Background Compass Coffee is a Washington, D.C.–based coffee roaster and café chain operating multiple locations across the region. They started in 2014, with the mission “to make peoples’ days better”. Unlike many other restaurants, all Compass’s ingredients are delivered fresh, including items other than coffee, such as bakeries and snacks. Project Description Project Opportunity Inventory decisions are currently driven largely by manual processes and store-level experience, with managers estimating required quantities based on recent sales and intuition. While this approach provides flexibility, it can lead to inconsistencies across locations and challenges in accurately anticipating demand. Variability in customer behavior, seasonality, and product composition further complicates the estimation of ingredient needs, making inventory planning a complex and error-prone task. This is an issue the team aimed to resolve. Project Vision At its core, the vision is to enable store managers to make informed ordering decisions based on forward-looking data insights rather than retrospective estimates. This includes translating item-level sales into ingredient-level demand and providing clear guidance on expected inventory needs. In the long term, such a system can serve as a foundation for more automated and optimized inventory management processes across locations. Project Outcomes The main project outcome is the introduction of a forecasting system that enables systematic estimation of ingredient-level inventory needs based on historical sales data. By transforming item-level transactions into underlying ingredient demand, the system provides a structured foundation for more accurate and consistent inventory planning. All components were integrated into a coherent forecasting pipeline, ensuring that data preprocessing, model training, evaluation, and prediction can be executed in a consistent and reproducible manner. Project Deliverables The project delivers an interactive dashboard that provides store managers with weekly forecasts of required inventory at the ingredient level. The dashboard presents forward-looking demand estimates in a clear and intuitive format, enabling straightforward use in day-to-day ordering decisions. To support the forecasting framework behind the dashboard, the project includes a fully developed forecasting pipeline, covering data preprocessing, model development, evaluation, and post-processing. This pipeline ensures that predictions can be generated consistently and updated as new data becomes available. All components are documented and structured to support future maintenance and extension, allowing subsequent users to adapt the system to additional data sources, modeling approaches, or operational requirements. Student Team Ziyan Wang served as a Data Scientist on the project, developing machine learning models for demand forecasting and leading model evaluation to align predictions with operational goals. She also coordinated team meetings and supported communication to ensure alignment between technical work and project objectives. Ziyan is currently a Master’s student in Statistical Science at Duke University.
Mentor: Dawn Strickland
Project poster (PDF)
