1. Customer Relationship Management (CRM) Chapter 4 Customer Portfolio Analysis Learning Objectives Why customer portfolio analysis is necessary for CRM implementation That there are a number of disciplines that contribute to customer portfolio analysis (CPA): market segmentation, sales forecasting, activity-based costing and lifetime value estimation How the CPA process differs between business-to-consumer and business-to-business contexts How to use a number of business-to-business portfolio analysis tools The role of data mining in CPA

2. Data mining for Market segmentation: Data mining can be defined as the creation of intelligence from large quantities of data. Specially on the CPA stage: It is used: For market segmentation Customer valuation purposes Particular use when there are large volumes of data that need to be analysed. This is true of B2C contexts such as retailing, banking and home shopping; such TESCO has 14 million Club card members in UK customer base Club Card includes: Demographic data that the customer provided to become a club member. Transactional data linked to the card member. If 10 million members use Tesco in a week and have an average basket of 30 items, Tesco’s database grows by 300 million pieces of data per week. This is certainly a huge cost, but potentially a major benefit

3. Data mining: Questions that need an intelligent answer in the CPA stage of CRM strategy development are: How can we segment our customers? Which customers offer the greatest potential for the future? Clustering techniques can be used to analyse complex data sets with a view to identifying segments of customers. Various clustering techniques are available for the segmentation task. Technique includes: They form clusters by allocating objects to groups (the customer is the ‘object’ in market segmentation research). The composition of a group is relatively homogeneous. An object is allocated to a group because it possesses attributes that are more closely associated with that group than any other group

4. Clusters: Once clusters have been formed they need to be given a CRM interpretation. Lifestyle market segments are outputs of cluster analysis on large sets of data. Given the availability of transactional data, it is certainly possible to segment according to historical value. Transaction data can be converted into revenue or margin data. Customers can be segmented into quintiles or deciles according to the revenue or margin value of the purchases they make. In general, it is assumed that the cost-to-serve is standard across all customers, so no further analysis is done to convert margin data into true customer profit data.

5. Cluster Technique: In principle, some customers do have a higher cost to-serve because, for example, they make more demands on the contact centre. In practice, in B2C markets it is often too difficult and costly to introduce procedures that trace costs to customers. Each decile or quintile will include different customer types. Two customers with the same historical value may have generated that value in very different ways. One may have made multiple repeat purchases of household staples; the other may have made a couple of high-priced speciality purchases. Customer relationship management strategists are more interested in what future value a customer (segment, individual) can yield. This is determined by their propensity to buy products in the future. Data mining expertise can be used to build predictive models.

6. Data mining Technique: Data miner examine historical data sets to generate predictive models about future behaviour. Predictive models can be generated to identify: which customer (segment) is most likely to buy a given product which customers are likely to default on payment which customers are most likely to defect (churn). Training the model: Predictive models are built using data from the past. The data include predictors and outcomes, both of which are known as training the model. Scoring: When the model seems to be working well, it is run on contemporary data, where the predictor data are known but the outcome data are not.

7. Predictive modelling is based on three assumptions The past is a good predictor of the future... BUT, this may not be true. Sales of many products are cyclical or seasonal. Others have fashion or fad lifecycles. The data are available... BUT, this may not be true. Data use to train the model may no longer be collected. Data may be too costly to collect, or may be in the wrong format. The database contains what you want to predict... BUT, this may not be true. The data may not be available. If you want to predict which customers are most likely to buy mortgage protection insurance and you only have data on life policies, you will not be able to answer the Question Two tool that are used: Decision Trees Neural Networks

8. Thank You!!! Q&A