Crio Bru Segmentation

RStudio, Clustering Algorithms, Classification, Consumer Segmentation, Feature Engineering, Decision Tree, Random Forest

In this project I used data that had consumers preferences for the coffee replacement brand, Crio Bru. With the data I used clustering algorithms to create consumer segmentation based on similar preferences that were distinct from other consumers. To determine these different consumer preferences I used three main attributes and multiple levels within those attributes to determine different segments.

I first created an elbow plot as a starting point to see how many different segments should be created. As can be seen below the elbow plot begins to decrease at two clusters, but then decreases more at four.

I compared solutions between four to six clusters, and felt that four clusters didn’t overlap in attributes compared to five and six clusters. Using four clusters, I created four profiles to help Crio Bru know how to target different segments of their consumers. The four profiles I created were:

Mint & Energized in Utah – This segment has the largest population in Utah. They value energy as the most important attribute of Crio Bru. Unlike the other profiles, this profile enjoys the flavor mint heavily above all other flavors.

Best of Both Worlds – This profile is the most price sensitive, but they also care the most about high quality and a trusted brand name. The individuals want the best of both worlds, which is the highest quality for the lowest price. They are the only profile that prefers the medium size over the large size. A reason to explain this could be their price sensitivity.

Living Large – This profile is the least price sensitive and earns the largest income. They prefer the large drink over the other sizes and are willing to spend more. They live large without having to worry about price sensitivity and can manage to get any size they want. They care about the features of Crio Bru, but not as much as the other profiles.

Health Freaks – This profile cares the most about Crio Bru’s features of health benefits, taste, antioxidants, all natural, and additive free. All these features are regarding to health, making this profile the health freaks.

I then used the feature engineering to fit a decision tree and a random forest to see the accuracy of predicting which profile a consumer would fall into based on their preferences of Crio Bru. After initially running a decision tree and a random forest, I then tuned the parameters of both attempts to build the most accurate model. After tuning the random forest model, the model produced 92% accuracy of predicting which profile a consumer would fall into based on their preferences. This model can help the brand Crio Bru use marketing automation to target specific ads towards the different profiles highlighting the attributes most important to the consumer.