Nothing is more painful than seeing your customer walk out the door. And churn is costly.
On this page we help you understand the concept of churn and churn analysis using machine learning
From churn analysis to predictive churn analysis - a guide
The basis to churn analysis is about "understanding your customers” – and the basis to predictive churn analysis is about being able to be proactive with your customer and act before they churn. This guide is aimed at companies curious to learn more about how to connect their churn analysis to their loyalty management activities, ultimately improving your customers’ satisfaction and life-time value
Regardless of whether you have a detailed churn strategy & plan in play or whether you are starting on this journey, this guide should give you insights and areas to focus on in relation to utilise the benefits of ML that can support your churn prevention activities in terms of budgets, tools, resources, etc.
Navigate the site
The term “churn” generally covers the concept of customers who have left you. But depending on your business, your definition on churn might differ; for a subscription business it is usually when a customer unsubscribes, but non-subscription business, churn will have to be defined in time since last purchase or transaction; the repeat-buy cycle varies dramatically across industries from opticians to clothing, to financial services, etc.
If you have no industry standard to lean on, but have a lot of internal transactional data, the first step could be to define the frequency which your best, average and worse customers buy with. For some a churned customer could be someone who has not made a purchase within 30 days, for others it might be customers who have not made a purchase within the last 12 months, etc.
Congratulations on your new customer! Someone just bought something from you or subscribed to a new subscription and now your effort to make them a loyal customer begins; schematically you have these pre-defined key customer life-cycle stages: on-boarding, active, dormant/win-back. It takes no rocket scientist to know, that “active” is where your customer is the most satisfied and you want your customers to be here for as long as possible. Therefore, one cannot talk about churn without talking about loyalty as well; churn prevention and loyalty are two sides of the same coin
Happy customers churn too; a certain level of churn is inevitable regardless of your best endeavours to keep them happy – this is usually due to key life-stage events for the customer – e.g. moving to a new house, getting a family, changing jobs, etc. Whether your churn rate is too high, depends on your industry standard or the KPIs in your organisation. It’s difficult to find publicly available benchmark numbers, but if you don’t know the specific churn benchmark for your industry, and would like to get insight from a broader global scale, then Recurly Research has a good reference point (covering 10 different industries all characterised by being subscription businesses)
By implementing clear churn prevention activities, you will improve your customer satisfaction levels, which in turn will also improve your brand. What’s not to like?
However, nothing changes if you do what you have always done! So the first stage is to ensure that your organisation is ready meaning that you need to prepare yourself and the organisation that the results of your churn analysis can and will have an effect across the organisation on customer flows, offerings and processes, systems, etc.
Calculate your current churn cost to make your business case
Knowing your current churn cost makes it easier to set up KPIs for your churn prevention/loyalty enhancement efforts. Below there is a simple calculation to give you a taste of how much a reduction in churn could save you in costs and revenue that in turn benefits the bottom line. In the “we can help you” section, we offer you a standardised calculation of how much we can help you reduce your churn cost with an ML-driven predictive churn model. However, the actual saving will depend on your current churn prevention activities and could be much higher
There is no question a churn analysis should be based on data; you need to base your future optimised loyalty management activities on the facts. For the best results with any churn analysis, the data-view should cover both internal data sets, and external 3rd party
There is a lot to learn about your customers’ churn from your internal data. And there is a lot of data – “big data”. Below is the most commonly used internal variables used for churn analysis:
- time on book
- what they bought
- at what value
- how do/did they pay
- customer channel interactions
- customer service:
- support requests
- customer satisfaction:
- NPS/customer satisfaction data
- marketing permissions
- newsletter/marketing mail engagement
- digital engagement
Getting data ready
Data is not always easily available, structured and ready to go. And some data can be really difficult to get hold off. As a rule of thumb, our experience in this space is to always weight up the need to have vs nice to have, vs the benefit, vs the ease of access. Basically, what do you need to have at an absolute bare minimum, and what could be done without. Of course, it can be that a data item is deemed need to have but once the complexity to extract is added in, then its overall benefit is lowered to be a nice to have. Many churn analysis projects never get off the ground as organisations start by looking for the ultimate setup and then spend too much time in trying to achieve this, instead of being more pragmatic
