Michael Kaminsky explores how one team uses causal machine learning to predict high-LTV customers instead of chasing volume. They cut lead volume but doubled customer value in three months by filtering Meta campaigns through a predictive model trained on their full customer dataset.
Chapters
So you mentioned a little bit earlier how important lifetime value is to the business and how much y'all think about really wanting to drive high lifetime value customers. How are you thinking about combining this idea of driving sales from incrementality versus this thing that you care about of longer-term lifetime value when that can't be measured in a month, right? How do y'all think about connecting those ideas?
Yeah, we teamed up with, with a partner who works with modeling of data and they're really good at causal machine learning. What they could actually help us do was actually do some calculations on all the customer data we have. And then through causal machine learning, look for patterns that we could not see, and then actually find signals we could give to Mesa and then use for our low-performance marketing to go after not just everybody, but go after high predicted customer lifetime value customers. This was really interesting because we could actually see it gave us more volume in sales. Both short-term but also longer-term, we were only able to track these customers with UTM tracking and cookies. We did not have access to communicate with these customers if they were not on Meta. We actually wanted to go after still our predicted best customers, but instead of having it in a short-term sales campaign, use that signal to go after the leads. So have a lead set up in Meta, give them the data around our best customers, and go look for our new best customers. We are running this test actually right now. We can basically see this is working. So we can see compared to another lead setup, we get fewer leads. They are more expensive, but we can see when we have the lead and they gave us consent, so we can track them. We can see over, for instance, 3 months, they become a lot more worth to us, like times 2 or something like that. So even though they are more expensive to catch, the long run or the long-term profitability is much, much better. So what we need to figure out now, how can we put even more scale into this? So we can find as many new best customers as possible. And I think if we nail this one and we can also put in offline data into this setup, we'll have an extremely powerful setup for the future.
