Forecasting in Analytics: Choosing the Right Approach - Issue 249

How to predict revenue, user growth, and key business metrics using moving averages, regression, and ML

Welcome to the Data Analysis Journal, a weekly newsletter about data science and analytics.

I will be in Seattle next week - reach out if you’d like to meet and chat about all things analytics.

Also, this year, I am partnering with Data Council, a “no BS data conference,” and have a 20% discount to share with my subscribers. Use the code daj20 to save on your ticket. The event takes place April 22-24 in Oakland, California. After quite a few years, it’s finally back home in the Bay Area. Hope to see you there.

---

One of my recent learnings is that it’s surprisingly difficult to find analysts experienced in forecasting and prediction. Which is unexpected, given that regressions and time series forecasting are extensively taught in most academic programs today. But I’ve noticed that analysts tend to either overcomplicate things - applying complex ML to forecast moving average (?) or simply do regression plotting without deeper analysis or understanding the coefficients.

There are many models for forecasting and predictions, each with its own caveats and context. I won’t be going over the basics of forecasts and modeling. Instead, I will focus on the practical side - how forecasting varies depending on the type of project. Forecasting revenue, for example, is very different from forecasting LTV, Churn, or subscription growth, even when using the same model and dataset.

Today, I will cover the different types of forecasts and ML, and the common use cases that require predictive modeling in analytics. I’ll walk you through the steps and models to forecast revenue, ARR, and paid customers to answer questions like:

• How many page views do we need to double signups?

• How many more games, streaks, or logs must users complete before converting to paid?

• How to forecast MRR for the next 2 years.

• When will we reach 1M of subscriptions?

• When will we hit $1M ARR with our current baselines and ad spent?

In another follow-up piece, I’ll share examples of my forecasts and explain how to adapt them to different use cases.