That Feeling When Costs Are Higher Than They Should Be
Most restaurant owners know their costs are "too high" but can't pinpoint why — where the waste is, how much, and what to fix.
I was the same, until I let AI analyze the actual data.
How AI Helped Analyze
Step 1: Gather the Data
You probably already have this in Google Sheets:
- Recipe specs — what ingredients each dish uses and how much
- Sales data — how many of each dish sold (from POS/GrabFood/delivery apps)
- Actual usage — how much raw material was actually used (from inventory)
Step 2: Calculate "Should-Use" vs "Actual-Use"
This is the core of the analysis:
Should-use = Sales × Recipe
e.g., 100 beef stew bowls × 0.15 kg/bowl = should use 15 kg beef
Actual-use = Actual withdrawals from inventory
e.g., 20 kg beef was used
Variance = (Actual - Should) / Should × 100%
= (20 - 15) / 15 × 100% = +33.3% 🔴
AI does this calculation for every ingredient, every dish, every sales channel all at once.
Step 3: Find the Culprits
The analysis looked like this:
| Ingredient | Should-use (kg) | Actual (kg) | Variance |
|---|---|---|---|
| Sirloin | 12.5 | 18.2 | +45.6% 🔴 |
| Pork belly | 8.3 | 10.1 | +21.7% 🟡 |
| Rice | 25.0 | 26.5 | +6.0% 🟢 |
| Cilantro | 2.1 | 5.8 | +176.2% 🔴 |
What We Discovered
The total variance across the restaurant was approximately +30% — meaning we were using 30% more raw materials than we should have been. Translated to money = tens of thousands of baht in hidden losses every month.
Root Causes AI Helped Identify
1. Yield Was Lower Than Expected
Fresh meat purchased at 1 kg doesn't yield 1 kg of usable product — tendons, fat, and waste must be trimmed. Our recipe specs had incorrect yield numbers.
2. Over-Withdrawal Without Checks
Staff withdrew inventory without anyone comparing it to actual sales volumes.
3. Delivery vs Dine-In Recipes Differed
Dine-in portions were one size, delivery portions another — but the recipe specs didn't distinguish between them.
The Fixes
📋 Fix recipe specs — use real yield numbers
📦 Set up withdrawal controls — auto-compare to sales
📊 Weekly analysis — don't wait for month-end
🔍 Separate delivery/dine-in recipes — different cost structures
No Consultant Required
All you need:
- Data in Google Sheets — recipes + sales
- AI asking the right question — "Given these sales, how much raw material should we have used?"
AI isn't magic. It just calculates in seconds what would take humans days to do by hand.
Key Takeaway
If you have recipe specs and sales data in any system — you can get AI to analyze your costs immediately. No consultant needed. No waiting for month-end reports.
Try It Yourself
- Open your Google Sheet with recipe specs
- Prepare daily sales data (from POS, delivery platforms)
- Ask AI: "Calculate should-use vs actual-use by ingredient"
- Look at the results — the highest variance is where to fix first
The data doesn't have to be perfect. Just starting the analysis reveals things you've never seen before.
