Deliveroo's Logistics and Load Management

Introduction to Deliveroo

• On-demand food delivery company
• Operates as a three-way marketplace:
  • Riders
  • Restaurants
  • Customers
• Partnered with tens of thousands of restaurants across 13 markets
• Utilizes a large delivery network of riders

Customer Order Flow

• Customer orders food via app
• Logistics Algorithms Team becomes involved
• Machine Learning Predictions:
  • Food preparation time
  • Rider travel time
  • Order assignment for optimal delivery efficiency

Order Assignment Algorithms

• Assign riders to orders
• Balances rider availability with optimal customer satisfaction
• Example: Central London Map
  • Orange knives and forks: Restaurants
  • Deliveroo logos: Available riders

Load Management

• Defining Load:
  • Relationship between number of orders and available riders
• Predictive Algorithms:
  • Forecast order volume
  • Predict rider attendance
  • Schedule rider shifts
• Load kicks in during unexpected high demand, e.g., bad weather or events
• Example scenarios:
  • Double whammy of bad weather affecting rider availability and increasing customer orders
  • Events causing a spike in simultaneous orders

Load Impact on Customer Experience

• Example graph depicting delivery network load throughout the day
• Load affects visible restaurant options based on real-time rider availability
• Algorithms adjust visible restaurant options to manage network load
  • Dynamic shrinking of the delivery area

Evolution of Load Measurement

• Early days: Ratio of orders to riders
• Current approach: Measure based on time to complete deliveries
  • More accurate reflection of network load and rider availability
• Example of how travel distance and delivery bundling affect load measures

Real-world Application of Load Measures

• Data showing correlation between predicted load and actual delivery delays
• Use metric to adjust visibility of restaurants in real-time

Managing Load in Practice

• Dynamic adjustments to the delivery area based on network capacity
• Strategies to handle high load:
  • Increase efficiency by prioritizing closer restaurants
  • Algorithm-driven adjustments using machine learning models
• Example: Comparison of rainy vs sunny days
  • Rain reduces rider turnout, increasing load
  • Algorithms adjust to maintain efficiency during high load

A/B Testing and Results

• Continuous experimentation and algorithm adjustments
• New algorithms have shown improvements in managing high load situations
  • Reduced delivery times and delays
  • Maintained or increased order acceptance during lower load periods

Future Work and Ongoing Projects

• Recommender systems and personalization
• Forecasting and scheduling optimization
• Dynamic pricing for optimal rider incentives
• Continuous algorithm development and experimentation

Conclusion

• Deliveroo uses complex algorithms and machine learning to balance order volume and rider availability
• Ongoing projects aim to further optimize delivery efficiency and customer satisfaction