Ryan ONeill | Order up! How do I deliver it? Build on-demand logistics apps | Pydata Global 2023

Python and OR-tools have become the de facto standard for optimization and logistics modeling

Three core components for on-demand delivery systems:

• Forecasting (predicting demand)
• Scheduling (matching supply/workers to demand)
• Route planning (optimizing delivery sequences)

LAD (Least Absolute Deviations) regression is preferred over least squares for forecasting because:

• More robust to outliers
• Can be customized through optimization modeling
• Implemented as a linear programming problem

Scheduling optimization aims to:

• Balance overstaffing and understaffing penalties
• Account for worker availability constraints
• Match economic targets like orders per driver hour
• Adapt to different time periods (morning/evening/etc.)

Route planning considerations:

• Vehicle capacity constraints
• Depot locations and delivery points
• Drive time and distance optimization
• Assignment and sequencing of stops

OR-tools provides multiple paradigms in one library:

• Linear programming
• Mixed integer programming
• Constraint programming
• Satisfiability solving
• Local search capabilities

Moving from manual/siloed processes to integrated technology platforms improves operational efficiency

Models can be customized based on business needs:

• Different penalties for over/undersupply
• Seasonal variations
• Economic conditions
• Geographic considerations