They are highly informed, they have soaring expectations and the global marketplace has given rise to an abundance of choice. This flow of customers leaving a product or business is known as customer churn. And for business leaders, stopping the flow, or at the very least reducing it, has never been more critical.
Identifying the issues, the levers to pull, and the metrics to track improvements is no longer a nice to have, it’s a necessity. And yet, there are still many companies that don’t do it well, or that have a very narrow approach to churn analysis.
Modest gains in improving customer churn can add significant additional revenue. Yes, there are costs associated with incentivising customers to stay (discounts, R&D, new product development, management costs etc.), however we can quickly calculate the high potential return of keeping existing customers.
Let's take a B2C Company with 100,000 customers with an average monthly bill of $50. This company has an existing churn rate of 12%. See how to calculate below:
If we can reduce the Churn rate to 10% we will retain 2,000 more customers. These 2,000 customers pay an average $50 per month.
This delivers $1.2million in additional revenue in the first year.
In addition, these gains are not once off. As your business grows, the benefit of lower churn rates grows exponentially. Year on year, 2,000 additional customers kept with many staying longer than 1 year will lead to millions in additional revenue.
Many of these customers will leave on their own, so we need to encourage them to stay with incentives or by improving the areas that prompted them to leave. This requires a customer churn analysis to understand the why?
Predicting customer churn has been a staple of many B2C companies with large customer datasets - mainly in the telecommunications, insurance, banking and utilities sectors. Within these organisations, Churn analysis projects have become a core task for their Data / Analytics teams. From these corporate b2C industries, a traditional approach has emerged that has become the default Churn Analysis template.
Using their customer databases, data teams build predictive churn models (A customer churn risk score attached to a customer number, based on who are most likely to leave) and then present insights back to stakeholders. These models then approved and implemented into CRM databases. They are primarily used for monitoring and for optimising direct marketing efforts.
It only treats the symptoms
These predictive models use a variety of data (account info, demographics, product holdings and behavioural) to assess a customer’s likelihood to leave. But that is kind of the only thing they do, they measure the symptoms. For example: the volume of calls by a customer to a service centre in the past 3 months is a strong predictor that a customer will leave, but it fails to identify the underlying problem.
Even with additional information such as type of call, duration, outcome, etc are included, we can start to identify the key issues. But acting on these insights falls outside the remit of the project - an additional project would be required. This creates inefficiencies, multiple projects disconnected to solve an issue that may not be the most important issue to solve first. Data analytics can unearth great insights but limited scope to solve the right problems renders many projects ineffective.
A lack of qualitative insights
As the definition goes, quantitative data answers the “what” and “how”, qualitative data answers the “why”. The traditional approach was established at a time when collecting customer feedback directly (through interviews, surveys, customer testing) was relatively expensive and combining this qualitative data with other sources was difficult and messy. But without exploring the “why” perspective you are missing at least half the pieces in your churn puzzle.
It is not news to marketers that customers don't behave rationally. There is a subjective and emotional aspect to their decision making, and this rings true when they are deciding whether or not to leave your business. Yes, quantitative data is growing exponentially and perhaps soon we will have the technology to quantify this subjective element. But for now, qualitative data needs to be embraced to create a truly holistic understanding of why valuable customers are leaving.
Data Analytics advancements
We now all live in the data age. The abundance of data that can now be applied for problem solving is seemingly endless. Coupled with new technologies available to us like cloud analytics platforms and data visualisation tools, we can now provide insight into customer behaviour faster and easier than ever before.
These opportunities were not available when the traditional approach was established. We can now leverage both customer and product databases (structured data) and embrace the new wave of video, voice and text data (unstructured data) to help direct effective churn solutions.
Example: Creative Analytics for customer service improvements
Companies with larger customer care needs such as Health Insurers are now deploying topic clustering (a natural language processing machine learning technique, also known as NLP) to quickly identify key topics discussed within calls that are classified for low customer satisfaction. These insights can then help streamline service to more digital resources or to adjust service offerings to avoid any churn inducing pain points. These improvements in customer experience directly impact churn and also reduce service call cost in one move.
Each of these areas highlight the need to create a new approach to analysing customers leaving your business. Churn is a customer-centric problem, so it’s in an organisation’s best interest to approach it this way. Pursuing solutions needs a robust approach, one that aims to not only predict, but prevent.
Our approach to churn centres on two things, understanding the why customers are leaving and in turn, creating validated solutions to address the root cause(s) of churn. We’re not removing data analytics, rather expanding its role into an enabling toolkit to find and refine the most effective solution.
There will be three stages to our approach; Discovery, Develop & Test a MVP and Embed Solution. We will put qualitative and quantitative data at the heart of our analysis and will strive to solve the most pressing driver for why customers are leaving.
This is designed to be an overview with key elements of each stage identified. There is much more detail used in practice that would stray too far into the weeds for this article. Details on different data science techniques, designing objective customer scripting to avoid bias, defining robust control groups, vanity metrics, etc may be addressed in future articles for more in depth perspectives.
To ensure the right problem is being solved, this approach requires a Discovery Analysis, that borrows some key principles from Design Thinking.
