Predicting customer churn

Sponsored by

When I first came into the role as business partner for the sales team, I was bombarded with anxiety by the sales leaders…

“Why are you asking me to go get MORE sales?? I’m already above my sales plan for the year and the increased customer churn isn’t my problem”

The truth was… they were right.

The customer churn that we loaded into the plan was way too low and we were seeing more customers churn than we ever expected.

Which meant we were going to miss our revenue targets unless we did something different asap.

Unfortunately, the ship had sailed on getting a better customer churn prediction for this year - the plan was set.

And I had to deal with the blowback.

But I could clearly see the cause of the issue: a broken customer churn model.

Why do I say broken?

Because the historical data in the model didn’t even tie out to what our actual customer churn was for the month.

Whoever built the model that I inherited was just being lazy.

So I made it my goal to rebuild the model and never again find myself in a position with not enough recurring customers to hit the plan.

It was time to build a better customer churn model…

Why is predicting customer churn important?

Customer churn is a metric that financial professionals help the business to predict, referring to the rate at which customers are leaving. The goal of a business with recurring customers and revenue is to minimize churn and build a base of repeat customers.

At a high level, a company’s revenue is made up of new customers and repeat customers. And having a bad customer churn forecast means you’ll have to achieve your revenue forecast via new customers to make up the gap.

Not an easy pivot to make when you are half-way through the year with no marketing budget left to go get more new customers.

Understanding churn is crucial for the business world; it's a financial indicator that can predict future performance.

• How many customers are you likely to lose over the next quarter?

• How much revenue is that connected to?

• What trends can indicate this customer loss?

When you can answer these questions, you equip yourself with data to influence strategic business processes that can mitigate this loss and focus on retention.

Here’s my simple approach to predicting customer churn:

Visual-based newsletter on business and tech

We explain the latest business, finance, and tech news with visuals and data.📊

All in one free newsletter that takes < 5 minutes to read. 🗞

Save time and become more informed today.👇

Subscribe with one click today

Construct a cohort analysis for deeper insights

The statistical method of cohort analysis has a way of cutting through the noise of raw customer data and providing clear, segmented insights into how different groups of customers behave over time.

Creating cohorts of customers is maybe the easiest way to get started with customer churn analysis and forecasting. It’s how I recommend you start your analysis.

What Is a Cohort?

A cohort is a grouping method that slices your user base into segments based on engagement or acquisition patterns in a given time frame.

Having your data arranged into cohorts will almost always immediately show you where the problems are in your customer churn.

For example, if you arrange your cohorts by months from sign-up date then you should quickly see that 60% of your churn happens at month 3 and if the customer makes it past that point they don’t churn for another 9 months (for example).

From this point you can jump into more detailed analysis, but this is the 20% of effort that gives you 80% of the insights.

Building Your Cohort Analysis

To construct an effective cohort, consider segments such as sign-up date (my favorite), first purchase date, or source of acquisition. Once you have these clusters identified, it's time to track their churn over time, looking for patterns and consistencies that can guide your predictive model.

Using cohorts is a great place to start with customer churn analysis. You can begin to predict calendar month July 2024 number of customers churned simply by looking at the average lifetime of your typical customers and matching that up with your customer acquisition data from past cohorts.

Better said, arrange your prior cohorts of customers down the rows

It’s a simple methodology that doesn’t require much data analytics and is easy for the business to understand.

But in order to improve our forecasted churn accuracy, it’s wise to go a step deeper into more customer segmentation.

Identifying the drivers of churn

Within each business and type of customer, there will be unique ways that you should analyze your data to better predict customer churn.

Variables like onboarding experience, customer service interactions, product usage, and recent service or price changes can be significant churn catalysts. By isolating and studying these, you'll be projecting the specific root causes of churn within your customer base for each time period.

Where do you find these drivers of churn if this is your first time trying to predict it?

• ChatGPT ‘what are the causes of churn for companies in ____ industry’

• Ask your business partners if they have any assumptions of what causes churn

• Explore the data and see what data elements you have that might be helpful

The goal here is to grab a few hypotheses that seem reasonable and spend time pulling the data to prove or disprove if they are drivers of churn.

For example: You may find that when customers stop using your product on a daily basis, they churn at a rate of 20% that month, 40% the next month, and 80% the following month.

By pulling your customer data by usage, you’ll be able to predict which of those customers are most likely to churn.

And the great thing about this data point is that you can work with your business partners to proactively reach out to the customers to offer specialized support to keep them on the platform.

The goal is not only to predict the churn but to help your business partners do something about it.

A word of caution: ensure your data is accurate when doing this kind of analysis. It can be tempting to pull all kinds of historical data into this, but as companies change over time the data quality may be spotty for customers.

Backtesting your predictions

As your customer churn analysis and model gets more sophisticated, you’ll find that you want to layer multiple drivers of churn together to build a legitimate predictive model.

But having a model like this can become a liability if you aren’t constantly revisiting it to understand its predictive accuracy.

More sophistication often means more maintenance.

This is where backtesting comes in, rigorously assessing your model’s performance against historical data.

Take a subset of your data's past and predict churn as you would for the present or future.

Did the predictions align with historical reality?

If so, your model’s predictive reliability is sound. If not, it’s back to refining the metrics.

Predicting customer churn begins as a 1-time event but should evolve into a constant evolution. Your model should be revisited and recalibrated as you acquire new data and as customer dynamics evolve.

I’ve found that building a mechanism to do a high-level backtest monthly is appropriate, but then doing a more thorough review quarterly or annually is ideal.

Case Studies for Inspiration:

These are fake companies and examples.

A pro move is to feed these examples into ChatGPT and ask it ‘how would a company with [specific problem] in [specific industry] implement the techniques found in these case studies to analyze and predict customer churn?’

Case Study 1: Bank of Tomorrow Implements Cohort Analysis

Bank of Tomorrow, recognizing the predictive power of cohort analysis, segmented its customer base into cohorts based on account opening dates. By analyzing the behavioral patterns and churn rates of these cohorts, the bank identified specific periods when customers were more likely to churn. Tailoring their retention strategies to these critical periods, the bank significantly reduced its overall churn rate, enhancing customer loyalty and profitability.

Case Study 2: GlobalTech Leverages Predictive Modeling

GlobalTech, a leading software company, developed a sophisticated predictive model to anticipate customer churn. By incorporating a wide range of variables, including usage frequency, support ticket history, and payment behaviors, the company was able to predict with high accuracy which customers were at risk of churning. Through targeted intervention strategies, GlobalTech not only retained at-risk customers but also identified opportunities to enhance user satisfaction and engagement.

Case Study 3: HealthFirst's Journey with Backtesting

HealthFirst, a subscription-based health and wellness service, faced challenges with customer retention. They implemented backtesting of their predictive churn models by comparing predictions against actual churn over the past year. This rigorous evaluation highlighted discrepancies in their model and pinpointed overlooked behavioral indicators of churn. By refining their model based on backtesting results, HealthFirst achieved a significant improvement in predicting churn, allowing them to proactively address customer needs and reduce churn rates.

In Summary:

Predicting customer churn is a complex dance of numbers and behavioral psychology.

Nonetheless, it’s a dance that every financial analyst must master.

By dissecting customer data through the lens of cohort analysis, identifying the drivers of churn, and meticulously backtesting your models, you arm yourself with the insights to not only predict customer behavior but to influence it.

In the digital age, data is power.

And bringing this level of insight to your business partner is where you’ll add immense value.

 Say hi 👋 on LinkedIn