As marketers we’re familiar with the concept of lifecycle marketing, and personalisation and relevance are intrinsic parts of our craft. How we can prevent customer churn and influence retention, however, may feel less attainable. And for a very good reason.
Identifying churn and building predictive churn models requires not only a large set of data – often gathered from different areas and silos of the business - but also a team able to make sense of it all: establishing time frames and trigger points which point towards the likelihood that a customer will churn. This is often beyond the reach of a marketing team due to resource, time, or skill constraints.
Once established, however, this information can inform tactics and programmes targeting those customers who are likely to leave, driven by the smart collection, understanding and manipulation of these data sets.
First things first - what is churn?
Customer churn (or attrition) is something that affects every company. At its core it represents the customers that are no longer paying for (or making use of) your products or services. It is often calculated as a ‘churn rate’ – the percentage of customers who’ve left over a specified time frame.
You can break it down further into ‘voluntary’ and ‘non-voluntary’ churn. As the name suggests, ‘voluntary’ churn is through a conscious decision by the customer to leave, while involuntary could be due to things like moving to another area.
Churn analysis and predictive modeling
Knowing that customers are leaving – and how many – is only the start of the journey. From here you will want to go through the process of analysing the data of those who have churned and identify the attributes or behaviours (risk triggers) that are common to those who have left.
When Qrious worked with Skinny Mobile they identified that there were nine separate risk triggers that indicated a propensity for churn. This included things like tenure, device, cost of plan, and whether they had added their age to their profile or not.
This then shapes their predictive churn model. By overlaying the data from current customers with these trigger points they were able to identify those likely to leave – and take action.
There is no single solution
Unfortunately, there is no single method or ‘model’ that will help identify the cause of churn and establish ways to combat it. Just as every organisation is unique – so is the model that will identify the rate and reason for churn.
As an example, one model is survival analysis. This estimates how long it takes for a particular event to happen. i.e. once a customer starts, how long before that customer stops? By assuming that the future will be similar to the past, based on an ‘homogeneity assumption’, historical data can help determine what will happen, and when.
Specialist data scientists are often needed to develop and understand the data and build the model that is relevant to your organisation and customer lifecycles. They will help you establish your organisation’s definition of churn, if there are any exclusions that apply - like low value customers or those who sit outside of your target market – and a time period on which to apply the analysis. From here they will be able to get an understanding of the behaviours or triggers that identify someone likely to churn so you can implement actions that will get them to stay.
As an example, when working with Skinny Mobile’s data, Qrious identified 149 data variables, of which only 110 were fed into the churn model. The other 39 were identified as having no relevance to the propensity for churn. They then used activity data from the two months prior to inactivity to identify the common risk factors. The model was applied to un-churned customers to calculate churn scores, identifying those that were exhibiting the behaviour that points to them leaving.
Whatever churn model you end up using, testing, learning and reiterating the model is crucial. Just like any other marketing tactic you employ, taking an agile approach and trying new things will help you realise if something’s not working and adjust.
Retention, retention, retention
Now that you know the customers who are likely to leave you can start targeting them with retention campaigns. You might find that you have to split the ‘likely to churn’ group into smaller sub-groups, as different churn indicators may require different tactics to keep them as customers.
It also pays to establish a control group who don’t receive any retention comms at all. This will confirm that it’s actually your marketing that’s luring them back, and not any other factors.
Beyond churn modeling
Knowing that customers are churning, the behaviours that point to churn, and establishing retention programmes is only part of a wider picture – one which identifies in more detail how customers feel about your products or services, which may be leading to churn in the first place.
An even deeper understanding can be established through NPS (Net Promoter Score) or exit surveys targeted at those who have churned. While uptake may be low, disgruntled customers can give you a glimpse into where you’re missing the mark on customer experience or available products, as well as what your competitors are doing differently (or better) and pointers on what you could do to improve.
As with any type of marketing, churn shouldn’t be kept as an isolated tactic, but rather incorporated into the wider customer lifecycle strategy, and as such continually tested and adapted to provide the best outcomes.
Key steps to combat churn
Once you’ve decided to undertake a project like churn, it’s good to break it down into manageable chunks. Here’s our step-by-step list to get you started:
Identify and understand common behavioural metrics that can identify that a customer is likely to churn
Build a churn model that will provide you with the metrics best suited to your organization or customer group
Test these metrics to determine impact on churn, and refine the model
Identify at-risk customers prior to them discontinuing business with your organization
Measure the effectiveness of the churn model and adjust as necessary