Inventory & Demand Forecasting with AI
Inventory in a restaurant is a slow leak. Over-ordered protein gets walked. Forgotten dairy goes off. Spec'd-down portions add up to thousands of dollars across a month. The right level of inventory is the level that just barely covers the week's forecasted sales — not a pound more.
AI gives even small operators access to forecasting techniques that used to require an expensive consultant. This lesson covers the practical workflow.
What You'll Learn
- How to use AI to forecast next week's sales by daypart
- How to build a smarter prep list and par sheet from forecasted demand
- How to identify slow-movers and 86 candidates from your POS data
- How to combine weather, events, and historical data into one prompt
Sales Forecasting in 2 Minutes
Your POS has the data. AI does the analysis.
Export the last 12 weeks of daily sales (covers, sales, average check, daypart split) from Toast, Square, or Clover into a spreadsheet. Paste into a prompt:
Act as my sales forecasting analyst.
Below is 12 weeks of daily sales data for my Italian
restaurant. Columns: date, day_of_week, weather,
local_events, lunch_covers, dinner_covers,
lunch_sales, dinner_sales.
[paste data]
Forecast next week's daily covers and sales by daypart.
Account for:
- Day-of-week pattern
- Holiday or event drift (Father's Day is Sunday this
week)
- The weather forecast: [paste forecast]
Output a clean table:
Day | Lunch Covers | Lunch Sales | Dinner Covers | Dinner Sales | Notes / risk factors
You'll get a forecast that beats most operator gut estimates within 5–10%. Use it to guide ordering, scheduling, and prep.
From Forecast to Prep List
Once you have the cover forecast, AI builds the prep list:
[paste house context]
Forecasted dinner covers next week:
- Wed: 65, Thu: 80, Fri: 110, Sat: 130, Sun: 70
Use last 30 days of menu mix (paste below) to estimate
how many of each dish we'll sell each night.
Then convert to a prep list for the kitchen:
- Bolognese (per portion 4oz) — total batch yield
needed
- Cacio e pepe sauce base — total quarts
- Marinara — total quarts
- Pizza dough — total balls
- Bread — total loaves
Format as a printable Mon–Sun prep grid with quantities
to make each day.
Menu mix for last 30 days:
[paste]
You just turned 90 minutes of chef-and-owner whiteboard math into a single prompt.
Smarter Pars
Pars (the minimum stock level to maintain) drift over time. They were set when you opened or after that one bad weekend. AI re-bases them quickly:
Act as my inventory analyst.
Below is 8 weeks of weekly produce usage (purchases
minus end-of-week count) and 8 weeks of sales.
[paste data]
For each SKU:
- Calculate average weekly usage
- Calculate weekly variance (standard deviation)
- Recommend a par level = average + 1.5 * stdev,
rounded to nearest case unit
- Flag any SKUs where current par is more than 25%
above your recommendation
- Flag any SKU we've stocked out on
Return a sorted table with: SKU, current par,
recommended par, change, reason.
This single prompt typically uncovers $400–$1,200 a month in over-ordered produce for a typical neighborhood operation.
Identifying 86 and Slow-Mover Candidates
Some menu items move so slowly that the prep cost, freshness loss, and menu real estate add up to negative contribution.
[paste house context]
Below is 60 days of menu mix (item, count, price,
food cost).
[paste]
Identify:
1. Items selling fewer than 1 per day on average
2. Items where total monthly contribution (gross
profit × count) is below $200
3. Items with food cost > 35% AND fewer than 2 sales
per day
For each, recommend one of:
- Cut from menu
- Re-engineer to lower food cost
- Reposition / rename and keep
- Replace with seasonal alternative
Show your reasoning.
Run this every quarter. Cut the dogs. Let the puzzles get one season to prove themselves with new copy or a new menu spot.
Layering in Weather and Events
Weather and local events move covers more than most owners account for. AI handles the layering:
Forecasted base covers for Saturday: 130
Layer in:
- Forecast: 88°F and sunny (typically pulls 8–12% of
covers to outdoor patios — we have one)
- Local: a major outdoor concert 3 blocks away
starting at 7:30 PM (typical impact: brings 15-20
walk-ins between 5 and 7, then covers drop sharply
after 8)
Re-forecast covers by hour from 5 PM to 11 PM.
The hourly cover model tells your floor manager when to push closes, when to staff up, and when to tighten the kitchen.
What AI Cannot Forecast
AI doesn't know:
- The new restaurant opening across the street next month
- That your dishwasher just quit and Saturday will run slower
- That the Eagles game is on and your sports-bar competitors will be packed
- The personality of your weekend dinner crowd
It models the past. You layer in the things only you know about the future. The combination beats either alone.
A Realistic Sunday Night Workflow
In 25 minutes:
- Forecast next week's covers and sales → 4 minutes
- Generate prep list from forecast and menu mix → 5 minutes
- Re-base produce pars from last 8 weeks → 6 minutes
- Run 86/slow-mover analysis → 5 minutes
- Layer Saturday weather + local event into hourly forecast → 5 minutes
Send the prep list to the kitchen. Send the par changes to the broadline rep. Sleep better than you used to on a Sunday night.
Key Takeaways
- AI forecasts next week's sales by daypart from your POS export — typically within 5–10% of actual
- Convert forecast directly into a kitchen prep list to cut over-prep waste
- Re-base your pars every 8–12 weeks; AI catches drift you'd miss
- Run a quarterly slow-mover/dog analysis on your menu mix
- Layer weather, events, and local context — it's the human edge AI can't replicate alone