Using Machine Learning for Customer Success

Machine Learning for Customer Success

The bread and butter of a great Customer Success software platform is identifying which customers need your attention, whether it’s to prevent churn, improve adoption, or suggest an upgrade at just the right time.

But how can software help CSMs predict customer behavior?

Let's start with a quick primer on predictive analytics.

Predictive Analytics

Predictive analytics describes a broad set of statistical and data mining techniques that attempt to predict the future by analyzing data from the past.

Linear Regression

One of the most common techniques used in predictive analytics is linear regression, which works by trying to understand the relationship between variables.

Consider the advertising example, where you want to determine the impact of increased advertising spend on sales. You start with an assumption (i.e. more advertising creates more sales) and analyze the effect that changing one variable has on the other.

You then use this analysis to predict the sales level that would be achieved for a given amount of advertising spent.

Linear regression starts with an assumption and tests the relationship between cause and effect to develop a model that can predict the change in one variable based on the change in other variables.

But what if you don’t know what causes something to happen?

For example, what are the causes or factors behind customers churning or buying more?

That's where machine learning comes in.

Machine Learning

Machine learning is a form of predictive analytics, but it turns the process upside down. Rather than start with an assumption of cause and effect, machine learning starts with an outcome and lets a computer uncover the causes that are driving this particular outcome.

It evaluates hundreds or thousands of possible factors, including complex interactions between those factors, to determine the best signals for a given outcome.

Machine Learning for Customer Success

In the case of Customer Success, machine learning models can be created to predict which customers are likely to churn, which are likely to upgrade or buy more and which are likely to convert from a free product or trial to a paid account.

The models can be built using the customer data collected by a CSM solution, including product usage, CRM, billing, support, etc. Machine learning models can even show a probability that a particular outcome (e.g. churn) will occur for each account.

Machine learning is true data science, analyzing hundreds, sometimes thousands of factors and the relative weight of each of those factors. These models become increasingly accurate with time, and automatically adapt to any changes in your customers behavior.

Machine Learning vs. Rules

In contrast, rule-based alerts use a straightforward ‘if this then that’ clause.

For example: "if a user doesn't login for two weeks, then send me an alert."

These rules are based on intuition or personal experience, and each can address a a limited number of factors.

Given the clear advantages of machine learning, do you still need rules? Yes.

Rule-based alerts are still valuable, but for different purposes. For starters, machine learning needs data to build these models. Some customers have historical data, so these models can be created fairly quickly. For others, it will take time to get valuable insight. In the interim, rule-based alerts will help.

The real advantage of rule-based alerts is helping CSMs drive customer behavior.

For example, are your customers meeting key milestones in a timely way? Are they progressing through a critical onboarding stage on schedule? Is product adoption on track with their users?

Rule-based alerts can be used to track customer progression through these lifecycle events. These are commonly called "engagement alerts" because they tell you which customers to engage with to keep them on the path to success.

Engagement Alerts

Engagement alerts are most powerful when used proactively. Let's say you have one feature of your product that’s greatly enhanced when used in conjunction with another. You can send a congratulatory message to users of the first feature, while suggesting the second.

You can notify users of a particular feature that it’s changing, and send them a link to a how-to video. Or use rules to track various deadlines, such as notifying you and the customer of upcoming renewals or a trial that’s ending soon.

Bottom Line

To truly anticipate customer behavior you'll need both rule-based alerts and machine learning.

Machine learning provides a powerful predictive engine to identify which customers need your attention, while rule-based alerts empower your CSMs to monitor customer lifecycle progression, and proactively engage customers.

Right now, there’s only one Customer Success platform that has both options. And you guessed it, it's Natero.

- - - -

In the next installment, we'll discuss how predictive analytics helps Customer Success teams and why serious Customer Success Software should have both predictive analytics and rule-based capabilities.

Want to learn more? Grab our eBook: Predictive Analytics for Customer Success.

Data-driven Customer Success teams run on Natero. LEARN MORE