If you had to choose a superpower what would it be? Maybe it’s just me, but I’d go for something like the power to understand a customer’s future cashflows. Even better, unlike flying or invisibility, this is achievable! We just need to calculate and forecast a customer’s LTV.
Here's a summarized view of some critiques of LTV by the folks at Reforge. Reforge is a phenomenal community and program; absolutely apply if you work somewhere with a training/education allowance. But I do want to respond to these points because I think they represent a consensus viewpoint.
1. The simplified calculation is deeply flawed. Correct, please don’t go for some hacky calculation. There’s no excuse for doing this! You have a couple of options here. If you’re handy with Python, use the Lifetimes library. If Excel is best for you, then follow along with Prof Daniel McCarthy’s video series here.
Additionally, there is a depth of academic literature on how to calculate and predict LTV. This isn’t just someone blogging and making wild claims. The best resource is cataloged by the LTV 🐐 himself, Bruce Hardie. The site is so basic you know the material has to be fantastic.
Here are some of his greatest hits:
- Creating an RFM Summary Using Excel
- A Spreadsheet-Literate Non-Statistician's Guide to the Beta-Geometric Model
- A Step-by-Step Derivation of the BG/NBD Model
2. It doesn’t help us operate day-to-day. If you believe that you can calculate and predict LTV, then you should reject this claim. You can operationalize LTV and LTV forecasts by:
- Understand where to dump marketing dollars. No one says, “I want the cheapest lawyers,” but that’s exactly what marketers do when they throw money at low CAC campaigns. What’s the value of the customers by channel? You need LTV to know that.
- More precisely staging reactivation/retention interventions. Take a cohort of users who have missed forecasted purchases and enroll them in targeted interventions.
- Find which features high LTV customers over-index on. Build more features that service this need. Experiment with driving lower LTV customers to these features to see if their LTV improves.
3. It doesn’t tell us about the effects of how, what and when we charge. If you have a model of your customers’ LTV, and you change the how, what, and when you charge, you will be able to see the impact downstream on LTV. It makes for a great A/B test! How did the LTV of segment X change when introduced into an experiment variant?
There are number of great (free) resources out there that help you rigorously measure and forecast LTV. LTV can and should be operationalized; see Etsy. This is a real superpower you can have!
*Standard disclaimer of being post-product market fit and enough customer history (in line with natural frequency) apply.