Churn Prevention

Brands use churn prevention campaigns to win back customers who have not purchased within defined time windows, such as 90 days, 120 days, or even up to 2 years. The length of the defined time window and the types of promotions and offers you will send depends on your goals for each audience within the churn prevention campaign.

It’s Not You, It’s My Data: A helpful book about churn prevention campaigns

It's not you, it's my data

Brands that don’t adapt—and marketers that don’t adopt new technological skills—will not last.

One of those primary skills is systematically leveraging data to grow customer loyalty and improve customer retention.

This book focuses on retention, and how to build a churn prevention machine that keeps your customers engaged for the long run.

Open the PDF to read the whole book

What is a churn prevention campaign?

A churn prevention campaign seeks to engage high-value customers as their likelihood of making a transaction diminishes. This type of campaign uses a series of messages based on thresholds to apply a variety of offers, discounts, and messages as probability of purchasing declines.

An established churn prevention campaign, as the following diagram represents, may have many thousands of customers spread across four churn thresholds, each ranked by lifetime and predicted customer value, and then each divided into specific categories that align to your typical marketing campaigns and strategies.

An established churn prevention campaign.

This can create a large number of specific types of campaign messages, offers, promotions, marketing channels, etc. that are all associated to a single churn prevention campaign!

The following sections describe the process of setting up an effective churn prevention campaign, including how to define your audience, test the campaign, extend the campaign by adding additional parameters and tools, and then optimize the campaign for all audiences and automation.

Tip

The probability of churn, as defined by a customer’s lifecycle status, is not monotonic. In other words, the probability of churn is not “always increasing or always decreasing” over time. Some customers who are at risk of churn may move to highly at risk, and then move back to at risk. Some customers who are highly at risk may move to lost. Some customers who are highly at risk may remain highly at risk.

Use the Campaign_Recipients table to ensure that customers are not sent repetitive churn prevention messages. For example, if you want to send a message to customers who are moving from at risk to highly at risk, use the Campaign_Recipients table to look back and identify customers who have already been sent that message, and then exclude customers who have not purchased within the past year from that segment.

How to build a churn prevention campaign

Building an effective churn prevention campaign starts with defining audiences that can be mapped to a series of churn thresholds. Experiment by sending an increasing number of promotions and offers to an increasing number of audiences (and sub-audiences) and marketing channels as you test, expand, and then optimize your churn prevention campaign.

Plan a series of promotions and offers that can be tied to each audience or sub-audience within your campaign. For example:

  • Modest discounts (10% off)

  • Substantial discount (25% off)

  • Free shipping

  • Exclusive offers

  • Buy one and get one free

  • and so on

Define an audience

Churn prevention campaigns focus on customers that fall into specific categories. To find these customers, start by building a segment that uses a combination of attributes that describe customer details, including purchase history, purchase recency, average order value, product channels, and loyalty status.

Then use the predicted customer lifetime value (pCLV) attributes to group them by customer lifecycle status: active, cooling down, at risk, highly at risk, and lost.

Churn prevention segments are often focused on smaller groups of customers. Use specific criteria to identify which customers belong in the segment, but remember that customers are likely to respond to a promotion or offer in different ways:

  1. A segment for newly at risk customers who have not purchased in the previous 60 days.

  2. A segment for customers who have not purchased in the last 90 days, but have an average order value of more than $200, who are members of your loyalty program, and with an cooling down lifecycle status.

  3. A segment for newly highly at risk customers who have not purchased in the previous 120 days.

Customer lifecycle status is represented by a probabilistic score–referred to as p(return) or “probability of return”–that identifies if a customer is active or if they are likely to churn.

A customer’s p(return) score determines the customer lifecycle status tier to which they are assigned:

Status tier

p(return) score

Active

p(return) score is over 60%

Cooling down

p(return) score is between 50%-60%

At risk

p(return) score is between 35%-50%

Highly at risk

p(return) score is between 20%-35%

Lost

p(return) score is below 20%

A customer is more likely to churn when they have been assigned a cooling down, at risk, highly at risk, or lost p(return) score, with increasing likelihood of churn as they move toward lost. A customer who is more likely to churn is the type of customer that should be the primary focus of a churn prevention campaign.

Test the campaign

Testing and measuring the performance of a churn prevention campaign can be a challenge. A sophisticated marketing campaign is run across multiple channels: email, direct mail, display advertising, Facebook, etc. and likely identifies many types of sub-audiences within a larger group of customers.

