Keeping customers active and engaged is essential for any business that relies on recurring revenue and repeat sales. Customer turnover—or “churn”—is costly, frustrating, and preventable. Carl Gold shows how to leverage data to avoid churn by identifying the warning signs and stopping churn before it happens. He talks about data-driven techniques for converting raw data into measurable metrics, and testing hypotheses, to ultimately maximize customer retention and minimize actions that cause them to stop engaging or unsubscribe altogether.
Introducing Carl, he’s the Director of Machine Learning Implementation at OfferFit. Earlier, he was the Chief Data Scientist at Zuora, where he created the Subscription Economy Index and the data science behind Zuora’s Subscriber Insights. He is the author of the Book “Fighting Churn with Data”. Before coming to Zuora, he spent most of his post-academic career as a quantitative analyst on Wall Street.show more
Now a data scientist, he uses a variety of tools and techniques to analyze data around online systems. Carl believes that AI is good at detecting churn. But currently, it is not adept at advising us on what to do with the data. Carl is bringing the analysis to the humans who are dealing with churn. He’s the author of the book “Fighting Churn with Data,” on which this session is based. Carl, welcome and thank you for joining us. I know it’s early for you. You’re on the other side of the United States, and I’m very excited because this is the only, quote-unquote, ‘not exactly a sales book’ in the sales and marketing lit fest. I thought it was extremely important to do this because we are so focused on acquiring customers that we forget about keeping them. And that’s where I think this fits right in.
Thank you. Please start, and I will disappear from the stage.
Okay, sounds good, and I’ll dive right in.
I’ll skip the “About me” slide since you just gave me such a great introduction, Subhanjan. Thank you so much for that.
Here’s an outline of my talk for today. I’m basically going to give you a nontechnical synopsis of the book in the talk this morning. The book is actually quite technical. It is focused on people in your IT department or your technology division, with actual code samples of how to do things in the book. But this is just going to be a nontechnical overview of the book— Churn and how to fight it. A little bit about churn rates because you’ve probably all heard of them, something about customer data, and finally, everyone always wants to talk about AI. So I’ll make a few comments about AI.
So, you probably already know this. I give this talk to a lot of people, but I’m sure you probably already know what churn is. Churn just means customers quitting, canceling, unsubscribing, or it could be unfollowing in a YouTube channel or just not coming back if it’s retail. So churn actually means your customers not coming back in a variety of contexts. It’s originally from something called the churn rate. We’ll say more about that in a minute, which means the percentage of customers that drop out. Churn is also now a noun and a verb. You can say the customer churned or make a churn report. So churn cancels your growth in a subscription or really any kind of company with recurring revenue. And that’s the key thing you want to know, that I have to tell you, all that probably is not.
So, what is fighting churn with data? Really, it means using your data to do things to prevent churn. And here are some of the top ways. The number one way, of course, to prevent customers from churning is to make a great product, and you can use your data to do that without surveys. You can also use your data for targeted marketing, which I think is probably the focus since this is a sales and marketing event. So you want to send customers informative and engaging content that’s going to reinforce their interest in the product and hopefully help them use it better. There’s also a new field called customer success. It’s kind of related to sales and marketing but a little bit outside, which is helping customers proactively rather than the traditional support function, which waits for customers to call. You can also use your data to improve your pricing.
And lastly, you can use it for acquisition targeting if you identify which channels give you the customer least likely to churn. You can then also target those channels for more acquisitions. That’s kind of another sales and marketing focus.
Why is churn hard to fight? Well, there are a number of reasons. The first one is that it’s hard to predict churn, and that’s because it’s a very subjective process for every customer. And even with the best data science and AI, it’s hard to know exactly when a customer will churn. It’s also harder to prevent churn than to predict it because, in a way, it’s not entirely a sales and marketing problem, to be honest. Because the customers already know the product, and you actually have to deliver more value to prevent them from churning. There are no silver bullets or tricks you can pull at this point because, like I said, the customers already know the product.
Another challenge with churn is that it’s a multi-team effort in any company. If you look at the things I mentioned that you need to do to improve churn, which is improving your product, having targeted marketing, customer success, pricing, I’m just naming every department in a modern company, and they all have to work together on churn. And they all have different tools and methods for doing so.
