ProfitWell Customer Health Scores: Likelihood-to-Churn (LTC)
What are ProfitWell Customer Health Scores?
ProfitWell's customer health scores predict how likely your customers are to churn.
We apply advanced machine learning models on your revenue and customer engagement data to generate a set of scores to predict how likely each customer is to churn that month.
Our health scores are 100% free and can be implemented without engineering help within minutes. To start, we'll need you to let us collect engagement data on your users. Get started now.
How do I use these scores?
Customer success teams are using PW Health Scores in two ways:
• Proactively rescue at-risk customers: Identify customers who aren’t engaged with your product, and are behaving similarily to your previously churned customers. Prioritize rebuilding these customer relationships before they churn.
• Upsell power users: Identify highly-engaged customers who display patterns similar to those who previously upgraded their plans or purchased more products. Proactively target these opportunities for expansionary revenue.
Where can I find these scores?
You can find these scores on each customer profile inside of ProfitWell.
Once you install the .js snippet, it will take up to 60 days for our models to build enough confidence in your health scores and display them on the profiles .
Additionally, in some cases our model will have insufficient data to calculate your LTC scores accurately. This can be due to a number of reasons, such not having a lot of historical churn data for us to use to train our model to generate accurate predictions. In these cases, you won't see a LTC score for a customer for that month, until they pass a reasonable threshold of accuracy in order to set you up as best as possible to take the appropriate actions on your customers.
How do I read these scores?
The score represents how likely the customer is to churn in the upcoming calendar month. Think of this as a range of probabilities that take all of your customers into account.
For example, if a customer's likelihood to churn is less than 1X, then that means the customer is less likely to churn when compared to all the other customers in this given month. If a customer's likeihood to churn is 2x, that means they are twice as likely to churn compared to the average user that month. The lower the number the better :)
How are these scores calculated?
The short of it is we use some fancy machine learning algorithms to look at your historical engagement and revenue data to predict churn.
However, you are probably here for the longer answer :)
We first analyze a number of historical data points such as customer log ins, page views, session length, MRR, age of account, plan, discounts, and more.
We then use machine learning models to analyze this data with advanced regression analysis to ultimately find correlations and trends in these variables and churn. These models make predictions about the customer's behavior in the upcoming month.
Each prediction is displayed as a probability— we take all the probabilities for each customer and then put them on an index. This is then displayed as a score on a given range of probabilities. The data is retrained each day, and a completely new/updated model is created each calendar month. This means that the scores will update every single day, and the overall prediction each month will get smarter than the last.
Your health scores will get more accurate over time, and adding support for custom variables and inputs in this model is on the roadmap for later this year.
How do I get started?
1. Set up your ProfitWell account in 3 minutes if you haven't already
2. Install the Engagement snippet here.
3. Good things come to those who wait :) Health scores will be displayed in your customer profiles 60 days after installing the ProfitWell Engagement snippet, as it takes a while to gather sufficient data for the algorithm. Scores become more accurate over time as the algorithm learns more about your customers.
More questions? Send us a note at firstname.lastname@example.org and we're more than happy to help!