Transforming Workforce Scheduling with AI
Helping hospitality operators save millions in labor costs while improving service quality through intelligent, data-driven scheduling.
Impact
€5-10M
Projected cost of labour savings in 2025
1
Enterprise customer signed during beta
TL;DR - The Solution
We designed an AI-powered scheduling system that surfaces critical forecasting data and provides intelligent recommendations—while giving General Managers full control and transparency to build trust in the system.
The Challenge
Labor represents the single biggest cost in hospitality operations, often consuming up to 40% of total revenue. With rising wages and chronically high turnover rates, operators face mounting pressure to protect their margins while maintaining the service quality their guests expect.
The Root Cause
A major driver of this cost inefficiency is poor scheduling by General Managers. Without accurate visibility into peak demand periods, staff are routinely over-scheduled during quiet hours and understaffed when business is busy—simultaneously wasting payroll and degrading the guest experience.
This wasn't just a hypothesis. Our data showed clear patterns of misalignment between staffing levels and actual demand, representing millions in potential savings across our customer base.
Research Approach
To build something truly valuable, we needed deep insight into how scheduling actually happens on the ground. I led a comprehensive discovery process combining multiple research methods.
Aligned with internal teams to understand business objectives, identify gaps, and surface existing pain points in our scheduling tools.
Spoke directly with General Managers to understand their needs, real-world constraints, and the criteria they use when building cost-efficient schedules.
What We Discovered
The current scheduling experience had several critical gaps:
- Usability Issues: GMs found the scheduling process extremely time-consuming, often taking hours to build a single week's schedule
- Trust Problems: GMs were skeptical of algorithm recommendations, believing they could make better decisions manually
GMs managing schedules with complex spreadsheets, indicating a clear market need for better scheduling tools.
Limited visibility into key metrics - forecasted sales data existed but wasn't surfaced in an actionable, user-friendly way.
Understanding how GMs interact with scheduling tools helped identify pain points and opportunities for improvement.
Key Insight
A critical question emerged from our workshops: “Businesses have budgets—expected sales per week with targets. How can we use this data to inform what we're recommending?” This became the foundation of our AI-powered approach.
Exploration & Ideation
With clear problem areas identified, I began rapidly exploring solutions through low-fidelity sketches. The goal was to test multiple concepts quickly and align the team behind a direction without getting bogged down in visual design debates.
Created rapid prototypes to explore user flows, test interaction patterns, and align stakeholders behind decisions without lengthy UI debates.
Detailed hourly view with optimal staffing recommendations and cost metrics
First iteration of AI recommendation interface
These sketches explored several key user flows:
- Creating schedules 2 weeks in advance
- Enabling smart scheduling features
- How AI suggestions would be presented
- Workflows for reviewing and editing recommendations
- Quick access to detailed daily breakdowns
Defining Scope
Based on research insights and technical constraints, I worked with the team to clearly define what we would tackle in this POC phase.
In Scope
- Visibility: Sales and order forecast metrics prominently displayed
- Visibility: Daily view with detailed breakdown drawer
- Usability & Compliance: Auto-assigning shifts based on AI algorithm
- Constraints: Define parameters and rules for the AI
- Trust: Modal explaining what's happening behind the scenes
- Control: Schedules created in draft mode—users can fully edit before publishing
Out of Scope (Future Iterations)
- Color coding departments for visual clarity
- Replacing legacy design system components
- Surfacing and editing custom scheduling rules
- Conversational AI agent for schedule adjustments
The Solution
The final design introduced AI-powered scheduling recommendations while maintaining the control and transparency GMs needed to trust the system.
Weekly View: Enhanced Visibility
The updated weekly view surfaces critical forecasting data directly in the scheduling interface. GMs can now see projected sales and orders for each day, making it immediately clear where staffing should be prioritized. It also breaks down people by department, a key insight from customer feedback that enables managers to ensure balanced coverage across all operational areas.
Daily Breakdown: Granular Control
When GMs need more detail, they can access a comprehensive daily view that shows:
- Hourly staffing recommendations by department (FOH, BOH, Management)
- Visual graphs comparing current staffing to optimal levels
- Real-time cost of labor calculations
- Clear indicators when over or understaffed
- Actionable suggestions for adjustments
Building Trust in AI
Recognizing the trust gap we discovered in research, we implemented several features specifically to build confidence:
Draft Mode
AI-generated schedules start in draft, giving GMs full control to review and edit before publishing.
Transparency Modal
Clear explanations of how the AI makes recommendations and what data it's using.
Visible Reasoning
Each suggestion shows the underlying logic—peak hours, sales forecasts, and labor targets.
Design Principle
Rather than replacing GM expertise, the AI acts as an intelligent assistant—surfacing insights and recommendations that GMs can accept, modify, or override based on their on-the-ground knowledge.
Impact & Results
The POC phase validated both the technical feasibility and market demand for AI-powered scheduling.
Business Outcomes
- Signed a major enterprise customer based on the AI scheduling beta
- Projected to save €5-10M in cost of labor across their operations
- Validated product-market fit for intelligent workforce management
Product Learnings
- GMs need visibility and control, not black-box automation
- Incremental trust-building is essential for AI adoption
- Draft mode gave users confidence to explore AI recommendations
Next Steps
Based on the success of the POC, the roadmap includes:
- Rolling out color coding and updated design system components
- Enabling custom rule creation and editing within the interface
- Exploring conversational AI for natural language schedule adjustments
- Expanding to additional enterprise customers
Reflection
This project reinforced several important lessons about designing AI-powered products:
Research drives impact. The insights from GMs about trust and control fundamentally shaped the solution. Without understanding their skepticism toward algorithms, we might have built a fully automated system that users would reject.
Start with lo-fi. Rapid sketching allowed us to explore multiple approaches and align stakeholders quickly. This speed was critical in a POC phase where we needed to validate concepts before significant engineering investment.
Transparency builds trust. By showing our work—explaining AI recommendations, providing draft mode, and surfacing the underlying data—we transformed a potentially threatening automation into a valued assistant.