Separate three operations
Restaurant service, a coffee rush and bakery production create different coverage and handoff needs. Bakery supply also supports other locations.
CASE 04 / Workflow design
AI-assisted restaurant scheduling
Demand, service coverage and manager judgment before automation.
DELIVERED
Used demand history and manager context to support planning, and prepared requirements, a baseline and evaluation criteria for a separate scheduling pilot.
EVIDENCE LIMIT
No verified automated schedule rollout, manager acceptance rate, labor savings or forecast-accuracy improvement. The current scheduling process remains authoritative.
The schedule has to work on the floor.
Recent sales can look similar while the shifts feel completely different. A home game, a private event or the mix of roles on the floor changes what we need. The coffee shop and bakery have their own needs, and bakery production supports other Spill the Beans locations too. A single labor target doesn’t tell the whole story.
What I owned.
I worked with managers to put the demand history next to their experience of the shift. I defined what an AI-assisted draft would need to account for and what a manager would need to check. We also prepared a separate pilot for evaluation; managers still own the actual staffing decisions.
Where I used judgment.
Keep planning and actuals distinct
Use a frozen comparison and identify missing history before treating a draft as a reliable baseline.
Hold the automation gate
Prepare evaluation criteria without replacing the authoritative schedule. Incomplete data is a reason to pause automation.
Who had to be in the loop.
I brought demand history, events and the differences between bar, coffee and bakery work into the evaluation. Managers retained staffing decisions; product/vendor access and complete historical records were dependencies for the separate pilot.
The manager knows what the sales total leaves out.
This describes existing manager-led planning and the separate evaluation gate for a scheduling pilot.
- Inputs
- Sales history, event demand, manager shift notes and different bar, coffee and bakery needs.
- Manager decision
- Use context about guest flow and coverage alongside the numbers. Keep actual results separate from an expected staffing plan.
- Separate pilot
- Requirements and evaluation criteria prepared. Generated output, acceptance and operating savings remain unproven.
What was actually delivered.
- Manager-led planning that combines demand and operating context.
- Separate pilot requirements, baseline and evaluation criteria.
- A readiness boundary: setup is complete only as setup; schedule acceptance and savings remain unproven.
What remains to demonstrate.
Evaluate a generated draft against a frozen baseline, record manager changes and measure review effort alongside service coverage.
Managers carry a lot of useful history in their heads. I want to get that knowledge into the planning conversation before I ask a tool to make a recommendation.