I recently read yet another very informative blog post from Corey Eridon over at Hubspot, entitled The Problem with Predictive Analytics. Along with an explanation of predictive analytics and its increasing use in marketing, her analysis got me thinking about the use of predictive analytics in customer success. If marketing is using predictive analytics wrong, then chances are the nascent industry of CS is likely doing it too – or rather placing too much emphasis on them.
Interestingly, Hubspot innovated the notion of predictive analytics for customer success with the introduction of the Customer Happiness Index (CHI). As highlighted by David Skok, Hubspot found that although useful, the CHI was not a good predictor of churn. Just because someone was using the product, they were not necessarily happy with it. But Hubspot was still on the right track in their thinking – we need to find a way to measure customer success happiness to be able to find unhappy customers and take corrective action before it’s too late. A possible solution: quantify business outcomes.
I decided to parallel Corey’s insights into dissecting the problem, the approach and how predictive analytics fit into the picture from a customer success perspective.
Let’s imagine that you have 10 customers who have not logged into your service for the last month, all of who have their contracts renewing this quarter. You might assume that they are at risk to churn because other customers who have exemplified the same behavior pattern all cancelled at renewal time. As Corey highlights, if this is your thinking and your expectation, then you’re using predictive analytics. Predictive analytics provide a means of taking data on past performance and using it to predict future performance. It is, however just that – a prediction – more grounded than the magic 8 ball but not necessarily the complete picture either.
Where the Problem Lies
As Corey points out with marketing – customer success isn’t a hard science either. Within our 10 customers, there are anomalies. Not all customers are exactly the same and not all of them use your service in the same way. 3 of those 10 customers might only login to your service once a month – this is actually a normal usage pattern for them. So without intimately knowing your customers or really having a relationship then everything else identifies them as a churn risk. Something could’ve also changed in the population – your customer champion left – which had very little to do with the usage patterns and their association with churn.
Accounting for the Knowns and the Unknowns
In our example, we identified insights that customer success should have known, therefore reducing the probability of churn down to 7. So let’s take those 7 into account as an unknown. But how can you when unknown is well, unknown. “You can try to anticipate them, but there’s no way you can account for all of them.” (well stated Corey). And since you can’t anticipate every unknown, you have to account for this when you are leaning on predictive analytics. Using predictive analytics is flawed when you use it as a basis for your decisions with the expectation of 100% accuracy.
Predictive Analytics don’t replace Relationships
In our example, there is a likely probability that the remaining 7 customers might churn. A risk has been identified using predictive analytics but then what? What is the triage and mitigation process? How is it defined and managed?
A better thermometer is useful but useless on it’s own. You need to establish a system and have a tool to enable or support it – a tool won’t solve your problems. As a customer success organization your system should identify how to triage and mitigate those 7 customers and then enable you to define and manage the process to achieve the desired outcomes.
As predictive analytics shouldn’t be used to make concrete predictions, tools that only provide predictive analytics shouldn’t be considered a solution to reducing churn (or managing the entire customer lifecycle). It is always the customer relationship that matters most.
Thanks Corey for guiding this marketer in the right direction and for providing a leverage point for applying marketing learnings to customer success.
What are your experiences with predictive analytics?