Soluto by asurion
Published in

Soluto by asurion

Churners gonna churn — how we filtered out guaranteed churners from our data

To fight churn, we had to understand churn. If you’re trying to reduce customer attrition in your digital products, you’ll probably find our story extremely enlightening

Churners gonna churn — how we filtered out guaranteed churners from our data

Asurion is probably the world’s largest tech-care provider, and Home+ is one of its more popular offerings. Home+ is a subscription-based plan which provides protection and support to all connected home devices, and covers smart TVs, smart home hubs, gaming consoles, laptops, printers, and more. Within Home+, my team created the digital product experience that provides customers with clarity about their plan and access to its features and benefits.

One of my team’s main KPIs was a reduction in 7 days and 30 days churn — lowering the number of customers who cancel the subscription within the first 7 or 30 days after enrollment. We planned to impact these numbers by giving subscribers added value through the features in the online experience.

Churners mess up the data

There are several behaviors that can be counted as interactions with the digital product — signing into the account, exploring features and benefits, contacting support via chat, and more. What I wanted to find out was whether a specific type of interaction can be associated with a lower churn rate.

However, I first found that there’s a sub-population of our customers that only interact with the digital product in order to churn. We called them “guaranteed churners” — users who changed their mind after enrolling, and sign into the digital experience with the sole intent to churn.

Guaranteed churners mess up piles of other wise perfectly good data. In their search for a cancel button, they click on other features and experiences, and add a lot of noise to the data, which impairs our ability to identify churn-reducing features. I had to find a way to identify guaranteed churners, and exclude them from the analysis.

How we identified guaranteed churners

My goal was to identify guaranteed churners with a precision of 70% or higher, based on internal business guidelines. My hypothesis was that guaranteed churners have two main behavioral characteristics:

  1. They will not explore the digital product in-depth, but rather look for a way to cancel the subscription as soon as possible. This will translate to a short time spent on the website.
  2. At the time we didn’t have an “unsubscribe” button, so customers who wanted to cancel their subscription would turn to the expert support chat. That means their first chat with an Asurion support expert will include a request to churn.

To prep my data, I filtered:

  • Customers that entered the digital experience within 7 days of enrollment
  • Who on their first visit quickly started a chat with an Asurion expert, with a request to churn

What does starting a chat “quickly” mean? I decided to examine different timeframes for time-to-chat: 0–60 seconds, 0–90 seconds, and 0–120 seconds.

In order to identify chat sessions with requests to churn, I created a query that searched for keywords associated with churn. I created a data set with a few hundred chat sessions, labeled them, and got a precision of ~75%. I then refined the query, repeated the process, and got to a precision of about 90%.

Customer Journey

The data showed that the parameters I chose were indeed connected to high churn rates. I chose to go forward with the 90-seconds variant, as it balanced accuracy and coverage, in comparison to the other variants.

Utilizing the new segment

Once I had the logic for singling out the guaranteed churners, I could filter that segment out of my analysis. With the freshly cleaned data I was finally able to demonstrate a significant churn reduction for customers who interacted with certain digital features.

Guaranteed churners are not a large portion of the enrolled population, but because they have to interact with the digital features in order to churn, they had a significant effect on our analysis. Consequently, the filter we created was a valuable asset for our product managers who could now easily filter out guaranteed churners for any given use case.

When performing an analysis there are often variables we cannot always control or account for, and that is why it’s important to analyze and identify confounds that might affect the results. Accounting for guaranteed churners with this segmentation enabled us to have more accurate analysis and make better data-driven decisions.



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store