To get there, start with a simple campaign using only customers that are cooling down, and then divide them randomly into three groups:

  1. A control group at 10%. This group will not receive a promotion or offer.

  2. A recipient group at 45% that receives promotion or offer A.

  3. A recipient group at 45% that receives promotion or offer B.

Test the churn prevention campaign.

Send this campaign to a single marketing tool, such as your email service provider. This is the location from which you will run the initial churn prevention campaign.

The most important measurement is identifying how many customers do not churn. This is done by comparing revenue per user, profit per user, and overall conversion between the control group and all recipient groups. If one of the test outperforms the other, consider testing that promotion or offer again, but with a new set of control and recipient groups, to help identify ways to improve it.

Expand the campaign

After you have established a rhythm with testing and measuring churn prevention campaigns for a small, targeted audience, it’s time to start expanding the campaign to add more customers, more value tiers, and more marketing tools.

Update the initial test campaign to include customers who are at risk. For both cooling down and at risk customers, apply predicted customer lifetime value (pCLV) to identify customers who fall into the high and medium value tiers, and then, for each value tier, divide them randomly into three groups:

  1. A control group at 10%. This group will not receive a promotion or offer.

  2. A recipient group at 45% that receives promotion or offer A.

  3. A recipient group at 45% that receives promotion or offer B.

Expand the churn prevention campaign.

Send these campaigns to your email service provider and another marketing tool, such as Facebook. These are the locations from which you will run the next phase of your churn prevention campaign.

Continue comparing revenue per user, profit per user, and overall conversion, but this time for each set of control and recipient groups.

Start to tailor your promotions and offers to the customer lifecycle status and to the value tier. Continue to test a variety of messages across a greater number of audiences and sub-audiences and keep using the control and recipient groups pattern for each audience and sub-audience.

Optimize the campaign

Update the expanded campaign to include customers who are highly at risk and lost. Apply predicted customer lifetime value (pCLV) to all stages and expand to include all value tiers. For each combination, continue to divide them randomly into three groups:

  1. A control group at 10%. This group will not receive a promotion or offer.

  2. A recipient group at 45% that receives promotion or offer A.

  3. A recipient group at 45% that receives promotion or offer B.

Optimize the churn prevention campaign.

Send these campaigns to your email service provider, Facebook, and then continue adding additional marketing channels, such as direct mail.

Continue comparing revenue per user, profit per user, and overall conversion across each set of control and recipient groups. Continue to tailor your promotions and offers to specific combinations of customer lifecycle status and value tier.

Continue to test a variety of messages across a greater number of audiences and sub-audiences and keep using the control group and don’t stop using the control and recipient groups pattern with each audience and sub-audience. Get more creative with the campaigns that are sent to your most valuable customers.

Consider using additional recipient groups (along with a control group) to allow for testing additional messages. Continue reusing promotions and offers that perform well. Look for ways to reuse creative assets and messages across your churn prevention campaign.

Start to automate elements within your churn prevention campaign with a goal of eventually automating the entire workflow. For example, start by configuring Amperity to automatically re-run the segments that power your churn prevention campaign, and then send those results automatically to your email service provider and Facebook.

Churn prevention attributes

Amperity provides a series of predictive attributes for predicted customer lifetime value (pCLV) that you can use to understand how your customers fit within the various stages of your churn prevention campaigns.

Start with the following pCLV attributes:

  1. Predicted CLV

  2. Predicted value tiers

  3. Predicted lifecycle status

and then use the individual components of pCLV – predicted average order revenue, predicted order frequency, and probability of transaction – to fine-tune your understanding of your customers, and then get more precise about where they fit within the various stages of your churn prevention campaigns.

Apply the following non-predictive attributes to the segments within your churn prevention campaign:

  1. Days between orders

Predicted CLV

Predicted customer lifetime value represents the total value of all orders a customer is predicted to make if they return to make another purchase during the next 365 days.

Each prediction is represented by an amount, where a higher amount represents an increase in the amount of money a customer will spend (in total) during the next 365 days: a 25.75 score is a likelihood that a customer will spend $25.75, a 140.45 score is a likelihood that a customer will spend $140.45, and a 295.25 score is a likelihood that a customer will spend $295.25.

