SaaS: How To Predict Customer Churn With App Usage Data

Churn kills SaaS companies. No matter how many growth hackers you hire into the marketing team or how many hustlers you hire into your sales team, eventually a ‘leaky funnel’ will at best, cause your company growth to plateau and at worst, cause your revenue to actually decline.

This is because eventually, you saturate out your niche markets and distribution channels. People begin to ignore your marketing messages as they’ve already “been there, done that, didn’t work for me…”.

There is also the reality that for most SaaS companies, the cost to acquire a new customer is not returned until the customer has been a subscriber for a few months (or longer). Most companies invest big into customer acquisition because they hope that the customer will stay around for years to come, paying their fees each month with the minimal of costs or effort involved.

But if we don’t solve churn, growth stalls. And to solve churn, we need to intercept before it happens.

Smart SaaS Companies Solve Churn Before It Happens

Thats why the smart SaaS companies solve churn early. They learn to catch it before it has chance to set roots and cost the company lost revenue every month. These “smart” SaaS companies are ones you’ve heard of too: Salesforce, Huddle, Box… (just a handful of the big name B2B SaaS companies who have had established customer success management departments for quite some time).

It’s no coincidence that you’ve heard of these companies and they happen to be huge pioneers in recognising and proactively combatting the effects of churn.

You’ve heard of these guys because:

  • They’re still around and growing fast, while their competitors who ignored churn have now closed down or plateaued into mediocrity
  • Their customers receive such great ROI’s from using their product that they can’t wait to tell everyone they meet at meetups, conferences and social media

Nearly all SaaS companies around today know they need to report on churn. It’s an important metric that tells your investors how well you’re doing (or not) and gives you a baseline to toast beers over at the end of each month.

But what the really smart SaaS companies know is that churn rate is actually the end of the story. It’s a trailing metric – once you’ve measured it, you’re powerless to affect it. They don’t just report on churn. They predict churn and do something about it!

These pioneering SaaS companies realised that a customer churns when they aren’t receiving success as a result of subscribing to the product anymore. And they noticed that when a customer’s success starts to decline, this correlates to changes and patterns in how they use the product.

Product Usage Correlates To Churn

By measuring how customers use your app, you can then create models to predict when a customer may be at risk of churning. On a super basic level, app usage is inversely proportioonal to churn. When app usage goes down, churn likelihood goes up.

Hold On!!! Yes I know this sounds like I just jumped into the madhouse of mathematics and you might be tempted to ‘switch off’ at this stage and forward this article to the “Woman in the corner who does SQL and spreadsheet stuff” but please… stick with me for a few more minutes! I promise you, this article has been written for people who are not data scientists in any way. This article is for people who understand people, and want to know how people’s behaviour can be measured to predict churn and prevent lost revenue.

With the right tools (OK here’s a plug for our tool, Trakio) this process can be made simpler and accessible to most customer success managers. In fact, once you’re setup and you understand what’s going on, the complicated stuff pretty much just runs on autopilot.

I’m going to run through each of the main signals within your app usage data that can predict churn. What I’m not going to do is talk about pattern recognition and machine learning algorithms.

That stuff sounds fancy on marketing websites and data sheets, and I’m sure you’ll hear that a lot once you go out looking to buy a tech platform for your own company (I mean just say it out loud with me, al-gorr-rhythm…) I’m probably guilty of shamelessly dropping those bombs around a few conversations myself.

But in practice, as a customer success manager who wants to get started in predicting churn signals in your customer base, you need to understand the fundamental theory first.

Whether you buy the tech from Trakio or another vendor, if you don’t know how this stuff connects together and how to get the most out of it in your company from your own data sets, then you’re at risk of wasting tens or even hundreds of thousands of dollars on technology, manpower and time.

Here are 8 key principles for using app usage data to predict your customer churn.

1. Segmenting Based On Usage

In most case cases, the simplest way to use usage data is to generate segments of your customers based on behaviour. These aren’t static lists like in your email marketing software – they are dynamic segments that are constantly updating in the background.

What this means is that users “move into” and “move out of” these segments, depending on the events they perform within your app.

