A Growth Hackers Guide to Building a Predictive Growth Model
Update: The growth model is now updated to version 1.6 with new features and ability.
At Mozilla, I’ve implemented analytics tools, performed deep analysis of our funnels, and launched countless A/B tests on our experiences. While I’ve learned from it all and improved key conversions, the question I would get asked over and over is “How will change [X] impact Firefox’s active user base by [Y] period of time?”. I wasn’t able to answer this question outside of saying that ROI of retention should be higher than acquisition if all things are equal, but that was not a satisfying answer to my stakeholders or even myself.
I knew that basic funnel analysis or understanding conversion rates wouldn’t be able to answer the impact question above. I also assumed that the size, type, and duration of change [X] should have a substantial impact on the outcome and I felt like I was on to something. All those questions and the curiosity that kept me up at night, let me to create a mashup of funnels and cohort analysis into what I called the Firefox Desktop Growth Model.
This growth model gave us a better sense of how Firefox’s growth levers impact our active user base at some point in the future. The key idea here is “at some point in the future”, which was possible with using time series data. The turning point for the model was when we learned how a seemingly big win on specific growth lever (input) had a much smaller and slow impact to our user base over time (output). We called these input vs output relationships our growth ratios and it helped size up opportunities based on the outcomes instead of focusing on potential vanity metrics or purely optimizing for local maximum conditions.
During most of 2016, I continued to improve the growth model, add new features, and it was surprisingly accurate to the trends we saw in our real monthly and daily active users. It was never exactly perfect, but it was close enough to aid with planning and prioritization of potential growth opportunities. When I shared my model with Growth friends at other companies, I was surprised to learn how few people had a growth model or what they called a model was just a glorified funnel or loop.
Then in early 2017, Sean Ellis and the team at GrowthHackers.com reached out to me to see if I would be interested in speaking at the upcoming Growth Hackers Conference. I said I’ve love to present and told them about the growth modeling I’ve been working on for Firefox. When building my slide deck for the talk, I realized that there is no way I could completely explain how to build a growth model from a presentation alone.
With the challenge of presenting on a complex topic like modeling, I did what any Mozillian would do and decided to take Mozilla’s non-profit, open-source, and public good mission and apply it to my growth model presentation. I structured my talk and gave simple visuals to ideally get the audience to have the “aha!” moment during the talk and that I would follow up with a call-to-action to follow me (Chris More) on twitter to get the open source growth model.
I’ve been working on the open source growth model for a few months now and I am happy to announce that Growth Model v1.0 is ready for prime time. Like any v1.0 products, it will improve over time, and the best way of optimize it, is getting others to take it for a test drive and provide feedback.
I’ve built the entire growth model in a Google Spreadsheet and it contains how-to instructions and over 100,000 calculations (v1.6 now has close to 2 million!) to be able to understand the impact of potential changes to your acquisition and retention metrics. Please note that it is very easy to hit the upper limits of a spreadsheet with modeling, thus remember to keep it simple.
If you want to do more segmented modeling, consider creating a separate model for different segments. In theory, you could create a separate spreadsheet model for all of your core segments and then use another spreadsheet to aggregate them all together into a comprehensive view, but that is a topic for a later post.
How to build a predictive growth model (presentation)
Predictive Growth Model v1.6 (spreadsheet as of March 2018)
Got questions about this growth model or modeling in general? Ask me on Twitter and I’ve love to discuss more.