Freemium businesses' marketing techniques are immensely powerful. They drive large amounts of users to try a product and convert some small fraction of those to paid, upending the enterprise sales model.

In some sense, freemium businesses are real world Monte Carlo simulations. Because of the large volume of users using the product, freemium businesses can generates gigabytes of interaction data and conversion-to-paid data, which makes these kinds of startups particularly well suited to data science, A/B testing and regression analysis.

Over the past week, I’ve been trying this first hand. I’m building a Bayesian model in to predict whether an account will convert to paid for one of the companies I work with. I hope that ultimately this model will prioritize leads for an inside sales team.

![image](https://res.cloudinary.com/dzawgnnlr/image/upload/q_auto/f_auto/w_auto/image_222_592692678" width="500px">

The process of building this model has forced me to refine my thinking about the metrics we gather, the conversion to paid upgrade process and customer segments.

I’m still working on this model but some of the questions I found myself asking:

1. Are we collecting the right data to build a model?
2. How can we supplement data about our users to give them more context for analysis?
3. What kinds of characteristics should indicate a customer will likely upgrade? Can I validate those assumptions in the data?
4. Do we have a good enough trigger to convert to paid system?
5. Do our pricing plans make sense for our user base?

Instead of pursuing a product centric analysis of our conversion funnel in which customer feedback and design judgement is considered, this process has forced me to perform a customer centric analysis of the conversion funnel.

At the very least, I’ve come away with a better understanding of our customers. But I’m hopeful that this conversion to paid regression will be successful, that revenues grow faster as a result and I can adopt the technique with more startups.