Connector

...

Question 6

a.

Confusion Matrix (Table 3.4) shows that the predicted customers for cluster 4 are likely to be 91% correct. This strengthens our conclusion that ConneCtor seems to be a good fit for cluster 4 - affluent professionals. This cluster desires even more features than ConneCtor currently offers. Albeit, the product in its current form offers enough value for these customers to pay a good price and replace their current PDA with a Connector. We must build up on its current form and make improvements on preferences desired by this segment, e.g. lighter weight multimedia capable device, in order to penetrate this cluster more effectively.

Connector is also a great fit for cluster 1, however cluster 1 will not use all its features and is unwilling to pay enough. Therefore, the firm could develop a more basic version of the product to cater to cluster 1. We recommend 2 tier product roadmap based on our analysis for the firm.

b.

In terms of quantity, we have enough data for a reasonable indication that our product will work with cluster 4. We should survey more people that fit the discrimination criteria of this cluster in order to gain more confidence in our ability to sell to them.

In terms of quality, we should test how likely cluster 4 will buy a device that offers a slightly smaller feature set than they desire. E.g. Connector does not have multimedia features. It would be interesting to see how much cluster 4 will pay once they realize this feature is taken away.

Supplemental Questions:

Question 1

Hierarchical clustering gives us the most sensible results in this case. Hierarchical segmentation starts with each data point in its own segment, iteratively combine two closest segments, and “stop” at a certain point. This gives us a sense of how different or similar the segments are, and also helps us determine how much information will be ignored if clusters are combined (through dendogram). This helps identify appropriate number of clusters well.

k-means algorithms are calculated based on the followings:

1. Number of clusters must be known

2. Each data point gets assigned to a cluster which is closest

3. Recalculate cluster centers by finding mean of data points belonging to the same cluster

4. Repeat step three until cluster centers start shifting

k-mean fails for non-linear data sets.

Question 2:

There are two possibilities here.

1. One or more of the other clusters give a clear indication of the product that needs to be built, are large enough that they are worth going after, and have clear discriminant data. In this case, we should create a product that caters to these other clusters and target them.

2. None of the other clusters are either large enough to go after or clear enough in their product preferences and discriminant data. In this case, we need to reject the five-cluster solution and continue segmentation because unless we are able to identify who the customers or what are their characteristics, targeting them will not be possible.

APPENDIX

1. Hierarchical Segmentation with 2 clusters

Table 1.1 Cluster Sizes

Size / Cluster

Overall

Cluster 1

Cluster 2

Number of observations

160

93

67

Proportion

1

0.581

0.419

Table 1.2 Segmentation Variables

Segmentation variable / Cluster

Overall

Cluster 1

Cluster 2

Innovator

3.47

4.27

2.37

Use Message

4.21

3.6

5.04

Use Cell

5.56

5.84

5.16

Use PIM

4.01

5.09

2.51

Inf Passive

4.45

4.47

4.42

Inf Active

4.5

4.53

4.46

Remote Access

3.99

3.18

5.1

Share Inf

3.71

3.29

4.3

Monitor

4.79

4.34

5.42

Email

4.72

5.82

3.21

Web

4.47

5.78