The market of ISPs is highly competitive. Optimizing the Cost of Customer Acquisition is key to most players. Of course digital marketing is heavily used by ISPs to gain traction and momentum.
However, the cheapest and most effective means of acquisition is “word of mouth”. An ISP who keeps track of numerous data about its customers wants to understand which criteria contribute to a customer making a recommendation.
The goal is to “predict” which customers are going to recommend the ISP and to understand what criteria are key in impacting this recommendation.
The dataset used contains 4369 customers’ records. And for each record 123 information.
Some of the typical information collected are:
So far, customer recommendations have been evaluated but not analysed nor anticipated.
TADA’s Machine Learning approach can help predict and understand what aspects of the service makes a customer willing to provide a recommendation.
It poses the following customer relationship question:
Can customer satisfaction be engineered and optimised?
The TADA predictive models’ are accurate in predicting customer recommendation in 79% of cases (R2 of 79%, a MAPE of 15%).
This means that in most cases, customer recommendation was well predicted by the model.
What is of high value to the ISP is why does a customer make a recommendation? What are the key criteria influencing this recommendation?
TADA has selected the following six main criteria out of the 123 available in the dataset:
Moreover, the what_if feature (Live Predict) has shown to this ISP that the quality of the service provider who installs the internet access is relevant, up to a certain threshold. Beyond a customer satisfaction of 60/100, it does not make a difference anymore for the customer to recommend this ISP. Since this ISP was working solely with a single service provider in order to secure a continuous level of quality well above 60, it became clear that he could diversify its pool of service providers. This in turn allows for the scaling of its operations and the scaling of this ISP’s revenue.
No need to be a skilled statistician to find the key impacting factors among the 47 types of information collected.