This is what I call the churn-fighting pyramid, which shows you what you have to do to fight churn, and it’s really based on your data. The foundation layer is your customer data, and hopefully, you’re storing the customer data in a CDP or customer data platform. And that’s just the foundation. Having the data is good, but to do anything with it, you have to build these additional layers. Making customer metrics, which are measurements of your customers, is the first level. Segments and targeting are the next level. You group your customers and then take targeted actions. And then the very highest level is the AI and automation approach. And each of these levels has to build on the other one in order to be successful.
So, the good news about fighting churn, I’ve been a little bit pessimistic before— it’s so hard to fight. You can’t predict it. The good news is you may never eliminate churn, but you can improve a lot. And a little bit of churn reduction has a dramatic impact on other business metrics, particularly your growth rate. And then growth compounds over time. So, a little bit of churn reduction really improves any business.
There’s what they call the 80/20 rule or ‘Pareto rule’, and I say there’s a weak form of that for churn. If you can make 25% of the effort, you get 50% of the churn reduction. And that would be to calculate churn, calculate your churn rates, and some customer metrics, and just pay attention to them. And that will actually improve your strategic thinking so much that it’s going to help your churn rate. Just from doing that, for most companies, you can get 75% of the churn reduction possible with say 50% of the effort. And that would look like making interventions targeted by simple metrics, maybe some simple A/B testing on your product. So the good news is really that doing something about churn is much better than doing nothing, and you don’t have to do a lot to get a lot of the benefit. So that’s the good news about churn, I guess you could say. Let’s talk a little bit about churn rates because everyone’s heard of it. It’s worth me spending a minute to make sure you understand it.
The churn rate is the percentage of customers that quit in some time period. It could be a month or a year, depending on your business. Consumer businesses will generally measure a monthly churn rate, while B2B or Enterprise businesses will generally have annual contracts, so they measure annual churn. Acquisition is not part of the churn rate. So new customers cancel your growth, but your churn rate only measures the customers that quit. The churn rate is like the leak in your bucket, that is what they say.
You might have also heard of customer retention, and it’s literally the flip side of churn mathematically speaking. So your churn rate plus your retention rate has to equal 100%. So retention equals 100% minus your churn, and churn equals 100% minus retention. So you can really talk about them interchangeably. They’re two sides of the same coin.
Usually, here’s my one tip. Usually, when people talk with their investors or their financial reporting, they give their retention rate because it focuses on the positive. Then you get to say, “Oh, I have a 95% retention rate or 97% retention rate.” Sounds good. When you’re focusing on reducing your churn, then you focus on the churn rate, which will be a small number, like 3% or 5% or something like that. And there you have to recognize that a small relative reduction in the churn rate is huge. Like if you go from 3.3% churn to 2.7% churn, it doesn’t sound like much, but that’s a 10% relative reduction in your churn rate, and that’s really huge. So it’s really relative to the churn rate that you’re at.
Here’s some data about what typical annual churn rates are. This is based on the Zuora Subscription Economy Index, a report published in March of this year. Basically, it’s showing that there was actually a real dip in churn in 2021, and then in 2022, it’s kind of gone back to normal. It’s in contrast to the early fears in the pandemic. For those of you who don’t recall, and I’m sure you all do, back in 2020, everyone was like, “Oh my God, this is going to be so bad for businesses.” But actually, for the services that are subscription services like the kind that are on the Zuora platform, like streaming, SaaS, and Zoom, the pandemic was really a boon for those services. 2021 was a record-low churn rate year, and then 2022 was sort of back to normal. I should mention this data is based on a cross-section of SaaS and streaming companies primarily, some content providers outside of streaming, but it’s kind of a broad cross-section of churn rates. So if your churn rate is higher than 23% annually, you know, don’t worry. You might be in a different vertical or something like that.
