Introducing Metrisim's new market segmentation module. Easily segment your market using a cluster analysis of partworths and automatically generate filters that can be applied to scenario runs.

 

You decide on how many clusters (segments) you want, and the cluster analysis algorithm will find them. Ultimately the goal of the cluster analysis is to group survey respondents into clusters where individual respondents in a cluster are as similar to each other as possible and as different as possible to respondents in other clusters. This can be used to develop different products for each segment to better meet the requirements of each segment. It can also aid in developing different advertising messages for each segment.

 

How many clusters?

Instead of a single mass market approach, segmentation may be an improvement. However, it is more costly as organizations must spend more on R&D and there are issues of skills and operational capacity to consider. Additionally you need to consider total sample size. Small survey samples should mean using less clusters if you want to generalize results from each cluster. The smaller the sample the larger the standard error for a given measurement approach and its associated statistic.

 

How does it work?

Metrisim uses a K-means cluster analysis algorithm. If you want ‘K’ different clusters, the algorithm selects initial centroids (an estimation of the most ‘typical’ respondent at the center of each cluster) using a newer approach which is more efficient than random selection from a uniform distribution. Each respondent is assigned to its closest centroid based on a measure of distance to each (Euclidean distance). The centroids are then recalculated as the mean (average) of each cluster, and then the distances are recalculated and then reassigned to new clusters again. This process keeps iterating until it converges on a ‘stable’ solution – meaning no or very little change in the calculated centroids from iteration to iteration, or a preset maximum is reached.