The Natero CSM solution includes a customer health score which is configurable. The meaning of the health score can vary based on how you want to use it. For example, if you are laser-focused on the activity level of your customers or on certain feature usage, you can customize the health score to focus on that aspect.
Not for identifying specific outcomes
But if you consider health score as an overall indicator of customer health, and that's how you should configure it. This is in contrast to trying to predict specific outcomes, such as who is likely to churn, expand or convert. If identifying those potential outcomes is the goal (and it is for most SaaS vendors), then a predictive analytics solution will be far more accurate approach than using a health score.
This is because separate machine learning models can each focus on a single outcome, while using a health score to identify multiple potential outcomes is unrealistic. For example, if you configure a health score such that a low score indicates potential churn, you can't expect that a high health score indicates an expansion opportunity. The indicators for each of these outcomes are not necessarily based on a common scale or the same set of factors. Moreover, machine learning models evaluate hundreds to thousands of factors to determine what the best signals are to predict a given outcome, while health score configurations incorporate a handful of factors.
While health scores are not good indicators of specific outcomes (e.g. likely to churn or expand), they are useful in understanding a customer’s overall health, as well as the health trends for an account, portfolio of accounts or the overall business.
Health scores are typically configured using a variety of other metrics, such as activity, feature usage, support history, NPS scores, financial history, CSM subjective scoring, etc. Given the varied nature of SaaS products and their usage models, its important that the CSM solution is flexible in how it calculates the health score. CSM teams should be able to select and weight the fundamental metrics that compose the overall health score, so it can align closely with their business model and product usage characteristics.
Decide key indicators of health
A good way to configure health scores is to start by deciding which of the available inputs to the score are the best indicators of customer health. In some cases it may involve activity levels or feature usage, while the health score for products that don't involve lots of user interaction may be better measured by NPS scores or other non-usage metrics.
You should think about which of the configurable inputs to a health score are most important for your business and most closely correlate with account health. For example, if you have a fairly new product and you get lots of tickets from your customers, this may not be indicative that they are unhealthy. They may be in love with your product, but based on its maturity level, early customers have lots of questions/feedback. In this case, you might not weight support tickets highly, or you might set a high threshold for how many tickets are an indicator of poor health.
Start from the accounts you know
Once you've decided which factors are most important to determine the health of your customers, its time to configure your health score accordingly, and weight the individual inputs. One way to do this is by looking at some of your accounts that you know are healthy, as well as some that seem to be struggling. You can get a sense for what constitutes good or poor metrics for each of the inputs you are evaluating.
Adjust as needed
Lastly, you should look at the results of your health score configuration and see if it is representative of some of your well understood accounts, both healthy and not. If you find some accounts where the health score doesn't seem to reflect your subjective view of the account, look at the various metrics for this account to see if some adjustment is required.
There's no science to configuring a health score, and some accounts may not perfectly fit whatever model you come up with. But with tweaking over time, you should be able to come up with something that does an effective job of telling you how your customers are doing.