Telecom GIS In Five Minutes Video Series - Episode 025 - Predicting Churn with Spatial Analysis

09-07-2021 11:03 AM
Esri Regular Contributor

Hello once again and welcome to this week's episode of TGI5. In the episode today, we are going to take a look at a workflow around Churn prediction leveraging some tools within ArcGIS and spatial analysis to help us understand what customers may be likely to churn. So without further ado, let's jump into the demo.

We're going to take a look at a newer tool in ArcGIS for assisting with Churn prediction. Now, ArcGIS offers hundreds of spatial and statistical tools, including a lot of those newer machine learning and prediction tool set. One of these tools is the Forest based classification and regression tool. So here's our scenario. I have a Customer 360 database with just over a million data points that were provided to us by every partner share tracker with some additional variables that we've generated internally, and we can see all those locations shown here in blue.

Now this heat map on top shows us those customers that have cancel their service and turned within the last year. And what I want to better understand is why. And if we can understand why they left, then we can start to locate other current customers who have similar Churn characteristics and put together some marketing campaigns to influence their decision to stay well before they churn. So using this Customer 360 database, as well as the Forest based classification and regression tool, we ran this machine learning tool against a variety of different variables to see how they impact or influence the turning of customers.

So some of those variables that we tested against were things like years as a customer, their total bill, or how much they spend on telecom services. Are there any competitive threats in the area? Do they have bundled services? How many devices at home do they have connected at once? What are the total number of reported technical issues? And we also took this data and we geoenriched it with some Esri demographic data. So we could also add additional demographic variables like the tapestry segmentation. And if we wanted to add an additional geographic variable like distance to central office, we could have also added that as well.

Now, the results of the analysis produce a trained data set that lets us know all the different variables that influence Churn and how much they influenced Churn. So it produces a level of accuracy or a level of confidence that these variables actually did influence a customer to leave our organization. Now let's take a look at these results within a chart. So the top factors or variables that influence Churn were the total bill of a customer or how much they were spending the dominant Tapestry segmentation and how many times they reported technical issues.

Now, if we like the level of confidence that the analysis had in these variables, we can then take this train data set and run it against our current customers in our Customer 360 database. This would then produce a result with a predicted probability that a customer is about to churn. Now, based on these analysis results, let's say we've implemented some campaigns to target those customers with a high probability of churning, and we want to monitor the impact of those campaigns. So what we've done is we have put together a dashboard to allow executive teams start to see how these campaigns are influencing Churn and look at our churn numbers month by month or year over year, so we can see for our entire service footprint how many customers have left this year versus last year, and then get a breakdown of how many of those customers disconnected their broadband or TV service. I can also drill down into a particular area and start to see the churn numbers for a region or a territory.

Thank you all for watching the TGI5 episode this week. I hope this was insightful and you saw some ways to use the forest-based classification and spatial analysis tools within ArcGIS to help your organization understand churn and predict likelihood that a customer will cut their service.

And until next time we're out.

-Patrick Huls

Solutions Engineer | @SpatialNinja | Telecommunications
0 Kudos
0 Replies