Mining Your Own Business in Retail Using DB2 Intelligent Miner for Data

94 Mining Your Own Business in Retail Using DB2 Intelligent Miner for Data

Figure 4-19 Scoring using the data mining tool directly

There may be situations where it is not possible to perform this operation directly

from the data mining tool or where scoring all of your customers is either

unnecessary or desirable. For example, the clusters may have been developed

using a subset of customer records held in a separate database from the one

containing the customer records you want to score. In such cases the use of the

second method of scoring may be more appropriate. This is discussed below.

4.7.3 Using the cluster results to score selected customers

While scoring all of your customers can be performed as some form of batch

process, if you have large numbers of customers this may introduce a significant

overhead on database performance. If you have to score all your customers this

may be unavoidable, but in many situations there may only be a subset of the

customers that it is appropriate to score at any one time. This is increasingly true

for e-commerce applications where you may only want to score customers as

they use a particular service. Using the exported PMML models, it is possible to

score selected customer records directly within the database. This type of

scoring is performed using available database functions and therefore can be

done automatically, for example, as customer records are being updated or at

Chapter 4. How can I characterize my customers from the mix of products that they purchase? 95

specified times. An additional advantage of this approach is that the PMML

models themselves are stored in the database, and therefore the configuration

control of which models were used and when, can be handled by standard

database administration procedures. The process is illustrated in Figure 4-20.

Figure 4-20 Scoring customers using exported PMML models

Using this type of approach for deploying data mining solutions into your

business offers many advantages. It makes possible the deployment of the same

models into distributed applications being used across your business. The same

models, although developed by your operational research department, could be

being used by your sales and marketing organization (in OLAP and other forms

of reporting tool such as Business Objects, BRIO and so on), at the same time

the models can be deployed into your CRM systems (kiosks, Web-shopping and

similar customer touch points) and as part of your campaign management tools

(direct mailing, telesales and so on). As the models are updated or improved,

they can easily be replaced in all of the tools that you are using and consistency

is then maintained.

96 Mining Your Own Business in Retail Using DB2 Intelligent Miner for Data

There are an increasing number of applications that can make use of the

capability to deploy data mining models in this way and as e-commerce expands

in scope, there will be many more opportunities to exploit this capability. In

Chapter 6, “How can I decide which products to recommend to my customers?”

on page 137, we give another example of how this capability can be used as part

of a personalized product recommendation system.

What needs to be stressed is that the deployment of the results of data mining

into your business can bring significant benefits and wide ranging benefits. It is

worth spending time considering how you are going to use the results before

embarking on any data mining activity.