A time machine that predicts churn
If you could rewind the customer journey to where churn starts happening, could you stop it? What would an early warning system look like? Read: "Everything Starts Out Looking Like a Toy" #252
Hi, I’m Greg 👋! I write weekly product essays, including system “handshakes”, the expectations for workflow, and the jobs to be done for data. What is Data Operations? was the first post in the series.
This week’s toy: a fun interaction that illustrates game theory visually. Ever heard of the Prisoner’s Dilemma? Try it out using different strategies and see how you do!
Edition 252 of this newsletter is here - it’s May 26, 2025.
Thanks for reading! Let me know if there’s a topic you’d like me to cover.
The Big Idea
A short long-form essay about data things
⚙️ A time machine that predicts churn
I want to build a time machine. Let me be more specific. I want to create a log of events that's so good it identifies the events that are very likely to predict churn. In other words, if you had a perfect memory as a system, you could travel back in time to the moment someone lost their confidence in your product and decided to churn.
That's cool, you say, let's consider some things that would have to be true to predict churn this accurately.
You might start by comparing events for users who decided to churn and the parallel events for other users in the same initial cohort who stuck around. One of the challenges with that approach is that people have different situational reasons for churning, and their underlying motivation for using the product might also be different.
For example, if someone logs in once a month and another person logs in twice daily, is the second person more or less likely to churn? It depends highly on how you deliver value in your product.
Let’s talk about value
People churn from a product when they no longer believe the trade of money for value is working.
It sounds abstract, and it is.
But go beyond this abstract idea to think about the exact steps you need to know to determine whether value is delivered. It's hard to pinpoint one moment – to paraphrase an old movie – when Tom Cruise could tell you that pre-churn has happened.
To make this concrete, consider how we can measure value delivery. Every product has key moments where value is either delivered or missed. These moments create data points to help us understand when customers get the value they expect.
For example, if your product helps users save time, we want to track how long it takes them to complete key tasks. If it helps them make better decisions, we want to define “a good decision” and objectively measure the quality of their outcomes. These observations are the foundation of our churn prediction system.
Today, that moment often comes as a surprise.
When a customer tells you they are thinking about leaving or considering other options, you have already passed the point of being in pre-churn. You're facing a churn risk.
Turn back the clock and search for clues
To go further in the past and search for the relevant “pre-churn” event, you'd have to know the answers to questions like:
What was their initial brand experience, and how did it go? Did the company make promises to them about responding that were met, or did they have a negative initial onboarding?
Were they slow or fast when compared to a definition of a successful customer in coming up to speed and finding value in the product?
Did they tell you anything about their experience, or were they silent?
The answers to these questions combine behavioral information (the telemetry of their data within your system, the tone of their communications as captured by a meeting bot or revealed from email transcripts) with preferences (how they like to be treated, which communication channel feels natural) and their situation (how did they feel in a particular moment when they used the product or talked to us, what was going on in their life at the time).
SPOILER: You won’t get an easy answer that tells you yes or no.
Churn comprises a multivariate question where part of the answer differs for every customer. Because every customer has their own journey (thank you for that insight, Alan Berkson), you will do best by understanding how the median customer you observe compares with your best customer.
Said another way, if you looked at some event logs, what would you see?
What does a great customer look like?
Your best customer:
might respond immediately to you when you ask them a question
has a workflow path time that is faster than the average, maybe much less
will sometimes tell you that they really like your service and deliver a high NPS or CSAT score consistently
But that best customer might also be pretty quiet. In a big business, you don’t have a complete list of these people today. (Hint: they are not always your highest value or longest-tenured customers.)
Depending upon their personality, your best customer might feel you deliver value for the money they spend, and their preferred outcome is not to hear from you at all. They believe that when something important happens and requires their attention, you either deal with it on their behalf (if given permission) or alert them immediately to take action.