For example, a Project Management company (think Basecamp) might create a segment of “Recently Engaged Users” which would include users who have used their ‘Reports Module’ and ‘Projects Module’ in the last week:

Screen Shot 2014-09-10 at 07.45.10

This means if a user stops using one of the modules, they will eventually leave that segment (once it has been longer than 7 days since they used one of the modules).

By combining multiple rules regarding the different events a user has, and hasn’t, performed in recent time periods, it’s possible to segment users this way.

Once customers are in these dynamic lists, it makes it easy to:

  • Quickly categorise accounts and customers into ‘Healthy, Worrying or dangerous’ based on the events they have, or haven’t, performed recently
  • Send team alerts when customers move in or out of a segment
  • Send automated emails, push notifications or SMS messages to customers if they move in or out of a segment
  • Send one-off newsletters to a hyper targeted list of customers who match that criteria *right now*

Creating dynamic segments is one of the first things you’ll want to do when you get setup with a customer analytics platform. The great news is, once you get these setup you can pretty much just leave them running on auto-pilot. As you learn more about your customer profiles and usage, you can come back and tweak the rules, make them more detailed and contextual etc.

2. Using ‘Sign In’ Events

When a customer comes to your application, they’ll trigger a “sign in” or a “session” event. While some CSMs disagree about the importance of tracking this behaviour, I think without doubt that having the customers mindshare on your product is a good thing.

Sure, “sign in” events and counts don’t guarantee the account is achieving success from your product. However, in most circumstances, if they aren’t coming and signing into your product at all on a daily/weekly/basis then it’s certainly an indicator that they aren’t achieving success.

  • No sign in events = almost definitely not achieving success
  • Lots of sign in events = it’s possible to achieve success

You should avoid creating segments that indicate a ‘Healthy’ account based on sign in events alone. However, you should definitely create segments for ‘At Risk’ accounts who have not signed in ‘X number of times recently’.

An exception to this rule: does your product require a sign-in to get the most out of the product? In particular, some tools like developer tools require a setup by the customer with an API and then are pretty much ‘set and forget’. In these cases we would track activity on the API rather than on the individual users signing in. More on this later.

It’s up to you to transfer what you know about your product into the platform so that everything is in context.

3. Engaging With Key Features

Although sign in events carry some exceptions around use cases in indicating customer success, usage of key features is always a solid indicator of customer success. Think about it: how can a customer be receiving value form your product if they aren’t even using it’s key features?

However, the one small trip up here is in knowing what your key features actually are. In most products, feature creep can take over. Just because you have 13 features in your app, your data might actually reveal that only 6 of those are key to the customer experience and required for the customer to achieve huge success.

Or it could even be that different groups of customers value different features. When building your dynamic segments, you could use property matching to separate out these user types and create different segment rules for each.

For example, companies with under 5 employees might find the “Projects” and “Tasks” the most important. But your large enterprise customers with 100 employees should also be using the “Executive Reports” module often.

4. Monitoring Trends vs Absolutes Counts

Deciding whether to segment a customer as “Healthy” or “Dangerous” based on whether they’ve engaged with a key feature less than or greater than 5 times seems a little arbitrary. Why 5 times and not 6?

First of all, using an absolute value is easier to explain and understand for most teams. You can transfer the knowledge you already have as a team and combine it with additional insights from within your analytics to come up with this ‘tipping point’ value.

However, if you only focus on weekly trends it can be possible to miss the bigger picture. Dropping in usage by 3% a month won’t trigger any alarm bells – but after 12 months that’s a 31% drop in usage!

There’s a story about boiling a frog in water without the frog freaking out: if you slowly increase the temperature by 1 degree at a time, over a long enough period, the frog never notices the temperature increase!

It’s important to keep an eye on the longer term trends around usage to get a full picture of usage. Look at the 6 and 12 month picture to spot any steady declines – no matter how gradual – as a sign that the customer is slowly losing engagement with the product.

Tip: you can also analyse these trends to discover refined values for your segmenting.

5. App Usage During Onboarding

Onboarding periods are the most critical as an indicator for the customers success. As many customer success managers will tell you, most customers who churn within the first 6 months had a poor onboarding experience.

During this critical time – which includes any free trial period, it’s important for you to have the data analysed to a highly granular level. You’re looking for signals that users are failing to get to the “Aha!” moments. These are the points along the onboarding workflow where the customer “gets it” and each employee within their team “gets it” too.

