Advanced analytics to monitor and improve delivery time in e-commerce

MOTIVATION FOR CHANGE

A pure e-commerce player has experienced two digits growth for years, reaching tens of thousands of orders per day. Delivery time has been set to a few days, and sometimes hours, to meet market expectations and to beat competitors, providing a best in class service to customers. A large network of carriers has been established to handle this massive flow of goods distributed throughout the country.

Transit time worsened causing customer dissatisfaction until carriers were unable to adequately handle the flow of orders to the regions, although volumes were within agreed contracts.

The causes of service decline were diverse. They ranged from the shortage of drivers to handle all deliveries to the unplanned sharing of carrier capacity with their clients reducing the ability to handle the contracted volume. The carriers’ poor performance was unacceptable and triggered a series of actions to prevent service decline.

ANALYTICS TO DETECT NEAR FUTURE BEHAVIOR

One course of action was based on the fact that significant failures did not begin immediately but turned out to be a cumulative deterioration process whose signs could be detected some time in advance.

This project took advantage of this context to implement a monitoring system that anticipates near future behavior and drives control actions to avoid deterioration of the carriers' performance.

Designed to capture data of over tens of thousands of orders and their status throughout the day, the monitoring system continually builds an extensive database composed of time series of diverse measurements of system status such as delivery time variance, carrier´s buffer size of orders not collected, differential inbound and outbound flow of orders and several others.

This time series database represents the dynamic status of anticipatory indicators which allows the analytics engine behind the monitoring system to learn about the behavior of each carrier in each micro-region, clustering them into performance stages, such as "normal", “attention”, “near failure”, “failure”.

DRIVING ACTIONS TO IMPROVE SERVICE

Management is automatically informed on where to act and with which degree of urgency and this information drives control actions to mitigate the potential failure before critical stages are reached.

Control actions can be taken within hours and, according to performance stages, can vary from alert calls with carriers or even the transfer of new orders to alternative carriers in better conditions to service the same micro-region.

Longer lasting lower performances triggers another class of procedures aimed at maintaining the supplier base at the desired quality level.

The project enabled the company to offer a better service to clients in accordance with its commercial strategy mitigating a high-risk situation where the carrier market was not entirely ready to pair with the fast-growing e-commerce business.