Predict Customer Churn More Accurately with 4 Key Data Points
As more and more businesses move to models that are subscription pricing on a monthly or annual basis, the importance of managing customer churn (also called customer attrition) continues to increase. Once a customer base has been built, any challenges with attrition, no matter how small, can impact top-line revenue more than the entire new account function.
However, being able to predict customer churn – at an account by account level – has remained an elusive challenge. The value is clear: any ability to identify which accounts are most at risk of churn, long before renewal, allows customer success teams to quickly rebuild relationships, engage executives, remove barriers to success, and highlight key points of value. This “fire fighting” approach, when done far in advance of renewal conversations, reduces churn rates significantly and therefore boosts both top-line and bottom-line revenue. The challenge is identifying which customers are at risk.
Better Algorithms or Better Data?
Early attempts to improve predictions of customer attrition focused almost exclusively on algorithms – machine learning and artificial intelligence approaches that looked through the data that was available on customers and tried to predict the likelihood of churn. This approach suffered from major limitations in the underlying data it used – they generally only worked with facts about the customer organization (such as size and industry), and sometimes added in product usage data (such as login counts).
Put simply: if the data available to a sophisticated machine learning algorithm doesn’t reflect the true reality of what’s happening, the algorithm will not be any good at making predictions. A common adage in machine learning circles is that 10% better data will outperform 100% better algorithms. This was the reality that faced early attempts to pin-point client attrition risk based on algorithms alone – customer facts and basic product usage are not predictive of churn for most organizations.
Data as a Guide to Reality
But what data was missing? The best path to finding what data points are needed is to look a reality first, identify the kind of problems you need to look for, and then find data points that give you a good indication of those problems.
Attrition is almost always about perceived value. If the right people at the buying organization perceive that they are getting sufficient value from your offering, they will continue to pay for it. However, although this seems simple, it can be quite complex.
- Customers might be using your solution regularly, but are frustrated by it and this negativity clouds their perception of value
- Customers might be using your solution regularly, but not in a way that creates value
- Customers might be creating value with your solution but not aware of the value that they are creating
- Customers might be creating value with your solution, but the people creating value are not the same as the people who sign the contract – and they are not communicating with each other
It turns out that there are three important pillars that are needed in order to understand if a customer is likely to churn.
- Are they using your solution
- Do those who are using it perceive it as valuable
- Do those who sign the contract perceive it as valuable
The first of these is often product-related data, and use of the product can be proxied well by looking at who is logging in and how often. More advanced scenarios can identify specific use cases or outcomes based on the product features used.
The second, and third, however, are more interesting. Understanding whether the users are perceiving value, or, more importantly, the executives in charge of the contract are perceiving value is tricky. Here, the art of customer success comes into play, as these are mainly determined through one-on-one conversations and relationships.
The value of building a predictive model for customer attrition is mainly in identifying customer churn risk where you don’t already know that a risk exists. You can assume that at any customer where you have a strong customer success relationship with the right person, you have already had the crucial conversations. Because of this, you will already know of any risk that exists at that customer. The algorithmic model then needs to identify the customer accounts where you don’t have the right relationships, and therefore would not know if any issues in value perception existed.
Key (New) Data Points For Churn Model
To tackle this challenge of value perception, churn models need to incorporate a few new data points to be able to identify risk. These are data points that identify where relationships with users or executives should exist, but don’t. The following four data points are the most relevant in most customer success organizations:
- Strong Relationships with Users: the number of relationships that are either strong or very strong, with people identified as the front-line users of your solution
- Relationships with Executives: the number of relationships with executives at the level of director or above
- Last Contact with Users: the number of days since a member of your team has engaged with the users
- Last Contact with Execs: the number of days since a member of your team has engaged with the executives
These four new data points immediately add a “relationship” dimension to attrition models. Assuming that in any strong relationship with a CS team member, problems would have been discussed, this allows attrition models – whether based on simple heuristics or artificial intelligence – to flag accounts that are likely at risk.
Customer Attrition Models Require an Intelligent Data Foundation
Accurate models require accurate data, and models for attrition are no exception. If the data does not reflect the customer reality, the model won’t help you predict churn, regardless of how sophisticated the algorithm is. Perception of value, with the key stake-holders is the critical part of customer retention, and the best data for that is whether your team has strong relationships with the right people at each customer.
An intelligent data foundation like Nudge is needed to achieve this by ensuring that every data point on every customer is collected and measured, objectively, and added to Salesforce, Gainsight or whichever system you use for modeling customers and predicting attrition. Facts such as:
- Every contact your team is interacting with at the customer
- The strength of each relationship
- The decision-making level of each individual at the customer
- The functional role of each individual
- The history of communication between your team and theirs
With this core foundation, you are quickly able to improve your churn predictions and focus quickly on any deals that are at risk, long before a renewal conversation occurs. Managing customer retention by being proactive about reducing churn is the shortest path to increasing revenue and profitability.