Why Your Churn Mitigation Strategy Might Be Too Little, Too Late


Ibby Syed

99.9% of businesses are in the business of selling something, and the person on the other end of the transaction is a customer. Now, different businesses have different numbers of customers: Delta flies hundreds of millions of customers around the world each year (I’m writing this on a Delta flight right now), while a company that creates missile defense systems might do so for  very few but usually very large countries over the course of a given decade.

Taken from Anu Hariharan’s YC talk - “Nine Business Models and the Metrics Investors Want”: https://www.ycombinator.com/library/8E-nine-business-models-and-the-metrics-investors-want-sus-2019

Most of these companies also experience what’s called “churn.” The term is most often attributed to companies that sell a subscription, but is also present in most other types of business, just with differing amounts of importance. For a subscription business like Netflix, churn is straightforward - it occurs when someone cancels their membership. However, a conventional retailer like GAP can experience churn when someone who buys a new pair of GAP jeans every August decides they’d quite like to give a pair of Bonobos chinos a go this year.

Customer churn refers to the natural business cycle of losing and acquiring customers. Every company — no matter the quality of its products or customer service — experiences churn. Generally speaking, the less churn you have, the more customers you keep.

(taken from here)

The Problem

As a data scientist, I’ve worked with tons of teams that are trying to mitigate subscription churn. There are entire software solutions that are built on “Winback,” which try to give folks who are at the point of churn an alternative solution. A great example of this is Audible’s “re-trial” with three months of discounted usage, or the gym’s “pause” feature that stops billing you for a few weeks while you’re away on vacation.

Old Audible winback modal - taken from here

In our experience, this approach doesn’t work that well. Usually, by the time a customer is ready to hit that cancel button, their mind is made up - they aren’t getting value from the product or service anymore, and it’s often not a cost issue. How many times have you signed up for a streaming service, watched a show, stopped using the streaming service, and not cancelled for months until you came across the charge on a credit card statement? 🙋🏻‍♂️ I have (this is literally the reason Truebill was created).

The real solution to this problem is identifying where the moment of lost interest was, as that is the pivot point companies need to take action on.

Implicit vs Explicit Customer Journey Metrics

There are lots of touchpoints in a customer journey that are very explicit and are shared across many different businesses. Some examples are:

  • First Touchpoint - the first time someone comes across your brand
  • Conversion - the point at which a customer purchases your product
  • Time to Conversion - the time between the First Touchpoint and Conversion
  • Churned - the point at which someone cancels a product or service
  • Re-won/Re-activated - the point at which someone returns to a product or service

However, there are also subjective points in the customer journey, which might be harder to pin down. A few of these include:

  • The “Aha” Moment - The point at which a customer realizes the value that you provide for them and are no longer “just testing the product.” They’re locked in.
  • Loss of Interest - Conversely to the “Aha” moment, this is when a customer stops utilizing your product and will, in some months, eventually stop paying for it.
  • Neutral to Advocate - This is when a customer goes from being a 7-8 on the NPS scale to a 9-10 - they’re becoming your champions. They’re actively advocating for your product with their friends and neighbors, who are also converting and getting good use from it.
  • Onboarding Failure Moment - This is when the onboarding process fails a customer. They haven’t reached their “Aha” moment but are more than likely going to either (A) churn out of their trial or (B) stop using the product altogether.

Of these, the explicit ones are easy to identify and measure. Most of the time, you can find statistics for all of them in your business intelligence tool (Looker, Hex, Tableau, Amplitude) and can therefore act on them. However, we’ve found that companies that identify and act on the “Loss of Interest” subjective moment rather than the “churned” objective one are more likely to keep their customers engaged, and that outreach campaigns based off of the subjective metric are exponentially more likely to yield in successful outcomes.

So how does one identify the “Loss of Interest” Moment? Let’s find out!

