A busy Saturday can look like a great day until you see the overtime, the discarded prep and the delivery fees. A quiet Tuesday can be more profitable than it feels. Your sales total tells part of the story. The useful question is what happened underneath it, and what you should change next week.
Data science brings measurement, statistics and sometimes predictive models to those decisions. For a restaurant owner, the starting point can be much simpler: put a few reliable numbers together, ask a specific question and check whether the answer improves service or profit.
You do not need a data team to begin. You need a repeatable routine that fits between opening and closing.
Start with a decision, not a dashboard
Choose a decision you make every week. Perhaps the lunch crew is rushed on Thursdays, the same ingredients keep appearing on your waste sheet, or a popular delivery item contributes less than expected.
Write the question in a way you can test. “How is the restaurant doing?” is too broad. “Would moving one prep hour from Tuesday to Thursday reduce lunch delays without increasing weekly labor hours?” gives you a clear comparison.
Name one person who will review the result and one action they can take. A beautiful report is less useful than a small change your manager can explain to the team.
Build a dependable weekly scorecard
Start with completed weeks and compare the same days and service periods. A Friday dinner should not be judged against a Monday lunch. Keep your sales definition consistent: net sales should exclude sales tax and tips and account for discounts and refunds. Check how your POS labels those fields before combining reports.
Bring together three sources:
- Sales: orders, items, discounts, refunds and order channel, split by daypart.
- Labor: actual paid hours, wages and the employer costs you include in your labor definition.
- Food: beginning inventory, purchases, ending inventory, recipe costs and recorded waste.
Reconcile a small sample back to the original reports. Check business dates, time zones, duplicate orders and missing timecards. Make sure a case, a bottle and an ounce are not being treated as the same purchasing unit. Document your definitions so next week’s numbers mean the same thing.
Four calculations make a useful first scorecard:
| Measure | Calculation | What to investigate |
|---|---|---|
| Labor cost percentage | Labor cost ÷ net sales × 100 | A change in staffing, wage mix or sales volume |
| Sales per labor hour | Net sales ÷ paid labor hours | Whether staffing matches demand, alongside service quality |
| Food cost percentage | Food consumed ÷ relevant food sales × 100 | Purchasing prices, portions, waste or inventory-count errors |
| Average order value | Net sales ÷ completed orders | Changes in menu mix, discounts or ordering channel |
For food consumed, use beginning inventory plus purchases minus ending inventory, with consistent adjustments for transfers and returns. Purchases alone are not the food used during the week. Separate beverage costs and sales if you manage them separately.
There is no single target that suits every restaurant. Service model, menu, wage rates and opening hours all matter. Compare your own consistent baseline first.
Forecast the next shift with a simple baseline
A forecast is a planning estimate, not a promise. Start with the last four comparable weekdays for each daypart. Calculate their average orders, then mark exceptions such as holidays, closures, promotions and unusually large catering orders. Keep the raw numbers; record why you excluded an exception rather than silently deleting it.
Suppose four ordinary Thursday lunches had 92, 100, 96 and 112 orders. Their average is 100. That gives your manager a starting point for prep and staffing. If the nearby office is closed next Thursday, adjust the plan and write down the reason.
After service, compare forecast and actual orders. A forecast of 100 against 110 actual orders missed by 10 orders, or about 9.1% of actual demand. Review the errors across several shifts. If you consistently underpredict lunch, investigate the pattern before adding a more complicated model.
Later, weather, reservations and local events may improve a forecast. Keep a simpler baseline alongside any new model. More inputs are useful only when they make the plan more accurate and easier to act on.
Use menu data to protect contribution, not just volume
A best seller can be expensive to fulfill. Review each item’s selling price after discounts, recipe cost, packaging and channel-specific fees. Be explicit about which costs your calculation includes.
Consider an illustrative $12 bowl with $3.60 of ingredients and $0.60 of packaging. It leaves $7.80 before labor, rent and other costs. If the order also incurs a $2.40 delivery commission, the amount left is $5.40. Neither figure is net profit, but the difference matters when you judge a delivery promotion.
Plot popularity against contribution per item. A high-contribution item that rarely sells might need clearer menu placement. A popular low-contribution item might need a portion review, a supplier check or a carefully tested price change. Before removing anything, consider whether it attracts guests who buy other items and whether your cost records are current.
Do not assume a discount worked because order count rose. Compare the contribution after the discount and fees, and account for customers who might have ordered at full price anyway.
Read labor and service together
Lower labor percentage is not automatically a better shift. Understaffing can produce longer waits, missed breaks, mistakes and a crew that cannot keep up. Track a service measure next to the financial one, such as ticket time, remakes or unfilled stations.
Break the day into useful operating windows. Daily totals can hide a short lunch rush and a long quiet afternoon. Include opening, prep and cleaning work when you evaluate staffing; those hours still matter even when the register is quiet.
Use the numbers to plan coverage, then ask the manager what happened. A bus group, equipment failure or training shift may explain a result better than a trend line. Check scheduling rules, required breaks and employee availability before making changes. A model should support judgment, not make disciplinary or staffing decisions on incomplete data.
Make waste visible while the reason is still fresh
A waste log becomes useful when staff can complete it quickly. Record the item, quantity, approximate cost and reason: overproduction, spoilage, preparation error or a returned order. Use consistent units and a short list of reasons.
At the end of the week, select one recurring source of avoidable waste. If prepared tea is repeatedly discarded late in the evening, test a smaller final batch. Track stockouts and guest wait times as well as waste so a lower discard total does not hide a service problem.
Inventory variance can also reveal a problem, but it does not identify the cause by itself. Missing transfers, inaccurate recipes and counting mistakes can look like loss. Verify the records before drawing conclusions about people.
Run one small experiment
Change one thing you can reverse. Set a baseline, a test period and a guardrail before you begin. For the Thursday lunch example, your plan might be:
- Baseline: four comparable Thursdays, using orders, paid hours and ticket times.
- Change: move one prep hour earlier without increasing total weekly hours.
- Primary result: fewer lunch tickets exceeding your service target.
- Guardrail: no missed breaks, added overtime or increase in remakes.
- Review: compare the next few Thursdays and record unusual events.
Small samples are noisy. A better week does not prove the change caused it. Look for a repeatable improvement and ask staff what felt different. Keep the change if the evidence and the operating experience agree; revise it if they do not.
Where AI helps, and where to keep control
AI can help summarize reports, flag unusual movements and turn a question into an analysis. Predictive models can help with demand planning when you have enough consistent history. Ask to see the date range, source fields and calculation behind an answer before using it to change a schedule or a price.
Keep access limited to what each role needs. Most operating questions do not require employee bank details or identifiable guest records. Use aggregate information where possible, check retention settings and avoid uploading sensitive exports to tools your business has not approved.
Clockout brings restaurant operations into one place. Explore scheduling, food cost and Ro, the AI manager, or talk through your workflow with the team. Confirm the data connections and reports available for your setup before relying on them.
Your next weekly review
Pick one recurring decision. Collect one complete week of comparable records. Check the definitions with the manager who knows the shift. Then choose one change and decide in advance what would count as an improvement.
The goal is a restaurant that learns from each week: what to prepare, where to put the next hour of labor and which menu decisions deserve a closer look.
All worked examples in this article are illustrative, not Clockout customer results or universal restaurant benchmarks.