What problem does it solve?
A restaurant manager building next week's schedule and prep list is really making a forecast by hand: how busy will Tuesday lunch be, how many burgers will sell before the game starts, how many people does that need on the line. Fourth's own case study on Chili's describes the result of doing this manually at scale: "Chili's managers' manual forecasting kept them chained to their laptops instead of out on the floor with their guests and staff," made worse by a spike in turnover after the Covid-19 pandemic that left newer managers struggling with inconsistent, inaccurate forecasting.
Get the forecast wrong and a restaurant is either overstaffed, which eats margin directly, or understaffed, which shows up in slow service, wasted or run out prep, and a harder shift for the team on the floor, which itself feeds the turnover that made the forecasting problem worse in the first place.
How does it work?
- Learn the pattern. The model trains on each restaurant's own point of sale history, the same way a long tenured manager learns a location's rhythm, but across every location and every daypart at once.
- Add what changes the pattern. Local events, school calendars, weather and known promotions or menu changes are layered on top of the base sales pattern.
- Forecast sales and traffic per daypart. The output is a projected sales and guest count for each shift, not just a single daily number, so staffing can flex within the day.
- Turn the forecast into a schedule. Labor scheduling software uses the forecast, plus labor law constraints and staff availability, to propose a shift by shift schedule for a manager to review and publish.
- Turn the same forecast into a prep list. The same demand number drives how much of each menu item to prep, so prep and staffing move off the same, single forecast instead of two separate guesses.
- Compare forecast to actual, every period. Forecast accuracy is tracked against actual sales so the model, and the manager's trust in it, improves over time.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Hours saved | Not pooled | 600 hours to 56,000 hours | 2 | 2 vendor |
Value drivers: Lower cost to serve, Employee productivity, Customer experience.
Indicative value
A casual dining chain with 1,200 restaurants
USD 936,000 to USD 1.4 million
Annual value of general manager time returned from manual forecasting per year
How this is calculated
Formula: restaurants * gmHoursSavedPerWeek * 52 * fullyLoadedManagerHourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Restaurants restaurants, restaurants | 1,200 | 1,200 | Fourth's Chili's case study: 1,200 restaurants. The source does not say company operated. |
| General manager hours saved on forecasting and scheduling, per restaurant per week gmHoursSavedPerWeek, hours per restaurant per week | 0.5 | 0.5 | Chili's, via Fourth: GMs saved 30 minutes a week on forecasting, or 600 labor hours a week across 1,200 restaurants. |
| Fully loaded cost of a general manager hour fullyLoadedManagerHourlyCost, USD per hour | 30 | 45 | Editorial assumption for a US restaurant general manager, fully loaded (base pay, payroll taxes and benefits). Replace with your own figure. |
What it leaves out: This values only the manager's own time saved on the forecasting task itself, as reported by Chili's. It leaves out any labor cost saved from more accurate hourly scheduling, any reduction in food waste from better prep forecasts, and the cost of the forecasting software and integration.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Chili's Grill & Bar (Brinker International)
United States · Travel and hospitality · 2024
Brinker International's Chili's Grill & Bar turned to workforce management vendor Fourth for AI demand forecasting after Chili's own general managers found manual forecasting time consuming and, following a spike in turnover after the Covid-19 pandemic, newer managers struggled with inconsistent and inaccurate forecasting across its 1,200 restaurants. Fourth's forecasting draws on more than 20 years of Chili's own HotSchedules data. Brinker International's VP of Asset Management, Jason Noorian, is quoted in Fourth's case study calling the accurate forecast "foundational" to getting the right team and guest experience.
- Hours saved: 600 hours, per week, nationally, across 1,200 restaurants
"Saved GM's 30 mins / week or 600 labor hours per week nationally"
Claimed by: vendor
Thai Leisure Group
United Kingdom · Travel and hospitality · 2024
Thai Leisure Group, the UK restaurant group behind the Chaophraya and Thaikhun Thai restaurant brands, moved from Excel based manual scheduling across its 16 restaurants to Fourth's workforce management platform. Fourth's case study describes the deployment as AI driven Revenue Based Scheduling that removes manual assumptions from forecasting and building schedules. Richard Simpson, the group's Director of Operations for Chaophraya, is quoted in Fourth's case study saying the system "always learns" as it gets more information about the business.
- Hours saved: at least 56,000 hours, in the 12 months reported, across the group's 16 restaurants
"22% reduction in over scheduling equating to 56000+ hours in 12 months."
Claimed by: vendor - Revenue uplift: 15.7%, period not stated
"Proven technology from Fourth allowed managers to stay front of house and focus on delivering excellent customer service and great food, resulting in a 15.7% sales increase."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- At least one to two years of clean, per daypart point of sale sales history per restaurant
- A calendar of known local events, promotions and menu changes per restaurant
- Actual staffing and prep outcomes to check the forecast against and keep improving it
Systems to integrate
- Point of sale system, for historical and near real time sales data
- Labor scheduling and time and attendance system
- Inventory or prep management system, if the same forecast drives prep quantities
Complexity: Medium
The forecasting model itself is a well understood problem if the restaurant already has clean point of sale history; the real work is integrating that forecast into the scheduling and prep workflow managers actually use, and building trust so managers do not simply override it.
