AI for Agriculture: What a Small Farm Gets Without Paying — What free AI tools can and cannot do for a small producer.
• How Small Shops Can Use AI to Forecast Stock and Reorder — How small shops pair a simple forecast with reorder points.
• Food Truck Inventory Management: Build Tomorrow's Prep List — A week of sales exports, a recipe matrix, par levels and waste notes turned into a Friday prep list, with the prompt that writes it each evening.
• Ice Cream Inventory: Forecast Tubs by Flavour, Weather and Weekday — Build a simple sheet that turns your sales log and the weather forecast into tubs per flavour, with a reorder rule and a spoilage check.
• Pop-Up Shop Business Plan: Stock, Rent and Staffing in ChatGPT — One fact block, five prompts and a hand check turn a rent quote, stock costs and a rota into break-even days and a go or no-go rule.
• Starting a Chocolate Business: Costing, Labels and Where to Sell With AI — Per-bar costing, allergen label checks and a margin comparison of farmers market, online and wholesale, with illustrative numbers.
• Microgreens Business Plan: Tray Costs, Restaurant Orders and Waste With AI — A microgreens plan built from one tray: what it costs, what it yields, what a standing restaurant order is worth and how a waste log changes the sums.
• Selling Beef Boxes Direct: Hanging Weight, Cut Sheets and Pricing With AI — Convert hanging weight to retail cuts, price a quarter, half or box, and build the order form and deposit steps with AI checking your sums.
• Selling Produce to Restaurants: Availability Lists and Chef Emails With AI — A weekly availability list template and four chef email scripts for growers, with AI turning your harvest notes into the list.
• Subscription Box Curation: Match Items to Surveys With AI — A 120-subscriber craft box walked through: survey fields, item tags, matching rules, stock shortfalls, manual overrides and a margin check per box.
• CSA Farm Shares: Sign-Ups, Pickup Reminders and Swaps With AI — From sign-up form to members sheet to weekly pickup reminders, with box notes and skip-week rules that an AI assistant can draft.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: standard newsvendor-style stocking logic (cost of a leftover against profit on a missed sale), applied to the reader's own market records; no external statistics used.