Audience Segmentation in Insights Reports

In this user guide, we explain what segmentation is, how you can cluster your audiences inside your Insights reports and which to choose depending on your use case

What is audience clustering, and why is it so important? 

Clustering your audience involves identifying groups of people (clusters) within a broader audience, allowing us to uncover unique insights about what they care about and what trends connect them. Understanding the difference between clustering and filtering is crucial, as often get misunderstood:

• Clustering (Segmentation): This is the output that occurs when an audience is clustered together by unique characteristics and connections that are data-driven and unbiased.

• Filtering: Often used as an initial step to define an audience or to select an audience post-insights, providing a predetermined audience.

  At Audiense, our 'unit of work' is the initial audience defined by specific criteria you choose to input, such as people found in a specific geolocation or those following accounts. Once your audience is defined, clustering enables you to dig deeper into the clusters within the audience, providing niche insights on each group. This helps optimize content type/reach, increase campaign ROIs, enhance lead generation, or uncover new market opportunities.

How does Audiense Insights cluster audiences?

Audiense offers two main types of segmentation:

1. Interconnections (clusters that are connected, who follows who, to find out what glues them together as a community).
2. Affinities (interest-based clusters, those following similar sets of accounts, therefore with similar interests).

Individuals can only be members of one cluster. So once the pattern of multiple clusters has been observed, each individual within an audience is assigned to the closest matching cluster. These methods identify naturally occurring clusters within your audience without bias.

1. Interconnections clusters

Interconnections clusters individuals based on "who knows who," i.e., how these individuals are interconnected, using the Louvain algorithm which is based on relationships (communities). This method looks at who engages with whom and clusters them together. For instance, if person 'A' follows person 'B', they’ll be clustered together. This helps to understand the personas of associated people who are connected due to mutual relationships. 

Best For:

• • Finding niche clusters within your audience
  • Location-specific analysis
  • Professional cohort discovery
  • General market overview
  • Finding viral group potential
  • Both B2C and B2B contexts, as it can break down markets into manageable clusters

This method selects the best-fit number of clusters based on discovered relationships. The results can provide between 4 and 10 clusters by default, but you can request additional clusters at the end of your cluster cards by clicking on the +, and add up to 20 if they are available.

• Cluster Size: The size of each cluster in the graph is proportional to the number of members it contains. Larger clusters indicate a higher number of members.
• Edges (Links/Lines): These represent the connections between nodes. For example, in the Interconnections graph, the edges show follows exclusively among audience members.
• Interconnections Graph: Nodes represent audience members, and all connections (edges) are exclusively between them. This graph visualizes how members of the audience are connected to one another.
• Node Size: The size of each node corresponds to the number of connections it has. Larger nodes have more connections. Hovering over a cluster in the interface highlights the connections for that cluster.

2. Affinities clusters 

Affinities provides unsupervised clustering, which breaks users out into clusters based on their unbiased interest behaviors due to their following patterns. Individual users are grouped into a single best cluster based on the handles they choose to connect with. 

This clustering method uses the K-means algorithm. It is useful for grouping items based on similar properties, instead of relationships. The main goal of this algorithm is to cluster data into several groups based on similarity.

Once the algorithm detects a group of users with a pattern of following a handful of the same accounts, that group becomes the signal for a particular cluster.

Best For:

• Brand segmentation

• Fandom analysis

• Partnership discovery

• Analyzing larger audiences to understand broader consumer media strategies

This method lets you select the desired granularity for uncovering hidden clusters. You can adjust between 2–20 clusters, or go with the recommended number based on the estimated size of your audience. Select fewer clusters to break out high-level clusters, or more clusters to uncover niche segments.

• Cluster Size: The size of each cluster in the graph is proportional to the number of members it contains. Larger clusters indicate a higher number of members.
• Affinities Graph: Nodes represent both audience members and influencers. Edges indicate connections not only between audience members but also between audience members and influencers.
• Node Size: The size of each node corresponds to the number of connections it has. Larger nodes have more connections. Hovering over a cluster in the interface highlights the connections for that cluster.
  
  

How to access and use Audience Segmentation?

1. After you define your audience, you will proceed to the segmentation step before creating an insights report.
2. Select either Interconnections or Affinities

1. If you select Affinities, choose the desired number of clusters between 2 and 20 to tailor the analysis according to your requirements, or opt for the recommended number of clusters, which is based on the estimated size of your audience, and launch your Insights report
   
   
   
2. For Interconnections, you won’t be able to choose the number of clusters, as we will deliver only those that are connected within the audience, which can result in 4 to 10 by default, but you can request additional clusters, up to 20.

Learn more about our Audience Breakdown Page (incl. cluster summaries, renaming clusters, merging clusters, and more).