Our customer insight fundamentals blog series aims to unpack important components of effective customer data analysis, prediction, and activation strategies. This article was curated from William Morris, Director of Data Science at Faraday.
At Faraday we make the distinction between descriptive statistics and predictive statistics. The latter frequently gets all the attention (“Seeing the future!”), but it can’t happen without the solid groundwork of the former. You have to understand what data you have at hand before you can leap into the predicted unknown.
The data survey
Customer insights — as practiced by Faraday — are an example of descriptive statistics, but there’s a lot more to consider when laying your data foundation. Most data scientists practice some version of what could be called a data survey; the goal is to surface meaningful patterns, gaps, and anomalies with an eye toward prediction. One of the more common ways for a data scientist to approach a survey is with a notebook, which allows for an exploration narrative, almost like a blog post with code and charts.
For better or worse, notebooks can be freewheeling. While there’s no real limit to how far you can dig into certain datasets, at Faraday, there are a collection of standard metrics we look at in the process of building insights:
Acquisition time series
The pattern of how a group acquires members is usually essential in a data survey. This shows us if there’s a general upward trend, slowing enthusiasm, signals of seasonality, or spikes associated with specific actions like marketing campaigns or strategic discounting. A time series can also serve as the basis for any forecasting analysis.
Geography is a crucial indicator in a data survey. If a group is highly-concentrated in one region, it may not be the best seed from which to grow a national predictive model. The United States is a panoply of economic, racial, and cultural diversity, and geography is often the uniting factor in a host of demographic variables and indicators. Geography is also frequently the canary in the coal mine of statistical bias, and a good starting place from which to examine the implications of any future predictions.
Profile departure from baseline
As a component of Faraday’s customer insights analysis, we look at differences between a target group (e.g. “customers”) and a baseline group, like the whole - U.S. population. This allows us to gain a sense of the general profile of a customer and what makes them unique.
Armed with a sense of what makes your data tick, you can confidently approach predictive analysis. Insights and surveys offer the full view you’ll need when adjusting a Bagged Decision Tree for regional weight, or tuning seasonality in a Prophet forecasting model.
Learn more about Faraday's customer insight discovery solution.