Hey, great read as always. It's refreshing to see someone actually break down the 'how' instead of just the 'what' – this practical approach to segmentation is so smart and genuinly useful for anyone looking at app growth. I'm curious, do you have any thoughts on how these segmentation principles might evolve with more advanced AI models predicting user intent even earlier, especially for a new user?
Great question! I don't believe the segmentation principles will change, I believe AI tools may help you answer these same questions with less work, and faster.
To me, figuring out early intent is something that we can do today without advanced AI models. Simple regression or ML models can do this pretty well. Maybe AI tools will help you put those together faster, but honestly, usually the indicators of user intent for apps are kinda straight forward: is the user engaged? Do they interact deeper? Do they do more? Is usage frequency higher? Unfortunately, it's rare there's some novel crazy predictor that comes out of this type of analysis.
Well, it's certainly not easy. I don't think everyone is doing all of these things all the time. In my experience, you're constantly jumping around and diving deep into one area at a time. For example, you notice a drop in retention, and that triggers a deep dive into what's actually going on and some of this segmentation.
Hopefully, you have some support from analytics teams, and over time, you build more dashboards to track this stuff on an ongoing basis. This is usually the best case scenario, where you slowly build out automated dashboards for each of these areas and you slowly improve them as needed.
I think anyone working in consumer marketing today needs some quantitative analysis chops, but to do all of this well, you likely need some data engineering support to stitch all of these things together. SQL never goes out of style for when you run into the limits of Amplitude or Mixpanel...
Hey, great read as always. It's refreshing to see someone actually break down the 'how' instead of just the 'what' – this practical approach to segmentation is so smart and genuinly useful for anyone looking at app growth. I'm curious, do you have any thoughts on how these segmentation principles might evolve with more advanced AI models predicting user intent even earlier, especially for a new user?
Great question! I don't believe the segmentation principles will change, I believe AI tools may help you answer these same questions with less work, and faster.
To me, figuring out early intent is something that we can do today without advanced AI models. Simple regression or ML models can do this pretty well. Maybe AI tools will help you put those together faster, but honestly, usually the indicators of user intent for apps are kinda straight forward: is the user engaged? Do they interact deeper? Do they do more? Is usage frequency higher? Unfortunately, it's rare there's some novel crazy predictor that comes out of this type of analysis.
Thanks Jacob - great article. I particularly liked your metrics break down.
Thanks, Jacob!
This sounds overwhelming... :|
How many people & what kind of skills do you think you need in your team to pull this off? Feels like an army to me...
Well, it's certainly not easy. I don't think everyone is doing all of these things all the time. In my experience, you're constantly jumping around and diving deep into one area at a time. For example, you notice a drop in retention, and that triggers a deep dive into what's actually going on and some of this segmentation.
Hopefully, you have some support from analytics teams, and over time, you build more dashboards to track this stuff on an ongoing basis. This is usually the best case scenario, where you slowly build out automated dashboards for each of these areas and you slowly improve them as needed.
I think anyone working in consumer marketing today needs some quantitative analysis chops, but to do all of this well, you likely need some data engineering support to stitch all of these things together. SQL never goes out of style for when you run into the limits of Amplitude or Mixpanel...