Defining trust and high expectations
Consistent value delivery through proactive service or quiet reliability creates something even more fundamental than satisfaction. It builds trust in a “one in a row” sequence. When customers trust you to deliver value consistently, they're more likely to weather the occasional storm or temporary setback. They know you'll be there when it matters.
Looking at our best customers, we see these trust-building behaviors in action. Quick responses, efficient workflows, and high satisfaction scores aren't just metrics - they're signals of a healthy, trusting relationship.
But trust isn't built by accident. It results from deliberate, consistent actions that demonstrate reliability and value.
As product people, we want consistent delivery of expected value and emotional safety during uncertainty. As customers, we want to feel "they've got this" and mean it.
Getting it right on every contact
But how do we operationalize trust?
It starts with setting clear expectations and then consistently meeting or exceeding them.
When you spot a potential issue in a customer's usage, reach out before they notice. Keep them in the loop about updates that might affect their work. Share insights that help them get more value. It's all about showing you're thinking ahead for them.
Next, respond predictably by setting clear expectations for response times and sticking to them. Follow up on promises, even if it's just to say "still working on it." Make sure they never wonder if their message was received.
This kind of reliability builds confidence.
You also need to tell the customer what’s going on (and measure whether they are paying attention). If there's an issue, tell them. Share your roadmap and how it helps them. When things go wrong, explain what happened and what's next. It shows you respect them enough to be honest.
Finally, make it personal and demonstrate that you see them as a human by remembering their preferences and interactions. It’s very annoying to explain yourself more than once to a rep, so don’t allow that to happen.
As you’re communicating, you also need to match styles with the customer while maintaining a consistent brand voice. (Yes, I know that can be challenging.) Use their data to anticipate their needs and demonstrate that you see them as people, not just accounts.
The key is that trust isn't built in grand gestures, but in the consistent execution of these small, reliable interactions. Each interaction is a data point that either strengthens or weakens the trust relationship. When we track these interactions, we can see patterns that indicate whether trust is being maintained or eroded.
How do you customize your time machine?
So, what does this mean for our time machine?
If we could travel back in time to prevent churn, we wouldn't be looking for a single moment of failure. Instead, we'd be looking for the patterns of trust erosion - the missed expectations, the unaddressed concerns, the moments where we failed to deliver value consistently.
Our time machine isn't just about predicting churn; it's about building a system that helps us maintain trust through consistent, reliable value delivery.
The real power of this approach is that we don't need to wait for a customer to tell us they're thinking about leaving. By tracking these trust-building interactions and value delivery moments, we can identify when the relationship starts to fray before the customer even realizes it.
That's the true magic of our churn time machine - it doesn't just tell us when someone might leave; it shows us how to keep them engaged and satisfied in the first place.
What’s the takeaway? Churn isn't about a single moment of failure, but patterns of trust erosion. By tracking customer interactions, response times, and value delivery, you might see warning signs before customers realize they're dissatisfied. You need continuous delivery of consistent, reliable actions that show you understand customer needs.
Links for Reading and Sharing
These are links that caught my 👀
1/ Chatbots and the “uncanny valley” - We’re all a little bit used to the idea of chatbots that can parrot answers and summarize information. But when a bot’s description of a physical place starts to sounds more like a travelogue written by a New Yorker writer, it triggers an “uncanny valley” response.
2/ AI workers can work autonomously for longer - With the introduction of Claude 4, a worker can be dispatched on a task for up to 8 hours. This sounds suspiciously like a benchmark that could be compared to human labor for the same amount of time. Why don’t we focus on whether the AI worker asks for help or doesn’t report back until it’s down a rabbit hole of circular logic?
3/ LLMs are making us dumber - Vincent Cheng shares some engaging thoughts on how LLMs are making him dumber. But maybe that’s ok, because they are also replacing some of that thinking with superhuman abilities. Balancing the cognitive impact of this dissonance is a project that belongs to every knowledge worker now.
What to do next
Hit reply if you’ve got links to share, data stories, or want to say hello.
The next big thing always starts out being dismissed as a “toy.” - Chris Dixon