A restaurant group operating six fast-casual locations in the Orlando area was throwing away food. Not dramatically — no dumpsters overflowing with perfectly good meals. Slowly and persistently. A little extra chicken grilled every morning. A few too many portions of rice prepped for the afternoon. Salad greens that wilted because the prep quantity assumed a busy lunch that never materialized.
The food cost across six locations averaged 34.2% of revenue. Their target was 30%. The 4.2% gap, applied to $4.8 million in annual revenue, was approximately $200,000 per year going into the trash.
The executive chef knew the problem. Every kitchen manager knew the problem. The problem was the prep sheet.
How prep sheets work
Every restaurant starts its day with a prep sheet. The prep sheet tells the kitchen how much of each ingredient to prepare for the day. Grill 40 pounds of chicken. Cook 60 quarts of rice. Prep 25 pounds of mixed greens. Make 8 gallons of house dressing.
The prep quantities are determined by the kitchen manager, who bases them on experience, the day of the week, the weather, any local events, and how much was left over yesterday. A good kitchen manager gets close. A great kitchen manager gets very close. Nobody gets it exactly right because predicting how many people will walk through the door on a Tuesday in March is not a solvable problem.
The systemic error is always in the same direction: over-prep. Under-prepping means running out of an item during service, which means telling a customer "we are out of the chicken bowl" at 1:15 PM. That customer does not come back. Running out during service is a catastrophic failure that kitchen managers avoid at all costs.
So every kitchen manager builds in a buffer. 10% extra on proteins. 15% extra on sides. 20% extra on items that spoil quickly because it is better to have too much fresh product than to serve something questionable.
The buffers are rational individually and wasteful collectively. Every location, every day, over-prepping by 10-20% on everything.
What the tool does
The tool generates daily prep sheets for each location based on historical sales data, day of the week, weather forecast, local event calendar, and recent trends.
It pulls the previous 52 weeks of sales data by item, by day of the week, by location. It adjusts for trends — if chicken bowl sales have been increasing 2% per month, the forecast reflects that. It factors in weather — rainy days reduce foot traffic by 8-12% at their locations. It checks for local events — a convention at the nearby hotel increases lunch traffic by 15-20%.
The output is a prep sheet with specific quantities for each item, a confidence level for each quantity, and a recommended buffer amount.
Here is where the design matters. The tool does not eliminate the buffer. It right-sizes the buffer based on the confidence of the forecast.
For a typical Tuesday with no events and clear weather, the forecast confidence is high. The historical pattern is consistent. The recommended buffer is 5% instead of the kitchen manager's typical 15%.
For a Saturday when there is a music festival two blocks away, the forecast confidence is lower because event impact varies. The recommended buffer is 12% because the uncertainty is higher.
The kitchen manager reviews the prep sheet, adjusts if they have information the tool does not (a regular catering order that has not been entered in the system yet, a staff shortage that will slow service and reduce throughput), and finalizes.
Why the kitchen managers trusted it
This was the hardest part of the project. Kitchen managers who have been writing prep sheets for 15 years do not want a computer telling them how much chicken to grill. Their experience is real. Their instincts are usually right. Telling them to trust a smaller number feels like asking them to risk running out during service.
We did three things that built trust.
First, we ran the tool in shadow mode for four weeks. The kitchen managers wrote their prep sheets as usual. The tool generated its own prep sheets. At the end of each day, we compared both to actual sales. The tool's prep quantities were closer to actual sales than the kitchen manager's prep quantities in 78% of cases. That number was convincing.
Second, we showed them the waste data. Each kitchen manager knew they overprepped. They did not know by how much. When the tool showed that Location 3 had prepped an average of 18% more chicken than it sold every Monday for the past six months, the kitchen manager at Location 3 saw his own waste quantified for the first time.
Third, we let them override. The tool recommends. The kitchen manager decides. If they want to add 10% to the chicken quantity because they have a feeling about today, they do it. The override is logged so the tool can learn whether their instinct was right, but the decision remains theirs.
After two months, overrides dropped from 40% of items to 12% of items. The kitchen managers learned to trust the numbers because the numbers worked.
The results
Food waste across six locations dropped from 34.2% food cost to 23.3% over six months. That is a 32% reduction in waste, translating to approximately $148,000 in annual savings.
No location ran out of a menu item during service in the first three months. One location ran out of a side item on a day with an unexpected catering order that was entered late. The kitchen manager considered this a data entry problem, not a tool problem, which is the correct diagnosis.
The tool also surfaced operational insights the team had not expected. Location 5 was consistently over-prepping salad greens by 30% on Mondays. Investigation revealed that the Monday kitchen manager was prepping based on Saturday's numbers (the previous busy day he worked) instead of Monday's historical average. A simple coaching conversation fixed a waste pattern that had persisted for over a year.
The cost
Six days of build time. The tool connects to the POS systems at each location and pulls a daily weather forecast from a free API. The event calendar is maintained manually — someone enters known local events once a week, which takes about 10 minutes.
The executive chef's summary: "The tool does not cook. It does not manage the kitchen. It tells us how much to prep based on data instead of feelings. The feelings were usually close. The data is closer. And 'closer' across six locations across 365 days adds up to real money."
Get posts like this in your inbox
No spam. New articles on AI strategy, governance, and building with AI for small business.