- 1
Start with sales and traffic, not the full schedule
Get the forecast itself accurate and trusted at the daypart level before automating the schedule or prep list end to end.
- 2
Keep the manager as the final approver
Publish a proposed schedule and prep list from the forecast, but require a manager to review and approve it, at least until forecast accuracy is proven at that specific restaurant.
- 3
Track forecast accuracy against actuals, per restaurant
A chain wide average hides restaurants where the model is systematically wrong. Track and review accuracy restaurant by restaurant, not only as a company average.
- 4
Feed local knowledge back in
Give managers an easy way to flag a known local event or closure the model would not know about, and feed that back into future forecasts for that location.
- 5
Measure manager time, not just accuracy
Fourth's Chili's results report minutes saved per manager per week alongside an accuracy improvement; instrument both, because a more accurate forecast that still takes as long to use has not solved the real problem.
Guardrails
- Labor law constraints (minimum rest, predictive scheduling rules, minor work rules) enforced in the schedule the forecast feeds, not left to the model
- A manager must review and approve every published schedule, at least during rollout
- Forecast accuracy monitored per restaurant, with an alert when a location's error grows
- No prep quantity automatically increased without a person able to override it for a known local event
KPIs to instrument
- Forecast accuracy against actual sales, per restaurant and per daypart
- Manager hours spent on forecasting and scheduling, before and after
- Labor cost as a percentage of sales, before and after, on comparable trading weeks
- Food waste or stock outs tied to prep quantities driven by the forecast
Human in the loop
The general manager reviews the proposed schedule and prep list before it is published or acted on, can flag a local event or anomaly the model would not otherwise know about, and remains accountable for the shift, not the forecast.
Common failure modes
- A chain average that hides bad forecasts at specific restaurants
- Company wide accuracy improves while a handful of locations get worse, unnoticed. Review accuracy restaurant by restaurant, not only as an aggregate.
- Local knowledge the model cannot see
- A road closure, a new competitor, a one off local event, none of these show up in sales history until it is too late. Give managers a fast way to override or flag it.
- Managers who stop reviewing and just approve
- Once trust builds, review can become a rubber stamp, so a data error or a bad input event goes live unnoticed. Sample a share of published schedules for a genuine second look.
- Optimizing labor cost at the expense of service
- A model tuned only to minimize labor cost can leave a restaurant short staffed against genuine demand spikes. Track service speed and satisfaction alongside labor cost, not labor cost alone.
What are the risks and rules?
EU AI Act
Depends on design
Annex III point 4(b) covers AI systems used to make decisions affecting the terms of a work relationship, its promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of workers. Working hours and shift allocation are terms of that relationship, so shift scheduling is the limb this reaches most directly. Classification turns on the system's intended purpose, not on whether a human signs off: Article 14 requires human oversight for systems that are already high risk, it does not exempt a system from the tier. A system whose only purpose is to forecast aggregate, restaurant level demand, with no output about any named worker, sits outside point 4(b). The same forecast, used by a system that allocates shifts to named workers or sets their hours, falls inside it regardless of any review step; staying out of that tier depends on the system's purpose, or on an Article 6(3) derogation, not on manager sign off.
Rules that apply
Controls to put in place
- A named manager reviews and approves every published schedule, not an automatic publish
- Forecast accuracy tracked and reviewed per restaurant, not only company wide
- Labor law rules (rest periods, predictive scheduling, minor work rules) enforced separately from the forecast, not inferred by the model
Frequently asked questions
- How much can AI forecasting improve restaurant scheduling accuracy?
- Reported results vary by chain and by what is measured. Fourth reports that Chili's improved forecasting accuracy by 20% and saved general managers 30 minutes a week, or 600 labor hours a week across 1,200 restaurants; Fourth separately reports a 7% increase in scheduling accuracy and a 22% reduction in over scheduled hours at Thai Leisure Group, the UK operator of Chaophraya and Thaikhun. Both are vendor reported figures for named customers, not independently audited.
- Does the AI decide the schedule, or does a manager?
- Neither Fourth case study on this page says who publishes the final schedule at that specific chain; that detail is not disclosed. As a general pattern, and what this page's implementation playbook recommends, the software proposes a forecast and a draft schedule for a manager to review before it goes live. Whether a human signs off does not by itself change the EU AI Act tier: a system that only forecasts aggregate demand sits outside Annex III 4(b), while one that goes on to allocate shifts to named workers or set their hours can fall inside it either way.
- Does the same forecast drive food prep as well as labor?
- It can, since both come from the same underlying demand forecast per daypart. None of the named deployments on this page discloses a food waste or stock out result specifically, only labor time and scheduling accuracy.
- What data does a restaurant need before starting?
- Clean, per daypart point of sale history, ideally a year or more, plus a calendar of known local events and promotions. Fourth's Chili's case study cites over 20 years of Chili's own HotSchedules data as part of what it used to build the forecast.
How to cite this page
Blits.ai AI Use Case Library, "AI demand forecasting for restaurant labor and prep scheduling", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/restaurant-demand-and-labor-forecasting. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 29 September 2026: First published