Inventory Management
Background
A Barcelona-based restaurant specializing in Argentine empanadas needed to optimize their stock management system. Their existing approach relied on maintaining fixed pre-cooked inventory levels based on historical sales, leading to two critical issues:
- Stock shortages affecting customer service
- Product waste due to expiration
The business owners, lacking gastronomy experience, wanted to implement a data-driven strategy for inventory management.
Problem Statement
Current Strategy
The existing inventory management relied on a simple minimum/maximum stock level approach. The kitchen staff would reorder products when minimum stock levels were reached, without considering sales forecasts or seasonal variations. This basic system, while simple to implement, failed to address the complexities of restaurant operations.
Key Issues Identified
The current implementation of the Reorder Point (ROP) formula(*), revealed significant limitations in practice. The formula's effectiveness was compromised by including waste in the output calculations and failing to account for lost sales during stock-outs. These data quality issues created a cycle of inaccurate ordering that perpetuated both waste and shortages.
(*) ROP = (Average daily output × lead time) + safety stock
Furthermore, the high volatility in daily sales patterns made historical averages an unreliable predictor of future needs. The existing stock management strategy showed no alignment with actual sales patterns and business cycles, leading to systematic inefficiencies.
Data Exploration
Methodology
The analysis focused on daily closing data, examining the relationship between dates and the number of empanadas sold. This simple dataset revealed complex patterns that had been previously overlooked in the restaurant's planning process.
The analysis revealed clear sales patterns influenced by both weekly and monthly cycles. Weekly trends showed consistent peak sales on Fridays and Saturdays, while monthly patterns demonstrated significant correlation with local salary payment dates. These findings contradicted the initial assumption of volatile, unpredictable sales.
Recommendations
Forecasting Model
The proposed solution centers on a machine learning application that integrates with the restaurant's sales software to provide next-day inventory recommendations. This application will suggest Reorder Points (ROP) daily, analyzing each day's sales data to optimize inventory levels for the following business days.
The initial model will process historical sales data, accounting for daily and monthly patterns while excluding waste-related distortions. It will be designed to be extensible, allowing for the future incorporation of additional variables that can impact sales patterns, such as weather forecasts and holiday schedules.
At the end of each business day, the application will:
1. Process that day's sales data
2. Update its predictive models
3. Generate inventory recommendations for the next days
4. Account for lead times in the supply chain, with the option to be manually overrided for exceptional circumstances (holidays, transportation strikes, weather events, etc)
The model's initial implementation will focus on:
1. Processing daily sales patterns
2. Accounting for weekly trends
3. Incorporating monthly salary payment cycles
4. Excluding waste-related data distortions
Future enhancements will enable the integration of:
1. Weather forecast data
2. Local holiday schedules
3. Special events calendars
4. Seasonal trends
This machine learning approach represents a significant advancement over the current static ROP formula, offering the restaurant a sophisticated yet user-friendly tool for day-ahead inventory planning that grows more accurate over time.
Next Steps
The immediate priority is validating the forecasting model through real-world testing.
Once the base model is validated and approved by the customer, we will begin expanding the system's capabilities by incorporating additional data sources:
Government Information Integration
The system will connect to official channels to automatically capture data about:
- Public holidays and celebrations
- Local festivities
- Scheduled public works that might affect foot traffic
- Other civic events impacting restaurant operations
External Data Sources
We will enhance prediction accuracy by incorporating:
- Weather forecasts through reliable API services
- Local event calendars
- Public transportation schedules
- Tourism data when available
This phased approach ensures a solid foundation before adding complexity, while the modular design allows for seamless integration of new data sources as they become available or relevant. Throughout this process, we will maintain open communication with the restaurant team to prioritize the most impactful enhancements based on their operational experience.