Serviceability under high demand
At Swiggy, our aim is to deliver orders to customers in a reasonable promised time regardless of when and where the order is placed. We are confronted with considerable challenges when faced with high (and sometimes unexpected) demand - think IPL weekend, rains, New Year’s Eve, competitor’s platform is down.
To deal with this we use an array of approaches, some of which are systematic and incorporated into the delivery system operational process, others that involve specific manual intervention strategies.
Our primary objective here is to ensure great customer experience and doing so with minimum business loss.
I will first describe the nature and scope of the problem and why it requires a multi-pronged effort to deal with it. I will show how ideas from time-series, operations research, machine learning and simulations come together for solving these problems. The topics covered will include:
(2)Characterizing stress of delivery system
(3)Real-time paring of demand
(4)Delivery leg predictions
(5)Order Queue Dynamics - Inflow and Outflow
(6)Batching of orders
(7)Supply side parameter control
I am a data scientist at Swiggy and have a background in theoretical physics (PhD. in statistical mechanics)and systems/mathematical biology (postdoctoral research). My interests lie in multi-disciplinary problems, especially those lying at the intersection of statistics, complex networks, algorithms and machine learning.