top of page

AI Case Study

Equinix predicts customer churn with 90% accuracy using a machine learning neural network model

Equinix's data science team have created a model which analyses different data for customer characteristics and provides the sales team with customers' risk for churn with a 90% accuracy, along with possible reasons which can then be acted on.



Internet Services Consumer

Project Overview

Equinix data scientists created a model which "runs about once every two weeks and provides and spits out the probability that each customer account might churn, along with potential reasons why. That information is fed into the company’s Inc. system so sales reps can see it while working with Equinix’s roughly 9,800 customers... alongside the risk score, a table shows sales reps reasons that might explain the risk, such as a decrease in power consumption or an extended period without using the customer portal".

Reported Results

For the first model iteration "it predicted churn with 50% to 60% accuracy when compared to actual churn data. After months of testing and refining, it now is close to 90% accuracy."


In-house; "Using a deep neural network, the team built a model that predicts the likelihood of customer churn over a 30-, 60- or 90-day period and says whether each customer is a high, medium or low churn risk."



Marketing Research Planning


"Equinix Inc. is turning to artificial intelligence to help sales representatives spot at-risk customers and intervene before they decide to leave. It’s part of a broader effort at the data center and colocation provider to inject analytics deeper into its business processes. The company hopes to rely less on analyzing data after events occur and instead give employees additional context."



"roughly 70 attributes that could contribute to customer churn, such as how many times a customer logs into a customer portal or placed an order over a certain time period, or whether power consumption increased or decreased"

bottom of page