We don’t just segment customers—we find every pattern in their behavior

 
 

Most statistical techniques work by breaking data apart to create distinct groups. The points within each group must be similar to one another, and different from other groups, by some threshold set by a data scientist.

The problem with this approach is that it erases nuance. All points in a dataset are forced to belong to a single group, to stand for one overarching pattern. In reality, in a complex dataset, all points have some connection to one another. And each point can be a part of hundreds of different patterns.

We don’t try to segment consumers into one group. We know each consumer has a huge variety of attributes and behaviors. We preserve that variety, creating dynamic Netflix-like categories for your data. A consumer can belong to many of these categories, not just one.

This allows us to find all the patterns that correlate with a metric. Take high retention as an example. Using traditional analysis, you would start by carving out a segment of high-retention customers and try to deduce which patterns might be used to explain that segment.

Unsupervised starts by identifying every possible pattern, not by segmenting consumers. We directly identify any and all patterns related to high retention. Then, we rank the patterns according to their degree of impact, so at the top, you see what has the biggest effect on retention.