It's six in the evening, the shop has just closed, and you're staring at the order notebook. You have a vague idea of what's missing, another vague idea of what's left over, and a customer who messaged on WhatsApp asking whether yesterday's item is back. AI can help, but only if you can first say with certainty what is on your shelves.
Genoa, 1340: the books of two officials
Let's go back nearly seven hundred years. According to the Italian Wikipedia entry on double-entry bookkeeping, the first accounts certainly kept in double entry are those of the massari of the Compagna comunis of Genoa, in 1340. There were two massari, who looked after the commune's finances; two other officials, the maestri razionali, audited their work. Amounts were written in Roman numerals, which is no fun for adding up. A recent study of the 1340 ledger appeared in the journal De Computis, a sign that historians are not done discussing it.
And you'll say: Professor, I sell shoes, hardware, gifts, I'm not running a city. Wait. The idea behind double entry is simple: every movement is written twice, from two sides, so errors surface. If a massaro recorded an expense, a matching entry had to exist somewhere else. And a second office checked.
Well, your shop today is in the same position: goods coming in, goods going out and goods sitting on the shelf must add up. AI is a good extra bookkeeper, but only if the books are well kept. With a wrong inventory, even the sharpest tool produces wrong forecasts, only faster.
Where AI helps small retail
- Reading sales data: exports from the till or management software, summarized in a sheet showing what moves and what sits.
- Reorder drafts: a list of what to order from suppliers, which you approve or correct.
- Customer messages: replies to recurring questions about hours, availability and returns, written in your style.
- Product descriptions for your website, online shop or social media, as in our product sheets guide.
- Seasonal reminders: what to reorder before a holiday, based on what happened in past years.
| Task | AI prepares | Stays with you |
|---|---|---|
| Reordering | A list proposed from recent sales | Final quantities, supplier negotiation |
| Slow stock | A list of what doesn't sell | Deciding discount, return or clearance |
| Customer replies | A draft with hours and availability | Checking against real stock, tone, exceptions |
| Product sheets | Text from supplier data | Checking sizes, materials, prices |
How to start without upending the shop
- Do a real inventory. Once, calmly, without trusting the software. Note the differences: they are your map of errors.
- Pick five products. Your best sellers. You test on these, not on the whole catalog.
- Export sales from the last few months into a sheet. If your software can't export, start recording by hand for two weeks.
- Ask AI for a summary, not a decision. For example: which of these five products sold best in the weeks with fewer customers?
- Compare with your experience. If a suggestion surprises you, check the data before trusting it or yourself.
A caution on customer data: if you upload names, phone numbers or purchase history to an outside tool, privacy rules come into play. Before doing so, read the guide on GDPR and artificial intelligence and the rules for employees using ChatGPT. For reordering, sales data without names is enough.
A typical scenario
Imagine a shop with two people at the counter and a modest stockroom. Every Monday the owner spends half an hour working out what's missing, ordering from memory. With a list of the week's sales and a reorder draft prepared by AI, the half hour shifts from remembering to checking. I won't promise how many minutes you'll save: you measure them, counting the time before and after a month's trial.
Mistakes to avoid
- Starting from the tool, not the problem. First decide what you want to know, then pick the tool.
- Trusting a dirty inventory. Forecasts built on wrong data are wrong with great confidence.
- Giving AI the last word on orders. The supplier, the season and the neighborhood are things you know.
- Replying to customers without reading. A draft always gets reread, especially on availability and prices.
- Uploading personal data without checking where it ends up.
What to do on Monday morning
- Physically count your five best sellers and compare with the software: note the gap.
- Export or record the last weeks of sales in a sheet, without customer names.
- Ask an AI assistant to summarize that data and compare it with what you already know: where they differ, investigate.
The massari of Genoa did not trust a single voice: they wrote twice and had it checked. Do the same with your books and with AI. If you want to bring order to your shop's inventory, reordering and messages, write to us for a free introductory call. You'll find our paths on the services page and the one for entrepreneurs.

