Customer Retention Model: Definition, Creation & FAQs

What is a customer retention model?

A customer retention model is a framework for predicting whether a customer will stay with your business. It helps you identify which customers are most – or least – likely to buy your product or use your service again. You can use retention modeling to predict the impact of marketing activities on retention and churn.

By running customer retention models, you can focus your marketing efforts on the people who are most likely to respond, and send personalized offers to different segments based on predicted behavior, while also improving ROI.

Download Now

We use your information according to our Privacy Policy. You can update your preferences at any time.

3 things you need to do before selecting a customer retention model

Scalable customer retention models are complex machine-learning models. As a marketer, you won’t develop them from scratch. You’ll need your data team’s help, but you can lay the groundwork.

Understand your customer retention goals and metrics

Your goals and metrics for retention will help you determine which models to use. Say you need to maintain a minimum average order value. Run models that will give you insights into how to increase each order’s total amount. For example, a response model predicts which customers will buy more when offered a specific promo. A next best offer model identifies products you can recommend right after a customer adds an item to their cart. 

📚 Learn more about customer retention metrics and KPIs in our article on How to Create a Foolproof Customer Retention Strategy.

Gather data

Prepare the data you’ll need to feed into your models. For example, if you want to understand how newsletter subscription influences purchase behavior, gather data on:

• Which customers are newsletter subscribers

• Customer actions in response to newsletters (unopened, opened, scrolled, clicked)

• Purchase data (dates, items, quantities, basket value, channel used for purchasing)

This information is first-party data, which you obtain from direct interactions between your business and the customer (across your website, app, and so forth). You’ll need the capability to stitch together data from different sources to identify the same customer across multiple platforms – for example, to know that a given newsletter subscriber is the same person who shopped on your app. Instead of doing this manually, use a customer data platform (CDP) like Segment, which automates identity resolution based on data from various sources.

Enrich your analysis by considering how customer segments and cohorts respond to marketing activities. A customer segment is made up of customers with similar behaviors, interests, and demographic or psychographic traits, while cohorts comprise customers who performed an action during a given timeframe (e.g., signed up for a subscription in August). You can also use a CDP to automatically assign customers to segments or cohorts based on your specified criteria.

Identify what makes your solution sticky

People stick with your product once they experience its core value. To help customers experience value, your product team should be aware of which actions are necessary for a person to take (and when). You may already have an idea of what those actions are, but it’s best to validate them by mapping out the customer journey.  

A customer journey map is a visualization of a person’s interactions with your business throughout their lifetime. Mapping involves gathering first-party data on customers’ interactions with your brand, tracking feature usage, running heatmaps, and conducting surveys and customer interviews. Journey maps reveal customers’ “aha moments” – when they take valuable product actions that result in an experience of your product’s core value. 

You want to nudge customers to perform those actions regularly. Once you’ve developed a marketing plan to get customers to take those actions, plug your planned marketing activities into a response model, logistic regression model, or uplift model to discover their potential influence on customer behavior and identify which customers you should target.

📍 Learn more about how to map and analyze customer journeys.

3 customer retention models to consider

No single model can explain and predict all the nuances of customer behavior. After all, factors beyond your control like competition, economic crises, and environmental disasters all influence customers’ decisions. Using multiple models helps you understand the factors that make your customers stay or leave. 

For example, that logistic regression reveals that subscribers to your newsletter tend to shop on your e-commerce site more frequently. When you apply an uplift model, you discover that a subset of those subscribers shop only after reading your newsletter – they never shop without a nudge from you. You also run a next best offer model to find out how you can personalize your newsletters to feature products each recipient will most likely want to buy next.

If you have an in-house data team, ask them to help develop and validate customer retention models. Your data team can also help you implement an AI-driven customer retention platform like Voziq, use an automated machine-learning platform like H2O.ai, or work with an AI and analytics consulting firm like Addepto to develop an automated analytics and reporting process.

Propensity models

Propensity models predict the likelihood of a customer performing an action by analyzing patterns in their previous behavior. Examples include:

• Next purchase model — Predicts how likely a customer is to buy from you within a given period – say, within the next three months – based on their past purchasing behavior.

• Response model — Predicts whether or not a customer will take action in response to a marketing stimulus based on how they’ve responded in the past.

• Next best offer model — Predicts what product recommendation a customer will most likely respond to after they buy a given item (say, for example, a pair of boots). The model recommends the next best offer by analyzing what other customers tend to purchase next after buying boots. It also recommends products that are frequently purchased together with a pair of boots.

Logistic regression model

A logistic regression model tells you the relationship between one or more independent variables and a dependent variable. Independent variables may be marketing activities like advertising and sales events, or customer characteristics like their age, profession, or shopping habits. Dependent variables are retention and churn actions.

You use a logistic regression model to predict binary outcomes: yes or no, true or false, opened or unopened, retained or churned. That means you’d phrase your problem in these ways:

• Are readers of our blog more likely to make repeat purchases? 

• How does the probability of a customer returning our app change for every additional star rating they give it? 

• Do app usage frequency and e-wallet usage influence the probability of a customer signing up for a premium membership?

The results of a logistic regression analysis are plotted along an S-curve. Here’s an example from G2’s explainer on logistic regression (check it out if you’re keen to learn more about how this statistical method works):

Image source: G2

Keep in mind that logistic regression predicts outcomes but doesn’t definitively explain their causes. Think of it the way dark clouds are almost always followed by rainfall. They’re a reliable sign for you to bring an umbrella, but they don’t cause the rain itself.  

Uplift model

Uplift modeling predicts how likely it is a customer will respond to a marketing campaign based on how they responded previously. It reveals four types of customers depending on their chances of churning with or without retention marketing:

Uplift modeling reveals 4 types of customers

You want to capture your “persuadables” – customers who will take action (e.g., buy from you or renew a subscription) only with marketing communications and incentives like discounts. Save money by excluding from your retention marketing campaign your “sure things” (people who will purchase again even without hearing from you) and “lost causes” (people who will leave whether or not you send them messages and incentives). 

Avoid targeting “sleeping dogs”, too – customers who’d leave your business if they receive retention marketing communications, perhaps because these messages annoy them or change their perception of your brand. Some “sleeping dogs” are profitable customers, but aren’t highly engaged with your product. Think of a streaming app subscriber who gets offered a premium plan upgrade at a discounted price. The offer makes the subscriber realize they’ve gone more than a month without watching shows on the app – and they didn’t miss it. They end their subscription instead.

That doesn’t mean you should no longer reach out to “sure things” and “sleeping dogs”. Engage them in ways other than retention marketing, such as through loyalty programs or interesting content. You may want to test those programs with a small subset of sleeping dogs first to be sure you won’t irk them.

📺 Data analysts use a variety of uplift models. Watch Bart Baesens, a professor of Big Data and Analytics at KU Leuven in Belgium, run through a sample uplift model here.