An explainer video is also available for this topic.

Every subscription business knows the feeling: customers who seemed loyal suddenly cancel. Whether you are running a digital publication, shipping a monthly subscription box, or offering a learning app, churn is a fact of life. And while some customer turnover is inevitable, reducing it can make the difference between steady growth and constant firefighting.

The trick is spotting who is at risk of churning and acting to retain them before they leave.

Why it’s worth the effort

Bringing in new subscribers costs time, money, and energy whilst retaining existing ones is usually cheaper and more sustainable. The challenge is that churn does not strike evenly: some people stay for years, others only a few months. Sending out the same retention offer to everyone rarely works - it risks irritating loyal customers who were never planning to cancel.

A smarter approach starts with understanding the people who already churned: how they used their service and signals that may predict the behaviour of your current customers.

Step 1: Look back to look ahead

Your past churners hold the clues to future behaviour. What did they have in common? A few things to check:

·      Engagement levels: Did they log in less often, or stop opening emails?

·      Feature use: Were there parts of your service they ignored?

·      Attitudes: Did survey responses reveal subtle warning signs? For example, in an online news site: 'I prefer to avoid conflict' or 'news should stick to facts'.

Even a small set of data points can highlight patterns you would not spot at first glance.

Of course, not all churners will be the same and you can segment them to identify discrete groups with similar churn signals. This will ultimately make the process of identifying potential churners, and strategies for retention, as productive as possible.

Step 2: Find the lookalikes

Once you have identified the signals, the next step is to see which of your current subscribers look similar to your churners. A useful way to think about this is in two layers:

·     Critical cells: the core features you can rely on to make fair comparisons across the whole base – things most subscribers have recorded, such as length of subscription, package type, or frequency of use.

·     Hooks: the extra details that sharpen the match and show differences in risk, such as detailed usage patterns, attitudes, or responses to offers.

Together, they allow you to see which active subscribers most closely resemble those who have already churned.

Step 3: Where Fusion fits in

Doing this properly means making full use of the data you already hold, such as subscription records, usage logs, or survey responses. That is where Fusion comes in.

Fusion is a statistical technique for finding the closest matching respondents based on shared characteristics. It compares records using critical cells and enriches the match with hooks . In practice, this means finding those active subscribers who most closely match the traits of churners, so you can apply insights reliably without relying on a single personal identifier.

In this use case, unconstrained fusion would be used. Unconstrained fusion designates one dataset as the “recipient” and the other as the “donor”. The Donor dataset is the churned subscribers; they are the ones you study to understand which traits (hooks and critical cells) are associated with churn. They “donate” their profile information (demographics, behaviours, attitudes) as the basis for comparison. The Recipient dataset is the current active subscribers. They are the population you want to enrich with “churn likelihood” information. Fusion compares their profiles to the donor set and assigns similarity scores / nearest-neighbour matches.

The outcome of the fusion is a structured, ranked list of subscribers who most resemble those who have already left - giving you an evidence-based way to predict churn risk and act early.

You can read more about unconstrained fusion and how it differs from constrained fusion by downloading our white paper.

Step 4: Act early and with relevance

With that ranking in hand, you can act before people cancel. For example:

- A newspaper might highlight content based on the preferences of individual at risk subscribers such as investigative or environmental content;

- A subscription box might include bonus items to customers identified as at risk of churn "you are entitled to a bonus item next month, choose from...!" ;

- An app might highlight features such as community spaces that are linked to longer retention.

The principle is the same: direct effort where it matters most.

Keeping churn in check

Churn will not disappear completely. But by learning from those who have left, applying Fusion to find their nearest lookalikes, and tailoring your response can keep more of your loyal subscribers for longer, without the constant expense of replacing them.

If you would like to see an example in action, our short explainer video walks through the process step by step.

‍

How RSMB can help

• Our fusion platform gives you the ability to identify potential churners by running your own fusions in house - with the added bonus that you can also use it to integrate your siloed datasets too.
• Or RSMB can do all the analysis for you: we can segment your churners and use fusion and other statistical techniques to attach a churn likelihood score to each of your subscribers based on the data you have available for analysis.