A Discovery Analysis is a short (often two week) sprint that aims to seek out the root cause of an issue - Why customers are leaving your business - and then identifies which areas to pursue based on value return and the difficulty/cost of implementing the solution.
What techniques should be used? A combination. Both qualitative and quantitative approaches are used to create a holistic view of customer churn. In each approach we keep top of mind that we need to find the root cause behind customers leaving your business.
Discovery Tools & Techniques
Qualitative:
Quantitative:
Your discovery analysis may uncover a series of associated issues or maybe a single underlying cause. While you might consider them all to be important, your discovery should identify the most promising problem area where you can make the most impact.
Below is a list of example causes that appear general, but each require bespoke solutions to fit organisational and industry context:
Following the discovery process you should have identified one key area to pursue. This is where the approach pivots from identifying to tackling the underlying causes/drivers of customer churn.
Before a solution can be implemented, it must be tested and refined. Without testing possible solutions, you run the risk of investing time and money into ideas that when implemented, are ineffective.
This is where a Minimal Viable Product / prototyping comes in. A proposed solution is deployed (a new process, a mock up of an improved product, a new pricing model, a revamped loyalty program) to a subset of customers who’ve been identified as likely to give feedback. Remember that no solution is optimal in a project plan, because effective solutions need testing and refinement to determine what is working and what is not working.
Example: Tackling early life churn for Broadband Provider (200k Customers)
Discovery: Two cohorts that drive customer churn were identified: early life churners who have a tenure of 0-3 months, and re-contracting customers at 12 month tenure. Through a business engagement, separate plans for changing the re-contracting process were discovered. The team pivoted to improving early-life churners.
MVP: The team chose to prototype an improved onboarding experience. They created a simplified process that makes the web portal and app easier to use for new customers. It was hypothesised that the new onboarding process will lead to faster installations and less service centre enquiries.
Impact measured: By resolving these customer problems, early-life churn was reduced by 80% in the testing environment with the unseen benefit of reducing the percentage of new customers who miss first bill payments from 15% to 3%. It is understood that these results may not be replicated when fully implemented, however even a reduction of 40% would return significant value.
Selecting the most impactful solution, rapidly prototyping and testing a MVP is not a novel concept. However, within analytics and especially for churn analysis these concepts have yet to become the norm. The traditional approach is applied and solutions are implemented without testing, leaving many churn analysis projects with underwhelming results.
Measuring your MVP Solution
Effective measurement will help to refine your solution’s impact. What is working? What is not working? Will this change customer behaviour? Many of the MVP metrics need to be specific to your solution and its development stage. All good metrics share key some attributes, here are some below:
They are action-focused - How do my decisions change when this metric changes? Is there a design decision for the MVP that we need to choose between?
They are comparative - Does this tell me which is better A or B? What is the difference? Splitting testers into two groups is not robust enough - You need to understand if the groups have an equal mixture of customer needs. Effective comparison needs to overcome some subjective bias for the best option to be selected.
They are behavioural - Is this metric this tracking actual behaviour or reported behaviour? The difference between what we report we do and what we actually do can vary enormously and these differences may hold the insight to refining your churn solution. Example of observing customers trailing a new process vs asking them their experiences.
They are understandable - the metrics you use need to be explainable to everyone. As you move into implementation, eventually your testing results will need to be explained to a wider non-technical audience. Complex metrics can often be broken down to easier to understand a subset of metrics.
If the MVP solution tests successfully, the solution should be adopted across the wider organisation. Every organisation’s attitude towards change differs, so prior to MVP testing it is important to consider any foreseeable implementation problems. There will always be some challenges to overcome in implementing your solution, but these are some essential components to consider:
Automate for scale - Many MVPs can potentially fail to scale if they need manual inputs. Avoid this by automating all processes - especially reporting - to remove unnecessary waste. This may not be always achievable at the beginning of implementation, but should be a goal as you refine your churn solution.
Create alignment - Stakeholder engagement and alignment is key to successful implementation. However, it's likely you will encounter some form of cultural resistance, and overcoming these issues can often be overlooked. However, a combined approach of inclusivity, workshops, demos and training can help a smooth implementation.
Continuous solution measurement and refinement - Implementing your solution is a massive milestone, but as with the MVP stage, you need to continuously measure your solution’s impact. What is working? What needs refinement? How can you measure value return? Compare results with control groups and ensure any results have statistical significance.
Capture customer feedback - When collecting customer feedback, it is important to measure both what they do and what they say they do. Eg. measuring how a customer interacts through a new UX of a website with Google Analytics may differ significantly from how they respond in a survey.
Okay so we’ve covered a lot of ground - Time to recap on our approach:
This approach is broad, but can be tailored more to the bespoke industry context, customer dynamic and business model. It looks beyond data analytics as a siloed approach data as an enabler for effective problem solving.
We now have easy access to customer’s perception of our products from digital touch-points and can collect quality customer feedback at a much lower cost. These changing dynamics are fuelled by better machine learning technologies, altogether contributing to an enormous opportunity to change our approach to customer churn analysis. So now, we can reap the benefits and the significant value potential of keeping customers around for a little longer.
