Role of Data in Solving Capacity and Efficiency Problems in Real-time Logistics
In the world of real-time Logistics, every minute counts. It shows up not just as customer experience (SLA-Compliance & NPS), but also as the overall ability to accept N orders (Capacity) and deliver them in optimal time using minimum resources (efficiency). Such systems need to react fast to on-ground changes such as traffic, weather, availability of delivery-executives and their proximity to demand areas, availability of shipment at the source (in Swiggy’s case, prepared food at restaurants) etc. Both, real-time and historical data play a significant role in choosing “what” to optimize for and “how”. In this talk we discuss some of these challenges at Swiggy, the nature of historical and real-time data, and our journey of using these inputs in designing optimal solutions for Capacity (when exactly should we stop accepting orders) and Efficiency (orders/delivery-executive/unit-time). We also discuss challenges with data accuracy, high-variance and the necessary trade-offs in designing an optimial system.
- Introduction and Context
- The Capacity Problem - what is it; why it is important?
- The Efficiency Problem - what, why and the necessary trade-offs
- Data and its Nature
- Challenges with Accurate Data Capture
- Challenges with high Variance
- Real-time Vs. historical data
- Representing Capacity
- Aggregated capacity (Zone-level)
- Point-in-time-capacity (Order-level)
- Journey and Results: Solving for Capacity
- Efficiency Levers
- Predictions and accounting for errors
- Optimal Assignment
- Aggregate Analysis Vs. Specific Analysis
- Pitfalls of Aggregate Analysis
Piyush works as Director Engineering, Delivery team, Swiggy. He is responsible for building tech platforms for real-time logistics. Prior to this he worked with Last Mile logistics division in Flipkart where he led the development of the Mapping solution and helped deliver last mile optimal-routing solutions.