To apply predicted CLV to your segment, start with the Predicted CLV Next 365 in the Predicted CLV Attributes table, and then set its condition to is greater than or equal to.

Choose the predicted CLV attribute from the Segment Editor.

Set the value for predicted CLV to the amount you want to use to represent the lowest amount of predicted value you want to use with your churn prevention campaign. You might want to try out ranges of values to see which ones return the right number of customers to use with your campaign.

Tip

Predicted CLV is the multiplication of three components: 1) predicted probability of return, 2) predicted order frequency, and 3) predicted average order value.

Each component of predicted CLV is also available as an individual score:

  1. Predicted probability of transaction represents the likelihood that a customer will return to make another purchase during the next 365 days.

  2. Predicted order frequency represents the number of orders a customer is predicted to make if they return to make another purchase during the next 365 days.

  3. Predicted average order revenue represents the average value of each order a customer is predicted to make if they return to make another purchase during the next 365 days.

Use these components individually to help identify customers that are more (or less) likely to fit within stages of your churn prevention campaign.

For example, look for customers with a high score for predicted average order revenue who have low scores for predicted order frequency and predicted probability of transaction. This combination often represents customers who are not currently engaged with your brand, but are likely to spend more than your average customer if they were to make another purchase.

These attributes are available from the Predicted CLV Attributes table: Predicted Average Order Value Next 365, Predicted Order Frequency Next 365, and Predicted Probability of Transaction Next 365.

Predicted value tiers

Predicted customer value tiers represent a percentile grouping of customers by predicted customer lifetime value (pCLV). Value tier groupings include:

  • Platinum: top 1% of customers

  • Gold: top 1%-5% of customers

  • Silver: top 5%-10% of customers

  • Bronze: top 10%-25% of customers

  • Medium: top 25%-50% of customers

  • Low: bottom 50% of customers

To find predicted value tiers, start with the Predicted Customer Lifetime Value Tier attribute in the Predicted CLV Attributes table, and then set its condition to is in list. After the attribute appears in your segment, set the list values to include all of the value tiers you want to use in your segment.

Choose the predicted customer lifetime value tier attribute from the Segment Editor.

Choose the is in list attribute, and then set predicted lifetime value tier to Platinum, Gold, and Silver:

The is in list condition for predicted value tier.

Predicted lifecycle status

Predicted lifecycle status represents the likelihood to purchase again, grouped by tiers: “active”, “cooling down”, “at risk”, “highly at risk”, and “lost”. These tiers are used for two types of customers: repeat customers and one-time buyers.

For repeat customers, lifecycle status groupings use the following thresholds:

  1. A repeat customer with an “active” lifecycle status has a greater than 60% likelihood to purchase again.

  2. A repeat customer with a “cooling down” lifecycle status has a 50-60% likelihood to purchase again.

  3. A repeat customer with an “at risk” lifecycle status has a 35-50% likelihood to purchase again.

  4. A repeat customer with a “highly at risk” lifecycle status has a 20-35% likelihood to purchase again.

  5. A repeat customer with a “lost” lifecycle status has a less than 20% likelihood to purchase again.

To find predicted lifecycle status, start with the Predicted Customer Lifecycle Status attribute in the Predicted CLV Attributes table, and then select the is in list condition. After the attribute appears in your segment, set the list values to include all of the thresholds you want to use in your segment.

Choose the predicted lifecycle status attribute from the Segment Editor.

Choose the is in list attribute, and then set predicted lifecycle status to active, at risk, and cooling down:

The is in list condition for predicted lifecycle status.

Days since latest order

A churn prevention campaign must measure the time that has elapsed since a customer’s most recent order and the current date.

Use the in between operator along with relative date ranges to associate customers to specific stages within your churn prevention campaign.

For example: if your churn prevention campaign has stages at 14 days and 1 month after the customer’s most recent purchase, add the following criteria to the 14-day segment:

"Days Since Latest Order" is between 0 and 14

and the following criteria to the 1-month segment

"Days Since Latest Order" is between 0 and 30

To find the number of days that have elapsed since a customer’s most recent order, start with the Days Since Latest Order attribute in the Transaction Attributes Extended table, and then set its condition to is between.

Choose the days since last order attribute from the Segment Editor.

After the attribute appears in your segment, specify the number of days. For example, to find customers for whom 90 days have elapsed since their latest order:

Find customers for whom 90 days have elapsed since their latest order.