The problem
Prepping for an event is guesswork: over-prep and food is binned, under-prep and you sell out by two o'clock. If a trader bins about 15 portions an event at $4£3€4A$6C$6 food cost each, over 8 events a month, that's $479£360€428A$691C$677 of food a month.
What the AI does
Keep a simple log after each event: date, event, weather, rough footfall, portions sold of each item from your till's item sales, portions binned and when you sold out. Before the next booking, paste the log for similar events, with no customer details, into ChatGPT, Claude or Gemini and ask for a prep estimate per item with its reasoning. You decide the final numbers from your own experience and the forecast. The assistant only does arithmetic on what you give it; it can't know a new event's crowd.
A worked example
A Greek street food trader in Brighton does about 8 markets and festivals a month and logs each one in a notebook. Before each event the owner pastes the last few similar events into an assistant and adjusts its estimate.
Before
About 15 portions binned per event x 8 events x $4£3€4A$6C$6 food cost = $479£360€428A$691C$677 a month
After
About 8 portions binned per event x 8 events x $4£3€4A$6C$6 = $255£192€228A$369C$361 a month
What it's worth
About $223£168€200A$323C$316 a month saved in binned food ($479£360€428A$691C$677 minus $255£192€228A$369C$361), if waste falls from 15 to 8 portions an event; that fall is an assumption that depends on keeping the log. The free plans of ChatGPT, Claude and Gemini can do this; a paid plan is up to about $27£20€24A$38C$38 a month for one person if you hit usage limits.
How to set it up
- After every event, log the date, event, weather, rough footfall, portions sold per item, portions binned and sell-out time.
- Before the next booking, paste the log for 3 to 5 similar events into ChatGPT, Claude or Gemini and ask for a prep estimate per item with its reasoning.
- Check the arithmetic and adjust for what you know: the forecast, the event's size this year, and any new item.
- Prep to your final numbers and log what happens, so the next estimate is better.
- Review the log monthly and drop items that are binned most often.
What goes wrong
- The estimate is only as good as the log. Skip a few events and it guesses from too little.
- An assistant can make arithmetic slips. Check totals before you buy stock.
- A new festival has no history; use the organizer's expected footfall and your own judgment, not the assistant's.
- Don't paste customer names or card details into consumer AI plans; the log only needs sales numbers.