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
اختبار قصير
التالي · الدرس 9 من 13
Restaurant Marketing & Social Media Content
قراءة 5 دقيقة
أنهيت 8 من 13 درسًا

اسأل Byte عن هذا الدرس
محادثة خاصة حول هذا الدرس. قرأ Byte الدرس كاملًا، ويجيب خلال ثوانٍ، ويُظهر القسم الذي تأتي منه كل إجابة. يمكنك السؤال بأي لغة.
الأسئلة والأجوبة
أسئلة عامة حول هذا الدرس. يجيب AI Tutor فورًا بلغتك، ويراجع فريق FreeAcademy الإجابات، ويمكن لكل المتعلمين قراءتها.

