The parameters OR attributes on which you can segment a customer are listed in a separate topic and also indicated in the customer dimension. Add to that list the following components of your dimensional model:
Dimensions on which you can slice OR dice a customer segment:
- Customer segment details for a given location
- Customer segmentation for a given channel. This means that you can do segmentation of the customer who will be approached by a certain channel, say your third party distributors.
- Customer segmentation for a given product.
- Customer segments for a given line of business.
- How a customer segment behavior changes with time.
Examples of Measures linked to customer segmentation analysis
- Number of customers
- Sales Revenue value
- Sales revenue numbers
- Number of Years in a location, job
Typical Analysis which you may do for customer segmentation BI
Customer segment Behavior analysis-
For each customer segment, what are various measures like sales revenues, sales profitability, average customer tenure etc.. This analysis helps you to validate the assumptions you have made on a customer segment behavior. Over months and quarters, you can re-validate these assumptions and fine-tune your strategy. For example, if you find that a given segment is buying more than expected, you may heighten the advertising budget for that segment. You may also like to go back and check your methods which you deployed in the previous customer segment to behavior analysis.
Customer segment movement over time:
You can analyze on how the customer moments happening across the segments over time. For example, over a year- %age distribution of your customers in various segments could change. This information will tell you, if your customer acquisition is as per strategy. Taking this example further- If you find that % of low-income, low-usage customers have gone up over last one year, it may not be matching with your strategy to go for high usage customers.
Another example is – Tracking on how are we converting high risk customers to lower risk, and medium value customers to higher value? For example- one of the measures of success of a 'customer value' initiative could be on how many customers have moved from low value to medium value.
The way to achieve it is to have a time-stamp as the attribute to the customer dimension. For example, if you are having a way to assign the Customer segment class every month, a new record will be added with a time-stamp.
New Customer segment growth:
This analysis talks on the profile on new customers which are acquired. For example- one may note that while existing customer portfolio is fairly moderate risk, the customers acquired in last three months are primarily high risk. This becomes important because in a normal analysis, last three months acquisitions may not move the averages, but it serves the shareholders to know that sales function is suddenly getting over-aggressive in acquisition..
With-in Customer Segment Movement:
If you are looking for a short horizon, you may not be able to move customers across the segments. Therefore, it is important to note the movements within a segment. For example one of the measures of success of a 'customer value' initiative could be to increase the average spend in the 'low value' segment by 10%.
Validating your results with a 3rd party research data
Another area you may run into is to compare your existing customer segmentation patterns (and also the marketing database data) with the studies done by 3rd party research outfits. For example, if a 3rd party research reports is saying that a given customer segment is generating 25% of the total revenue in the industry, and your own experience out of 1 million (say) customers is coming out to be different. This means that there is a difference of assumptions, methodology OR the reference data. One needs to explore these differences, if they are significant.
Business Modeling/Data Mining
Data Mining for Customer Segmentation:
Please go through the customer dimension to get a back-ground. Data mining is used to create the models to assign “Classes” attributes to customers. This modeling will frame the rules, assign scores, assign class category to each customer. One can use cluster analysis, nearest neighbor & rule induction techniques to arrive at the classes (these techniques will be explained in data mining section, planned to be published by June, 2008).
Once the customer class is assigned, the data mining can now work on clustering the customers into segments. One should understand that there is a difference between the “Class” and “Customer Segment”. Class is the input to find customer segment. For example the “payment/renewal history” class, “Year at location/address” class, “Years at current job” class, “product relationship “ class etc.. ALL of these classes contributed to work out the risk profile of the customer.
Data Mining to characterize the customer behavior of a given segment
The main purpose of segmentation is to find the expected behavior of a customer segment, so that we can devise actions to entertain that customer behavior. Data mining techniques can be used to develop behavior predictions along with the level of confidence and level of probability. This subject will be dealt in greater depth in Data Mining section (expected to be published by June 2008)