Leveraging the Power of Predictive Analytics to Control Churn
PAs and CS
The subscription economy has forced a huge shift in the way companies approach their customers.
The game is no longer a race to acquire new customers, but rather to hone in on retention and upsells. A Pacific Crest 2014 Private Software-as-a-Service (SaaS) survey found that it costs $0.18 for every $1 of upsell revenue, and $0.12 for every $1 of renewal revenue.
The same study found that it costs $1.07 to acquire every $1 of revenue from a new customer.
This doesn’t take into account the intangible marketing benefits of happy customers — referrals, success stories, and testimonials — that help you grow your business.
The Growth of Customer Success
There is an accelerating recognition of the value in customer retention, as evidenced by the number of SaaS companies that now talk about their churn metrics.
This is one of the reasons companies are making huge investments in Customer Success. Currently, there are more than 200,000 job openings nationwide for Customer Success, rivaling established functions like sales and marketing.
It has overtaken account management as the preferred method of maintaining customer relationships, and is swiftly developing into a science.
What can companies do to prevent churn, ensure renewals and drive upsells?
The first step is to know which customers are likely to churn, and which ones are likely to renew and/or expand. This sets the stage for predictive analytics to become the hero.
Tapping Predictive Analytics
Predictive analytics includes a variety of statistical techniques from predictive modeling, machine learning and data mining that analyze current and historical facts to make predictions about future or otherwise unknown events.
We have been using predictive analytics for many years now. But the technologies have become more sophisticated with the advent of advanced statistical modeling and computation techniques.
In fact, it is far more prevalent in our lives than we realize. For example, your credit score is an application of predictive analytics. Credit scoring looks at historical and current facts — your payment trends, account balances, application history and more — to make a prediction about the future; that is, the likelihood of you making future credit payments on time.
When it comes to predicting churn, there are many factors that indicate churn: drops in product usage, increases in support ticket volume, degrading sentiment in customer communications. With a wealth of data points and advanced models that have been honed over the years, it would seem that predicting churn would be a slam-dunk.
But as any Customer Success professional will tell you, this is not the case at all.
What’s Wrong With Predictive Analytics?
The credit score example reveals one of the limitations of predictive analytics. Over the course of our lives, every one of us has had at least one time in which our “credit score” did not reflect our future credit worthiness.
A student with his first job will not have it easy on credit applications because his credit score will take time to reflect the change in his circumstances. Someone looking for a mortgage, personal or business loan may be rejected because there have been “too many recent applications” on her record, even though there is no material change in her ability to pay.
The examples are endless.
Another problem is a mismatch between the perspectives of the creator of the model, typically a data analyst/scientist or a statistician, and the user of the model, who is the average business user. Let me provide an example from my experience.
A few years ago, I signed up with a then-leading vendor of predictive lead scoring. Their promise was that they would analyze a list of my existing customers, and predict the next set of customers.
A week later, they came back with a list of parameters that they thought were representative of our customers, one of which was that the prospect had Secure Sockets Layer (SSL) turned on for their corporate website.
As a business user, that made no sense, and made these parameters less credible. The modelers were probably using the presence of SSL as a proxy to identify these prospects as SaaS vendors themselves. I’m not sure how they came to this finding as a critical predictor of my future customers, but that is exactly my point.
Does this mean that predictive analytics do not work and should be discarded? Absolutely not.
That would be akin to throwing out the baby with the bath water. Just like making credit decisions based on the credit score, the key is to recognize the limitations of predictive analytics, and adjust for the elements that could lead to decision-making biases.
Why ‘Variables’ Are Problematic
Predictive analytics are based on complex models that consider many “variables” that may or may not be independent. I listed some of these variables earlier as they related to predicting churn. There are three issues with this approach:
1. The model may not include all relevant variables, and more importantly, may not account for interdependencies between these variables.
2. The model may not be able to incorporate data, especially in siloed environments, because it is hard to get relevant data in real time.
3. As the model becomes more complex, it creates a false sense of security (the model is very complex, therefore it is likely to be right). The reality is that complexity often equates to fragility.
In addition to the above, the Black Swan theory (BST) can also come into play. The BST, in a nutshell, states that no predictive model can account for what you don’t know. Unfortunately, the unknown can have a significant, and possibly destabilizing, impact on the model.
Predictive Analytics Done Right
The question becomes, how can you leverage the power of predictive analytics to control churn? Here are four key steps that can help you achieve Customer Success with predictive analytics:
First, stop trying to reinvent the wheel.
You may have the most unique business in the world, but if you have a SaaS or subscription based product, there are several features that you will share with other SaaS vendors.
Every Customer Success automation (CSA) vendor has developed their own predictive models based on their knowledge and experiences with their customers. Some will be better than the others, but chances are that they have something that you can start using out of the box. There’s no need to create your model from scratch!
The key question to ask is if the model can recognize the nuances that are specific to your situation. In some instances, support tickets play a major role, in others, usage activity contributes in a big way, and in some, customer interactions could govern the health of the relationship.
You want to make sure that the solution you choose meets your specifications, and can be configured to reflect the specific characteristics of your product or service.
Second, realize that it is only a prediction, not a guarantee.
You know that every predictive model has its limitations. Nothing can substitute for the human judgment of your Customer Success professionals (as of now). At the very least, the model will help you narrow down the set of customers that are likely to (or not) churn.
You will still need your professionals to make the final judgment call. More importantly, your Customer Success solution will need to monitor your top tier customers to identify the false positives, the customers that were unlikely to churn as per the model, but may actually be on the way out.
Third, move from “what-if” to “what next”.
Ideally, your CSA solution should go beyond just predicting churn. It should tell you what to do next, with automated triggers and playbooks, if a customer is identified as likely to churn.
For example, an increase in churn rate should result in an automatic notification to the Customer Success manager, and possibly a personalized email or call to the customer. The intervention may need to be even more drastic if the contract is coming up for renewal in the near future.
Fourth, make it a cooperative, closed learning loop.
You have to be constantly testing the results to identify the variances. Some CSA products allow you to configure the predictive model by changing the weight of the variables. It’s more likely that you will need to work with your vendor to highlight the variances, and have them fine-tune their model so that you can derive the benefits of the improvements to this model.
Predictive analytics are valuable tools in the hands of Customer Success professionals for reasons beyond just churn prevention.
However, it’s only a part, albeit a vital one, of the complete toolset that is available to them. It is meant to make life easier and complement a CSM’s judgment, but not replace it.
When used correctly, and in full cognizance of its capabilities and limitations, predictive analytics can indeed be a powerful tool.
It’s the closest Customer Success professionals will ever get to a crystal ball.
Read the original article on CMSWire here.