External 3rd party data
External data usually covers 3rd party data that can add to the knowledge you have on your existing customers. The main reason for why 3rd party data is so beneficial is that it helps you map the missing parts of the puzzle in relation to insights into your existing customer – by having this view you are able to understand the customers’ needs and what they do outside of your environment or universe – basically giving you the full 360° view
This data can vary in their origin and structure, but what they have in common, is that they help they shed light on your customers’ characteristics. This data strengthens your churn analysis considerably, as churn is often caused by changes outside your environment; a family moving, divorce, the birth of a family’s first child, etc. Due to legislation & GDPR, you cannot get 1:1 views on this data, but there is a variety of 3rd party data views (small cluster statistics) and triggers that can be added to your model and describe your customers more
List of external data commonly used for churn analysis:
- consumer profile
- consumer preferences
- digital behaviour
- data on their housing type, size, etc.
- triggers (moving, one person leaving the household, etc.)
Accessing external data
External data from 3rd party sources can either be accessed through different sources, either individually from the data source or via data providers distributing multiple data sources into a ‘hub’. What’s important to keep in mind is that the external data needs to be structured and accessible in a way compatible with your internal data and is easy to access . You are reading this on the website of a data provider that can support you with access to a wide variety of 3rd party data via our data-lake solution. Learn more about that here
A churn analysis will identify the characteristics of your churning customers. You will learn when they churn, what variables are most explanatory and what characterises your churners as customer segments. This will help you understand your churn and build campaigns and flows taking these learnings into account
In our Data Analytics Platform we offer a data blending tool that dramatically reduces time and resources spent in the preparation stage ahead of your churn analysis by blending your internal customer data with our Nordic data lake (counting 800 3rd party variables) and ranking them in terms of their predictiveness across your own data and the external data. This gives you an overview of the data best suited to include in your churn analysis that we can help you perform in our analytical model builder Predict. It helps you develop better models and speed up the time-to-market of your churn prevention activities; we have all the structure and framework around the non-direct modelling tasks, and we generate the models in the self-service environment. You can call it automated AI if you will. Click here to learn more
If you are ready with a more advanced set-up, a predictive churn model using machine learning is what could move your churn prevention activities to the next level; deploying a predictive churn model to your customer flows could trigger data-defined recommendation codes based on your customers’ actions or in-actions. This enables you to be proactive with your customers and act before they churn
The benefits of an ML-driven predictive churn analysis is the agility of the model; over time it will adjust to the learnings of new data added to the model and your customers’ behaviour meaning you also learn more about the effects of your churn prevention flows and can adjust them at a lower cost than if you had to re-do the customer churn analysis all over again
Our Data Analytics Platform is designed to make it easy for you to build a predictive churn model with machine learning; our data blending tool dramatically reduces time and resources spent in the preparation stage by blending your internal customer data with our Nordic data lake (counting 800 3rd party variables) and ranking them in terms of their predictiveness across your own data and the external data
Whether you want to build your churn model using our ML-driven model-builder Predict or build it in your own environment, this fast track your process of building or enhancing your predictive churn model. Click here to learn more
The process from analysing your customer churn to deploying actual action codes in your customer flows is highly complex; even though you identify the common behavioural patterns of a customer about to churn and you can set up triggers, it might not be as simple to define exactly what action is the right one, the right message etc. But with a thorough churn analysis using the right amount of data, you have some good help for this next task
Steal the knowledge of others
Here are three use cases where we helped customers with their churn analysis that have different complexity levels to implement
In-creasing win-back on dormant customers
Should we talk about how to reduce your churn?
Sign up below or get in touch with me directly