All public courses using SAS are held at SAS, Marlow:

and cost for two-day courses 1090 GBP (plus VAT) per person which includes lunch and refreshments. Reduction to 980 GBP (plus VAT) per person for multiple bookings from the same organisation.

To book a place on a course either click here, or phone +44 (0)118 934 4265, or print and complete an application form (pdf format 115 Kb) .

Course dates for 2008/2009 :

• Propensity Scoring Models: 13th-14th Nov 2008, 12th-13th Feb 2009, 11th-12th June 2009, 5th-6th Nov 2009
  
• Segmentation of a Customer Database: 3rd-4th Nov 2008, 26th-27th March 2009, 24th-25th Sept 2009
  
• Planning & Interpreting of Marketing Campaigns: 18th-19th Sept 2008, 4th-5th June 2009
  

Segmentation of a Customer Database

using the SAS^® System

The Course

How should I group my customers? By how they look? By what they do? Using what kind of data? How many groups should there be? Too many and they will be uncomfortable to manage, too few and important differences will be lost. What kind and size of sample should be used to determine a viable segmentation? How can that be extended from the sample to the rest of the database? We discuss the different ways that the data can be pre-processed, how business knowledge can be used to help to drive the solution. We will look at different ways for describing and profiling customer segments. We concentrate on the main clustering tools provided within the SAS^® System, namely PROC FASTCLUS and PROC CLUSTER, though we will also consider discriminant analysis (PROC DISCRIM) for allocating customers to clusters when direct scoring methods are not possible and PROC CORRESP and PROC PRINCOMP for presenting visually the relation between segments and descriptive variables. You will get the opportunity to work through the whole process guided by the instructors, where necessary.

We will present all the tools that you need during the first day of the course. We discuss the different types of segments you might want to create: by lifestyle or demographics, by transaction behaviour or by attitude, and bring out their business pros and cons. The course will then concentrate on using customer transactions or customer demographics, or a combination of the two, to create discrete groups of customers. Using data of realistic structure we will explore what kind of data pre-processing (eg principal components or correspondence analysis) might enable the grouping process to work more effectively and we consider what level of detail will yield the best results. We will show that there are opportunities to incorporate business knowledge into the process, perhaps by aggregating variables in an intelligent way, perhaps by giving extra weight to some of the variables.

The segments will usually be determined by working with a sample of customers and we will discuss the different means by which the remainder of the customers can be allocated to a segment. This allocation process has to be flexible enough to accommodate 2-stage clustering, where a largish number of clusters is produced by a disjoint clustering method and these are then reduced to a smaller number by a hierarchical method.

Once the segments have been created, what then? How can you best display the properties of the segments? We will show you how to use visualisation techniques like correspondence analysis to present segment properties pictorially; we will discuss different approaches to calculating indexes which can bring out the salient features of the different customer groups.

During the course there will be several short exercises where you will practice the techniques which have been discussed. There is also an ongoing workshop throughout the course when you can work on data of realistic structure and quantity to solve an extended segmentation problem. The final session of the course will take the form of a brainstorm discussion where you will have the opportunity to present any special problems you may have, or have discovered during the workshop, and exploit the experience of the trainers and the other participants.