How to identify when a customer lost interest:

Some rules before we get started:

  • Filter for Homogeneity - this method works best if you look at data for similar customers. For instance, if you’re a product manager at Adobe Creative Cloud and you’re running this process, a customer who primarily uses Photoshop is very different from one using Premier Pro. So run the process for each product individually!
  • Timing matters a lot if you’re an early stage company as well. Chances are, your customers that signed up this quarter are probably different from the customers that joined a year ago.
  • This process is specific to companies with a “user sessions” measure. You can run something similar for other types of businesses (APIs translate pretty easily, whereas e-commerce is a little different)
  • This method requires that you have a decent amount of data in order to come up with a generalizable CJM. If you aren’t able to manipulate data in SQL or Python, you might want to get someone on your data team to help you.

Step 1: Prepare session data

PSA: this next bit is about to get really mathy, really quick.

For this example, I’m going to pretend I work for a mature streaming service - let’s call it “Peaflix Plus” - people pay a subscription fee of $10.99/mo for, and in return, get access to original content like “Strangest Thing” and “19 Reasons Why Not,” as well as classic movies like “Ferris Butler’s Day In.”

We want to identify the criterion for our “Loss of Interest Moment,” so we’ll start by pulling user sessions. We want to make sure that we get a good sample population, so we’ll add these filters:

  • User activated between 18 and 20 months ago (cohort)
  • User is female (more of our users are female than male)
  • User is between the ages of 35 and 44 (our biggest age demographic)

Once we’ve identified these criterion, we’ll move on to the business of pulling user sessions. These might be stored in your event system or through a service like Mixpanel or Amplitude. The data might look something like this:

Step 2: Normalize the data

We need to process the data so that we know how many sessions each user had in a given period - for example, monthly or weekly. You can also choose to bucket them as number of days in a month with a session, or by buckets of time.

Here’s an example where the sessions are bucketed by month (note: this is just a snippet, your data will be many orders of magnitude larger).

After you’ve done this, an easy thing to do is make a column for the number of sessions that came in the period after the one that’s being calculated:

Tip: depending on how sparse your session data is, you might want to reduce the rows to only ones that fall within 25%-75% of the total distribution.

Step 3: Get Probabilities

The last part is simple. For each distinct “number_of_sessions,” calculate the probability that the next period is zero (or whatever you have dictated is a good “low usage” barometer - feel free to experiment!)

You might end up with a diminishing curve resembling the one above - and you can tell that the “loss of interest” moment hovers around 11 monthly sessions. The increase in probability from 13 sessions/month and 11 per month is minimal, but the probability quadruples when the sessions per month drops from 11 to 9.

This is also the point where the most customer LTV is lost. If we assume that the probabilities hold true, and we use a standard calculation of LTV…

LTV = Average Revenue Per User / Revenue or Customer Churn

… then the loss of LTV between 13 sessions/month and 11 sessions/month =

(10.99/.108) - (10.99/.12) = $101.76 - $91.58  = $10.18

Whereas the loss between 11 sessions/month and 9 sessions/month =

(10.99/.12) - (10.99/.434) = $91.58 - $25.32  = $66.26

One is a $10 loss per customer, while the other is a $66 loss per customer — quite a large difference!

And there we have it — you now know that for a large chunk of your business, the average “Loss of Interest” moment (which is a high precursor to churn) is about 11 sessions per month. Something many companies we’ve worked with do is center their messaging around preventing folks from staying below that “loss of interest” threshold for too long because it’s exponentially easier at that point to re-engage a customer than when they actually go to cancel their subscription.

If you want to do all of the above automatically, plus other things like automating outreach for customers that fall below the “Loss of Interest” threshold or understanding which of your product’s features correlate to high engagement, let us know. Cotera enables teams to access and act upon below-the-surface data, which is pivotal to developing and refining your business’ strategy and growth.

Let me tell you a story about customer experience ...

We run a podcast interviewing best in class CX teams. We get to know how they are helping their businesses win.

We'd like to share some stories of what good CX looks like, and what, uhhh! - less than good looks like. Are you in?