You should setup segments and alerts for users who don’t get to these stages and “drop off” in your funnel. This could be a segment of users who “Have imported their subscriber list, but have not created their first campaign newsletter”.

You probably already have a model hypothesis of your customer workflow during onboarding. The key is to setup multiple dynamic segments that describe milestones within this workflow so you can take action to push, and pull, users through who fail to meet those stages.

6. Did They Invite Their Team?

There’s a direct correlation for most apps that if a customer has a full team on your platform (i.e. full license adoption) and that team is active, then the customer is likely to be achieving success.

From this, you can assume the reverse is true.

If a user fails to invite their team – and their team don’t accept those invites and create accounts – then you should highlight out these accounts (using dynamic segmentation) and initiate workflows to get them all onboard (personal follow ups, automated emails etc.)

One area of caution though: if the customer has all of their team on the app (lets assume 10 people), it could be that not everyone of those 10 people needs to be using the product to achieve success.

Consider the situation where the customer has an office manager who comes into the product once a quarter to download their invoices or update their billing details. Or the IT Manager who was extremely active at the beginning to set the product up and train others in the company how to use it, but now has very little need to go in and use the product on a daily basis. Or the executive manager who only needs to login once a month to view the KPI dashboard.

You get the idea.

You should add some context to your segments by using “roles” within companies. This allows you to create different rules based on the expectations of that user within the company. Are they a manager? Admin? IT Manager?

For example, a newsletter application might setup a segment where “If anyone with role = `marketing manager` becomes inactive, then mark this account as `At Risk`”.

7. Tracking API Usage

Some products don’t rely on the individuals in a company to login and use their web app (or mobile app) in order to be successful with the product.

Consider developer tools and how they work. If an app developer needs to send emails, she would sign up to Sendgrid and spend maybe a week getting setup, reading through the documentation etc. Our customer analytics would show a highly active user who is getting through each of our “Aha!” moments in our onboarding flow.

But then once they are setup – they might never come back and login to the application.

In these cases, you should think beyond segmenting activities of the individual employees. You need to also log activities on the account itself – events triggered not by any individual but by the API or some background processes.

A good way to log API usage is to send an event each day with an aggregate count of how many calls were made on the API (this is how we track our own API usage at Trakio). You would then build a segment that monitored declining activity on the API to look for accounts that may not be achieving success anymore.

8. Tracking 2nd Generation App Usage

Building on from tracking API usage, you can also track ‘2nd Generation Events’. By this I mean tracking events performed by your customers customers.

So in our previous example of Sendgrid, they would start monitoring the open rates and click through rates that an account was achieving. Sure, an account might be sending thousands of emails a day, but if they are getting poor open rates or click through rates, then it could be an indicator that the company isn’t achieving success. In this case, the CSM could reach out and offer links to best practice guides about writing great email copy and subject lines.

If the customer improves their email performance overall, then they will become more successful as a company and increase their spending with send grid overtime (i.e. Upsell themselves).

Not all products are well suited to tracking 2nd generation events, however there is a strong case for it when possible. Companies selling a platform to conduct marketing campaigns might track the performance of those campaigns. Companies selling an ecommerce store platform might monitor the basket completion rates of their customers stores.

The key to remember is we want our data to inform us of our customers success, and that can be more than just how they engage with our web app.

And I’d Be A Terrible Founder If I Didn’t Also Add…

…that Trakio is a perfect platform to implement these tactics. We don’t replace your CRM or marketing automation platform (in fact we compliment them and to plug into them!).

What we do is compute, calculate and analyse all of your app usage data in real-time and show you the customers and accounts who need your attention, right now.

There’s a 14 day free trial and you can start piping data to us in within a few minutes if you’re already a user. There’s also a super simple developer API that will have you up and running within hours (some platforms take weeks!), so there’s no reason to not give us a try!

Discuss On Hacker News

Published by

Liam Gooding

Liam is the cofounder and CEO of Trakio. Previously an engineer, he writes about growing subscription companies using data-driven techniques and inside glimpses to Trakio's own growth journey. He wrote a book, "Growth Pirate!" which discusses data-driven growth strategies for startups.