Predicting LTV with ML - Issue 217
ML models for predicting LTV in freemium apps: research and case studies
Welcome to the Data Analysis Journal, a weekly newsletter about data science and analytics.
Thank God for LinkedIn and Paul Levchuk, who enriched us this summer with his research on numerous academic studies on LTV. Now, it’s my turn to pick up the baton.
During my blogging tenure, I’ve tried to stay away from LTV in my newsletter because:
LTV is the most over-discussed and over-studied KPI. It feels like everyone is writing about it.
There are 100 ways to forecast LTV, 90 of which are very complex. There isn’t one perfect and universally appropriate method to calculate LTV; each has a lot of nuance.
You’re not an analyst until you deliver an LTV prediction. Sooner or later, everyone does it. It’s like graduating into a dark world of under-budgeting and overspending. Fortunately, marketing analytics is not my passion.
Most importantly, you don’t need LTV to build a great product and bring it to market. Investors need LTV. Like I said, it’s a dark world of under-budgeting and overspending.
So today, instead of offering my own method for forecasting LTV (and I have 4 main models running, each returning completely different values for the same dataset 🙃), I decided to share 3 case studies on using ML to predict LTV for freemium apps, inspired by and borrowed from Paul Levchuk research.
What I appreciate about Paul is that he’s one of the rare analysts today with strong attention to detail, critical thinking, and thorough due diligence. In other words, you can’t bullshit him with uncontrolled A/B testing, sloppy research, or flawed logic. I enjoy watching him call out inaccuracies in studies, incomplete analyses, or shallow comprehension, whether it’s a junior analyst or an operating partner at OpenView with the largest PLG newsletter.
