Why segment?
Classifying data makes a complex amount of information more manageable; it allows us to make more effective use of the data available to us and take more precise action as a result. A segmentation is potentially one of the most well-known data classification techniques you can carry out in quantitative research, aiming to turn a broad audience into smaller subgroups that share common behaviours or characteristics.
Once you have your segmentation, you can see how much of the market is potentially available to you, prioritising and understanding these segments for better targeting, marketing and retention. The segments can be used at a broad strategic and planning level right down to feeding into regular CRM communications with messaging tweaked to resonate by consumer segment.
How to segment
There are different ways to approach a segmentation depending on your objectives, but these can be broadly put into two buckets:
- Setting the criteria that you want to use to create your segments, these criteria might be demographic (e.g. age, gender, income, household composition etc.) or geographic, but are known and existing parameters that can be used to group people together
- Finding behaviours, attitudes or emotional characteristics that group people together. These segmentations are behavioural or psychographic, and in consumer research are often referred to as needs-based segmentations. They require a rich understanding of consumers in order to group them by the benefits or motivations that are relevant for your category and product
We’re going to focus on the second approach, as you probably don’t need us to tell you how to group people with similar demographic traits together! This means that what we’re actually going to focus on is cluster analysis – the process of creating new segments based on relationships between data points.
Cluster analysis (usually K-means clustering)
Cluster analysis finds data points that are similar within their group and distinct from other groups. The most widely used method of clustering is known as k-means clustering, which is a type of unsupervised learning algorithm (there are no labelled / defined desired outcomes). This approach looks for naturally occurring commonalities between data points and works by finding centre points (the k-mean) and allocating data based on its proximity to these points. When looking to group people based on behaviours, attitudes, or needs that are relevant to your product, there’s potentially a broad range of relevant data that you might want to throw into the mix, even after an exploratory qual stage. For this reason, segmentation questionnaires intentionally capture a broad range of relevant data, and factor analysis is used to prioritise this ahead of running cluster analysis.
The number of clusters desired is set manually so the outcome is part science, part art. Often a range of different cluster solutions are explored until the most meaningful iteration is found. The definition of meaningful in this sense will be a combination of how distinct and relevant the segments feel, but also how many segments it’s feasible for the business to engage with – there’s a balance to strike between segmenting your audience and getting too niche or nuanced!
Once you have the ideal cluster solution for your aims, each cluster can be profiled and understood in relation to the other clusters – these are essentially your segments! In the first instance, clusters should be profiled on the key behaviours, attitudes or needs that have helped to form them so the essence of each can be understood. Wider traits such as any shared demographics can then be profiled, along with useful information such as brand engagement, likelihood to purchase and media profile. This lays the foundation for a detailed strategy that can guide you on who to target with what benefits, and where to best reach them.
