Research has pointed out that, out of all leaving customers, only 9% left for a competitor. It is the indifference of employees and product quality that cause customers to leave. The saturation in telecommunication service goes in hand with higher expectations from customers. Furthermore, they have various providers to choose from and can easily switch to another should they find fit, resulting in the increasing number of customer attrition.
As such, most businesses find reducing this number to be crucial: as leaving customers means reducing profits, and on a higher scale, reducing the business’s reliability to customers.
Investing in attracting new customers and keeping old ones are thus both important in maintaining a stable development, and the ability to predict which customers are planning to leave will enable in-time actions from businesses. As a result, many companies are looking into this method.
In acknowledgment of this problem, Ms. Vo Thi Hong Phuong and her project team have researched and launched a Customer Attrition Predictive program for Internet providers. The project started in May 2017 and was officially launched on.
The Project’s experimental model
The program has been built through various stages, as illustrated below :
Similar to many scientific data problems, data preparation, and analysis, as well as feature engineering is extremely important, taking long periods of time throughout the process. Precise attention on data leverage techniques is necessary, seeing that the customer attrition rate of a business is generally low.
Experimental tests are implemented on different models during the training process. Best grouping results can be derived from accurate statistics evaluation, recall, F1-score, and XGBoost model.
A list of customers who are most likely to leave will be extracted and sent to the customer care department after a process of optimization and feature selection with a proper model. The end result (after customer care) shall then be recorded and later utilized for the optimization of the predictive model.
Certainly, the current system is still flawed in its inability to pinpoint exact reasons to the leaving and priority for maximum effectiveness, but this leaves an area for later improvements.
Knowing the right customer who leave the internet
Before the prediction analysis program, the system performs customer care service according to the subjectively extracted list, in large numbers, randomly selected, discrete localized, not comprehensive evaluation. When the program was launched, FPT Telecom promptly supported the problems that customers were experiencing. For example: technical errors, network quality, freight rates, service packages … Bring a better experience to customers, help customers feel more interested, prevent early intention to leave the network or make policies persuading customers to stay when they request to cancel, the analysis results also provide input for some other customer care programs in the system.
The system went into operation to increase the rate of knowing the right customers leaving FPT Internet network to 43.1% of 215,760 customers who were cared for and similarly leaving FPT Television network with 32.5% of 64,758 customers being taken care of.
The system by Ms. Vo Thi Hong Phuong and his colleagues have won the “Creative award of the Year” for 2018 and held a total of VND 120 million (including 70 million prizes from the Organizing Committee) and 50 million bonuses from the unit).
Vo Thi Hong Phuong – Thao NguyenRelated posts: