Recommending an alternate delivery model for Amazon to cut down on shipping costs by minimizing the the distances travelled by its delivery agents. The concept allows anyone with a car and a valid driving license to work as Amazon's delivery agent, similar to Uber for taxi. The model selects the optimal delivery agent(s), assigns them the shipments and provides them optimized delivery routes. The compensation is decided by customer ratings and efficiency factors in a feedback loop.
Optimization strategy for Amazon's Uber like delivery service
1. An Optimization Strategy for Amazon’s Uber-like Delivery
Service
Arsalan Qadri, Hanzhuo (Andrew) Li, Tao Feng
Instructor: Dr. Ted Stohr
Introduction Methods and Algorithms
1.Use Logistic Regression to select a set of suitable delivery agents for each
shipping task.
Predictive Variable: P - The probability successfully completing the delivery task.
Independent variables:
X1 – Distance from agent’s current location to seller site.
X2 – Agent’s reliability.
X3 – Agents punctuality.
X4 – Agents customer rating.
2. Calculate the distance between available delivery agents selected from
above and the corresponding sellers using GPS coordinates. Use Assignment
Model to assign one or more (closest) agents to each seller’s shipment task.
Assignment Model for minimizing the total distance from sellers and agents.
3. Find the shortest routes for the shipments from a seller to one or more
buyers, using Genetic Algorithm.
This model is aimed at minimizing the total distance (Z) travelled by one or more agents
involved in a single shipment task. The constraints for each agent in this model includes
capacity (Q) and the number of shipping sites (L).
(This model is adapted from: Study of the optimizing of physical distribution routing problem based on genetic algorithm by LANG Mao-
Xiang. We adapted this model by changing vehicle capacity, route description and other variable meaning of prior model)
4. Use the weights from AHP method to calculate the rating of each vehicles
for their compensation ratio
The compensation for the drivers is calculate using the agent rating from the AHP model.
This is done on the basis of customer ratings and efficiency factors. e and customer ratings
on factors such as reliability, punctuality and courtesy, the compensation for a driver is
calculated.
Business Intelligence & Analytics
http://www.stevens.edu/howe/academics/graduate/business-intelligence-analytics
Optimization Methods
“Flex, Amazon’s new on-demand delivery service, promises to get your
packages to you even sooner by hiring independent drivers to bring them to
your house” (Washington Post, Sep 30th
, 2015).
The optimization model for Amazon’s Uber like delivery service minimizes
the delivery costs by allowing anyone with a car to work as a delivery
agent. After the necessary background checks, agents are assigned ratings
in aspects such as reliability, punctuality and behavior. Every time a
shipment is completed the customer rates the delivery agent, these ratings
become the basis for deciding the compensation of the delivery agent.
The model is primarily based on minimizing the distance travelled by the
delivery agent to make the product deliveries. This is coupled with the cost
of delivery per mile that is derived from the customer ratings and quality of
service records.
The model selects the closest delivery agents to the seller on the basis of
the former’s GPS coordinates, checks their availability and introduces them
into the model for further processing.
A seller with multiple shipments can have one or more than one delivery
agents allocated, to ship orders to multiple buyers.
Run every
½ hour
Run every
delivery
task
Run every
½ hour
Run every
½ hour
i: The buyer (shipping site)
K: The number of delivery agents involved in the task.
Z : Total distance travelled by all agents involved in a delivery task.
Qk:Capacity of the agent
Qr ki:The weight of goods to be delivered to site i, via agent k’s delivery route.
dij:Distance between buyer i and buyer j.
doi:Distance between seller o and first shipping site j.
nk: Agent k needs to ship to nk buyers in a trip.
L: Number of buyers in the shipping task.
Rk: Agent k’s route.
Rki: The position of shipping site i in agent k’s route.
Sign(nk) : When agent k has more than one buyer to deliver to, it shows that Sign(nk)=1
and otherwise Sign(nk)=0.
Formula to calculate compensation
C=(K+C1M+C2H)R
C: Compensation for agent per task.
K : Fixed compensation
C1:Compensation rate for total
distance in the task.
C2:Compenstation rate for driving
time in the task.
H: Driving time (minutes)
R: Agent rating