What Is A Market Segmentation?
For any business, a market segment is a group of customers who share the same interests, behaviors, traits, characteristics, needs, wants and expectations. Segmentation is a consumer-oriented process and can be applied to almost any type of market.
Market segmentation helps companies improve ROI, increase profits, reduce costs, successfully develop new products, enter new markets and server their customers better.
Market Segmentation Techniques
The segmentation technique will depend on your goals, the level of segmentation, available customer and industry data, the extent of market research and analysis you choose to pursue, and whether your products or services are intended for a retail market, business-to-business (B2B) or a mix. Market segmentation can be carried out using geographical, demographical, behavioural and psychographic data variables and the source of the data could be transactional, clickstream, CRM and survey data. Good segments have high inter-segment heterogeneity, within segment homogeneity, measurable, reachable, and profitable.
Segmentation Based On Machine Learning
Here are three different approaches to machine learning based market segmentation.
The best approach to a segmentation will depend on the type of data you are working with and how you plan to use your results. The below mentioned techniques can be implemented using coding languages and tools such as Java, R, SPSS, Python, SAS, Neo4j, etc.
- If your data is mostly continuous…
K-means is a very popular segmentation algorithm that clusters respondents based on how similar their data is to each other. It works best with continuous numeric data, for example, a customer’s transaction amounts, age, or annual income. K-means creates k “centroids”, or combinations of data points, and then classifies your respondents based on which centroid each they are closest to. The algorithm will adjust the positions of these centroids and re-classify until it has a stable model. K-means is popular because it is very easy to classify new respondents and does provide some information about how to describe each segment.
- If your data is mostly categorical…
Latent class modeling uses categorical data to find traditionally unobservable–or latent–variables to distinguish respondents. Examples of categorical data are income bracket, profession, or region. This analysis uses probabilities to show which variables in your data contribute to each segment. Latent class is very helpful for understanding an existing group of respondents but can be difficult for segmenting new respondents.
- If you want to understand segments and potential subsegments…
Hierarchical clustering creates many layers of segments, where each respondent is paired with the respondent, they are most similar to. Then, these respondent pairs are grouped by which pairs are most similar, and so on, until all respondents are in the same group. Hierarchical clustering is helpful for understanding how segments may break apart into smaller segments.
At ENGINE, segmentations have helped understand the different aspirations of Millennials & Gen Z, and to target product development to improve conversion rates. ENGINE helped GMAC in building segments of graduate schools, including international schools, based on applicant attributes, scores, industries, geographies and college profiles. We then built AI models to find out the probability for applicants to apply and be accepted for colleges in each segment. GMAC can use these models and segments to sell best applicants to their client colleges, thus bridging and connecting colleges to right applicants.