For a newly launched subscription-based SaaS product, how do you prefer to calculate LTV?
There’s lots of ways to calculate LTV for subscription-based businesses, but I’ve been studying two in particular.
The first is very simple:
LTV = (Customer Value x Average Customer Lifespan)
Average Customer Lifespan = (1 / Average Churn Rate)
While this is very straightforward to calculate IF you have the correct Average Customer Lifespan, which for a new SaaS product, you’re estimating/projecting, and if your Churn Rate is uniform over time, which for many SaaS products probably isn’t the case.
In my experience, customer churn numbers after month 1 follows a power law distribution like y = nx^k, where k < 0. I suppose if you calculate the Average Churn Rate over the first N months, it might be good enough for a rough approximation of LTV.
The second approach involves a lot more data analysis.
I group customers into cohorts by signup month, then calculate the actual churn numbers by month, and use curve fitting to predict the future churn numbers.
Using only 6 months of data, this gets me R^2 > 0.83, which is good enough for this exercise. R^2 > 0.88 if I use 9 months of data.
I then apply the Kaplan-Meier estimator to calculate survival rate using my actual and estimated churn counts.
Then, to calculate LTV for each cohort:
LTV = Sum over time of (Survival Rate x Value)
I’ve seen some calculations that include a Discount Rate to account for Net Present Value and network effects of longer-lived subscribers, but since this is also another wild-ass guess, I chose to ignore it for now.
What I found super interesting is, between the two approaches, the LTV I calculated using the first approach compared to the projected first year LTV using the second approach yielded results that only differed by 0.59%.
The first approach, using an educated guess for Average Churn Rate, took a minute or two to compute.
I suppose the time it took to measure and calculate the churn rate to be able to come up with an approximation for the Average Churn Rate took some time, but that pre-work was already done.
But, this approach definitely relies on luck to choose the right Average Churn Rate number. If I had chosen a value that was just +/- 2.5% different, that would have given results that would have differed by +16.2% or -17.0%, respectively. My educated guess was incredibly lucky, and I wouldn’t have known how lucky if I hadn’t done the deeper cohort-based analysis.
The second approach took a bunch of hours of segmenting customers into cohorts, then tabulating their actual churn histories, then the curve fitting, then setting up the model to calculate the LTV for each cohort, and so on.
However, the outcome of this more thorough analysis is that it’s more closely dependent on the actual data and not on my ability to correctly guesstimate a value that is used in the calculation.