Many important business questions are not about whether something will happen, but about when. How long will this customer stay with us before leaving? How long until this machine needs maintenance? How long, on average, does a new customer take to make a second purchase? These questions about the time until an event are of enormous value, but they require a different approach from the usual analysis — a technique with the curious name of survival analysis. Despite the name, which comes from its medical origin, this technique applies to any situation where it matters how long passes until something happens, and it is one of the most useful and least known analytical tools outside the statistical world.
Survival analysis studies the time until the occurrence of an event. It was born in medicine, where the "event" was, typically, a patient's recovery or death, and where it mattered to understand how long patients "survived" — hence the name. But the same math applies to much happier events in a business context: the "event" can be a customer cancelling, a machine failing, a user making their first purchase, an employee leaving. Whenever the question is "how long until this happens?", survival analysis is the right tool.
This article explains what makes survival analysis special, why a normal analysis does not work for these questions, and where it creates value in companies.
Why a normal analysis is not enough
At first sight, it may seem these questions about time can be answered with a simple average: what is the average time until a customer leaves? But this approach has a fundamental problem that makes it, often, misleading. The problem is technically called censoring, and it is this: at the moment we do the analysis, many of the cases have not yet experienced the event. Many customers have not yet left — they are still with us. How do we include these customers in an average of "time until leaving", if they have not yet left?
Ignoring them would be a serious error, because we would be analyzing only the customers who have already left, which biases everything — those who stay longer are underrepresented, precisely because many of them have not yet left. Including them as if they had left at the moment of analysis would also be wrong, because they did not leave. This situation — having cases that have not yet experienced the event at the moment of analysis — is what survival analysis knows how to handle correctly, and it is the reason simple approaches fail. Survival analysis takes advantage of the partial information from these cases ("this customer has been with us for two years and still has not left") without treating them wrongly.
What survival analysis reveals
Instead of a single number — "the average time until leaving is X" — survival analysis produces a much richer picture: how the probability of the event not yet having happened evolves over time. It shows, for example, what probability a customer has of still being with us after a month, after six months, after a year, after two. This curve over time says much more than an average, because it reveals the pattern of how and when the event tends to happen.
And that pattern is often revealing. It is often discovered that the risk of the event happening is not constant over time. A customer may have a high probability of leaving in the first months — the critical phase in which they have not yet become loyal — and then, if they survive that phase, a much lower probability of leaving. This information about when the risk is greatest is immensely useful for acting, and it is precisely what an average would hide. Knowing the danger is in the first months completely changes where retention effort is concentrated.
Where survival analysis creates value
- Customer retention: understanding how long customers tend to stay and at what moments the risk of leaving is greatest, to act at the right time.
- Predictive maintenance: estimating how long until a piece of equipment is likely to fail, to maintain it before the costly breakdown.
- Customer lifecycle: knowing how long, on average, a new customer takes to reach important milestones, like the second purchase.
- People management: understanding the patterns of how long employees tend to stay and when the risk of leaving is greatest.
From pattern to action
The real value of survival analysis is not only in describing how long things last, but in enabling action on that knowledge at the right moment. If the analysis reveals that the risk of a customer leaving is greatest in the first three months of the relationship, then that is where the onboarding and loyalty effort should be concentrated — not distributed uniformly across the whole relationship, but focused on the critical window. This targeting of effort to the moment it matters most is what turns the analysis into a practical advantage.
Survival analysis also lets you compare groups: do customers arriving through one channel "survive" longer than those arriving through another? Does one supplier's equipment last longer than another's? These comparisons, done correctly with the technique that knows how to handle the ongoing cases, give reliable answers to questions of great strategic value, which a naive comparison of averages would distort.
A concrete case
A subscription services company wanted to better understand its customer retention, to reduce churn. The team's first approach was to calculate the average time a customer stayed before cancelling. But they quickly realized that number was misleading: they could only calculate it from the customers who had already cancelled, and that left out the enormous number of customers who were still active — precisely those who stayed longest. The resulting average was biased downward and did not reflect reality. They then decided to use survival analysis, which knows how to take advantage of the information from still-active customers without treating them wrongly. The result was much richer and more revealing than an average. The survival curve clearly showed that the risk of a customer cancelling was not constant: it was very high in the first months of the subscription and then dropped substantially. In other words, if a customer got past the first critical months, they became much more likely to stay long term. This discovery completely changed the company's retention strategy. Instead of spreading loyalty efforts uniformly across all customers, they concentrated them intensely on the initial phase — the onboarding of new customers, close follow-up in the first months, ensuring they quickly reached the value that would make them loyal. The analysis further revealed, on comparing groups, that customers arriving through a certain channel survived much longer than those from another, which also informed where it was worth investing in acquisition. By concentrating effort on the critical window that survival analysis had identified, they managed to reduce churn far more effectively than if they had dispersed resources. The value came not from just knowing how long customers stayed, but from understanding the pattern of when the risk was greatest and acting precisely there.
An underused tool
Despite its enormous value, survival analysis remains one of the least used analytical tools outside specialized circles, largely because of its intimidating name and medical origin, which make it seem more complex and more distant from business than it really is. This underuse is a shame, because many of the most important questions companies face are, at heart, questions about the time until an event — exactly what this technique answers better than any other.
Recognizing when a business question is, in essence, a "how long until?" question is the first step to taking advantage of this tool. Whenever it matters to know the duration until an event, and especially when many cases have not yet experienced it, survival analysis is probably the right approach — and using a simple average is probably misleading. This simple awareness opens the door to much more reliable analyses for a whole class of valuable questions.
In practice
Next time you face a question about how long until something happens — how long customers stay, how long until a breakdown, how long until a conversion — resist the urge to answer it with a simple average, especially if many of the cases have not yet experienced the event. That average will almost certainly be misleading. Survival analysis gives you a much more reliable and rich picture: not only how long, but the pattern of when the risk is greatest, which is what lets you act at the right time. Are your business's "how long until?" questions being answered with the right tool, or with averages that hide the cases still ongoing?