Okay, 2022 was back to normal. Now, let’s actually go into the section where I’ll talk a little bit about data and make examples of using data to fight churn. I’m going to make a case study from a company called Broadly. Broadly manages a business’s reviews on an online platform. So Broadly is a SaaS company that other companies use to manage their reviews. And they have data here, I’m saying— I called it events. This is basically the types of data that they have, which is when they ask customers for reviews and get reviews. So they present a review ask to the customer, then there’s a decision, and the IT may ask to be fulfilled, and then there’s a positive review or a negative review that comes out of that. So this is what Broadly does. It puts these pop-ups to get reviews. Now we’re going to talk about the churn of users of Broadly, which are businesses that are trying to manage their online reviews.
Here’s some data on how a customer metric predicts churn. You’ll see in the slide, I’m showing the churn rate on the Y-axis, but I’m not showing you the numbers because I can’t actually tell you what Broadly’s churn rate was. But what this shows in the data is that the more positive reviews a company gets, the lower their churn rate. And you can see it’s quite dramatic. At 0 positive reviews, the churn rate is very high. It only takes a couple of positive reviews per month for the churn rate to fall dramatically. This is for the companies that use Broadly, churning from Broadly.
You also see that the churn rate plateaus out or rather dips out in a valley where more and more positive reviews do not further reduce churn. This is a very common pattern to see with the usage of any product. A little usage goes a long way in reducing churn, but then it levels off after a certain point. In the book, I refer to this as a metric cohort analysis where you group customers by a metric and then calculate the churn rate in each cohort. It’s a very simple analytic technique that everyone in the company can understand, and that’s why I recommend it.
Alright, we’re going to look at one more aspect of Broadly’s churn, which is kind of interesting. What about negative reviews? We said the more positive reviews a company gets through asking for reviews, the less likely they are to churn. What about bad reviews? Here, the data actually shows that the more negative reviews a company gets, they are also less likely to churn from Broadly. That’s a bit counterintuitive. You might think, “Wow, someone’s getting bad reviews. Why are they using this product to get reviews?” Well, there’s a technique I teach in the book called Making a Ratio Metric. In this case, we look at the negative review rate, which is the percentage of negative reviews. Now you see that actually, the higher the percentage of negative reviews, there is a higher churn rate for Broadly customers. The reason that having a lot of negative reviews looks like a low churn rate is that those companies just get a lot of reviews. If you get a lot of reviews, you’re going to get a few negative reviews. But only having a high proportion of negative reviews is bad for churn from Broadly’s point of view. Again, Broadly is the company that helps them collect reviews. This is a technique I teach in the book about making ratio metrics, which can give you great insight into churn, especially in clarifying the impact of disengaging things that happen on your platform. More details about this in the book, of course.
Lastly, let’s talk about the subject of using AI to reduce churn. It’s a very common topic, and I do go over techniques for this in the book. You can use some types of AI systems, and this will give you the maximum churn reduction possible. But you have to build this on all of the other foundations. You have to have a customer data platform in place or CDP. You have to calculate your churn and customer metrics. You have to have a customer success program running and be doing some A/B testing with your product. And then you can use AI for some types of churn. But you have to be careful with AI for churn because there’s no one-size-fits-all churn intervention. In a lot of academic courses, they give an example like, “Okay, predict churn and send a discount to everyone who’s likely to churn.” Now, I didn’t mention it here, but sending discounts is actually the worst way to reduce churn, and you really only want to send discounts to the customers who need them. And a standard AI must evaluate well on which customer the discount is going to be worth it. That’s what I mean. And a standard AI model is not predicting that. So you need to target your churn-reducing messages and offers. That’s why it can be hard to use an AI model to predict churn unless you’re using a very advanced form. So simple churn risk models should not be used for targeting. To be honest, that’s all I teach in my book. The real state of the art of using AI for churn is actually to use something called reinforcement learning, which is the focus of my current company. My role there is on using reinforcement learning to pick the interventions. It’s a different standard. A propensity model for churn will tell you who’s going to churn but not the impact of your interventions. The AI really has to predict the impact of the interventions. That’s the focus at my current company called OfferFit. As the name implies, it’s finding a fitting offer for your customer, and that’s using reinforcement learning in a variety of marketing contexts.
