Why you should #ClusterEverything. Clustering 101

If you've worked with great Audience Segments you'll know how powerful they can be as a tool. I've used them to change the economics and culture of global businesses and to help pop stars, TV shows and retailers to find growth. But how do you explain the power of clustering to someone that hasn't worked with them?

In this article I want to explain in as simple terms as possible what clustering is and why clustering your audience can so powerfully unlock insights you'd have otherwise missed. Here goes ...

It's hard to get insights from person-level data

First, let's start with some audience data. Here we look at 15 people who agreed or disagreed with a number of statements (from a recent 10,000 person study I did into luxury consumers). The same principles apply to any data set based around an audience whether it's transaction data, app usage data or survey responses: It's hard to see patterns with the naked eye.

We've all struggled with person-level data. What are the patterns? Where are the insights? It can often just look like a jumble of numbers. I don't see any patterns here, do you?

Population averages usually don't help

What usually happens in a situation like this is that an analyst will look at averages across the population. This gets to 'interesting' and allows some graphs to be drawn, but I still don't think it qualifies as 'insightful'. I.e. I don't think it will inspire any action yet. It still feels like a jumble of data with unclear 'so whats'.

Manually cutting the data sometimes helps

Analysts will often then manually cut the data and see if patterns emerge. What if we look by age or gender or region? Sometimes that gets you closer to insights. But it isn't guaranteed to lead you there. In my experience, clustering is a method that does guarantee you'll get to insights.

Clustering usually gets you insights

Clustering is where you cut the data into groups, but ones that are determined by an algorithm rather than by your manual determinations. Clustering takes all the people in your data set and groups them into clusters that behave similarly. In the abstract, here is a beautiful illustration of what clustering does. (Credit: Affinio)

Clustering our luxury data

If we apply clustering to the luxury survey data we used above, it groups the 15 people into three clusters, each of which has distinct behaviours. A clustering algorithm will find different groups without any human guidance. Clusters will 'emerge' from the data whether or not you have a strong hypothesis about the clusters that exist. It is almost like cheating.

Like magic, a pattern emerges

If I reorder the questions and the people in the way the clustering algorithm suggests, instead of seeing a jumble of people we now see a clear pattern emerge. We can relatively easily come up with names for each of the three distinct clusters that emerge.

Each group of respondents has two or three statements that all of their members strongly agree with and that define the group. From these statements alone you can easily picture each group of people and extrapolate to the kinds of things they'd want from your brand. You can imagine you'd do a very different marketing campaign for Aspirational Admirers than you would for Brand Loyalists, eh?

But the real power then comes when you look at all of the other survey responses that each group gave. You then insights into the brand, marketing, channel and style preferences of each group. You get to see whether some groups skew younger or skew older. And you get to build detailed product and marketing plans for each group that will be much bigger impact as they'll speak to their needs in a more focused way. Insights that drive action.

Summary: clustering is like a magic wand

In this example we started by seeing how hard it is to draw insights from person-level data ... until we clustered it, which felt to me like waving a magic wand over the data and watching insightful groups of people emerged as a result.

This doesn't just work for survey data. I've worked on similarly insightful clustering of social media followers and using transaction data from retailers. It is such a powerful tool, I believe we should #ClusterEverything.

I hope this has helped to bring clustering to life. The more people that are excited about this as a tool, the better we'll be at putting audiences first and the better served audiences will be.