And I think I’m just about out of time. Sorry if I talked so quick. I get excited, and I wanted to pack a lot of content in. But top takeaways from today: Churn is hard to fight, but there’s a lot you can do. The main churn-reducing strategies are product improvement, engagement marketing, customer success, pricing to deliver value, and acquisition. Customer metrics that you calculate on your data provide actionable insight and alignment between departments. And you can use AI for churn, but there are many pitfalls. Consider talking to the team at Offer Fit. Actually, I mean, I’m not here to promote OfferFit, but I’m just gonna mention that if you are interested in that, then reach out to me there, alright?
I see Subhanjan’s back on. I must be fully out of time now.
No, no, you did very well. You did very well, Carl, and you know, this was such a… I don’t know what you did, but you really simplified it a lot, and it was very effective. And I really cannot emphasize enough that attending to existing customers and ensuring that they’re getting the value they signed up for is at the core of this value transaction, which we call sales or buying, whichever way you look at it, right. So we are delivering some value, they’re buying, taking by paying for some value. That value must be delivered on a consistent basis. I mean, that’s what you would expect as a customer yourself. No reason why it should be different for your customer. So yeah, this is great, and I’m so glad that you joined. You could have told me, ‘Subhanjan, this is a sales and marketing lit fest. What am I going to do there? Don’t keep me out of it’. But thanks for accepting and coming on.
A lot of sales teams now are responsible for the renewal. And if you’re responsible for the renewal, that means you’re responsible for churn.
Absolutely, absolutely. We have a couple of questions. Let’s have them, and then we move on.
Yeah, okay. Let’s see.
For smaller organizations, is it possible to hire the skill required?
Wow, that is a tough challenge for everyone hiring talented people. I don’t know what to say, but I think it’s possible, and definitely, I mean, I know this year the layoffs in big tech companies made hiring opportunities for smaller companies. Being in Silicon Valley myself, and I have friends who were affected by the layoffs, they got laid off from a big company and they just went to work for a smaller company. So hopefully, it was a little bit better in 2023. Yeah, definitely, hiring is a challenge for everyone, and I work at a startup.
One more question. Okay, that’s it.
Let’s see, fear that we’re turning into a data plan, losing out on the human touch.
That’s a reasonable question also. Well, hopefully, the data captures a lot about, you know, the people really. But you’re right. I mean, the truth is, the data is only true on average and, in a way, it’s related to my point that there’s no one-size-fits-all solution, and the fact that you, the data complements a customer success department, definitely in enterprise sales and companies. Your customer success is going to be the human touch, and they’re going to have the data, you know, in the background, but they’re still going to reach out. And hopefully, in a smaller company, again, it’s the marketing department that’s going to provide the human touch by… I mean, I said you use targeted marketing, but you know who actually writes the emails? Who actually writes the messages? Nowadays, people are talking about using chat GPT, but I’m not… I still think most companies are going to, you know, have a human who actually determines the messaging.
Yeah, absolutely. And at this point, I think, fascinating as it is, what ChatGPT can do. I mean, it really writes some nice limericks and stuff like that, and it is good. But thinking that it’s the solution for all problems, that’s stretching it a bit, I guess. Okay, Carl, again one more question.
What extent can queries be fully automated? I’m actually not sure. I mean, there are a lot of areas of AI that I’m sort of an expert in, like propensity modeling and reinforcement learning, but I’m not really… Yeah, I actually don’t know. I’m just going to tell you what I don’t know, and I don’t know about automating queries. Really sorry for that.
Okay. Carl, thank you so much again. I really enjoyed the session and thank you for joining us. And I hope to see you in the next lit fest as well.
Thank you. Alright, thanks. My pleasure.
Thank you, guys. We’ll be back with Ted Olsen in a bit. Don’t go away.
Carl Gold is the Chief Data Scientist at Zuora, Inc. where he created the Subscription Economy Index and the data science behind Zuora’s Subscriber Insights. He is the author of the Book “Fighting Churn with Data”. Before coming to Zuora, he spent most of his post-academic career as a quantitative analyst on Wall Street. Now a data scientist, he uses a variety of tools and techniques to analyze data around online systems. Carl believes that AI is good at detecting churn. But currently, it is not adept at advising us on what to do with the data. Carl is bringing the analysis to the humans who are dealing with churn.