1	Management Summary

2	Personalization Conception

2.1	Automated Personalization

2.1.1

Genesis

2.1.2

Techniques

2.1.3

B2C and B2B

2.1.4

Limitations and Challenges

2.2	Influence on Customer Life Cycle

2.3	CRM Processes Opened to Personalization

2.3.1

Marketing

2.3.2

Sales

2.3.3

Customer Services

2.4	Protection of Privacy and Personal Data

3	Personalized Customer Channels

3.1

Web

3.1.1

New York Times

3.1.2

FedEx

3.2	Mobile Devices

3.3

Telephone

3.4

IVR

3.5

Email

3.6	Printed Materials

3.6.1

International School of Management, Dortmund Study

3.6.2

Union Gospel Mission of Seattle Project

3.6.3

Rochester Institute of Technology Study

3.6.4

The Negative Effects of Personalization on Response Patterns Study

3.6.5

Other Relevant Projects

3.7	Interactive TV

References

"2.1.2 Techniques
Before we outline techniques that are currently used for personalization delivery we should mention the essential prerequisite for each one. For all techniques that use a customer profile it is extremely important to have the profile complete and up-to-date. The issue is not only about the integration point, i.e. a profile that covers all the departments’ points of view and all the attributes in all the databases, but it is about sufficient coverage of the business situation for which that particular technique is used and that is expressed in profile schema as well.
In practice there are several approaches toward automatic personalization implementation. The first one concerns rules-based filtering; the second concerns content-based filtering, and the third concerns collaborative filtering. An approach expressing implicit customer information gathering is known as a non-intrusive personalization. Artificial intelligence systems and a variety of analytical tools can be used as tools for all the techniques mentioned above.
Rules-based filtering systems are run using demographics and statistical data (e.g. from registration process) connected to the target subjects. A system of rules based on specific context and customer profile’s attributes links appropriate content and presentation for communication (e.g. Siebel Applications).
Content-based filtering links appropriate content and presentation according to the affinity of the customer and the object profile. An important prerequisite is the ability to implement the same representation schema for both the customer and the objects or translation methods. The relevance in this case is a level of affinity or similarity between both profiles. An excellent example here represents ArtMecca.com, where based on a similarity of attributes (in this case numerical attributes), and their position in a proprietary profile system, the ArtMecca.com recommendation engine can offer users another art object that should be appealing (the first step in the process is projection of the customer’s profile into selected art object). Such a technique could be successfully implemented in confection, etc. [Dav00]. The disadvantage of content-based filtering is the costly process of generating profiles for all of the objects, if there is any reasonable and usable technique and method. From the target subject point of view, zero future diversification, and new target filtering, where no past record exists, are issues to be dealt with.
Collaborative filtering (e.g. Net Perceptions) seems to overcome the time lock of content-based filtering. This model overcomes the experience of a single user through the following hypothesis. If we can find for a target subject another subject with a similar past profile we can suggest evaluations for the second target that are based on the past profile of the first. One advantage is that there is no need for explicit objects’ profiling and the more subjects we have in our system the more appropriate recommendations we can provide. However, correspondingly there are two major challenges with this technique. Because we can recommend only what was evaluated in the past the first (latency) challenge corresponds to the inclusion of a new object into the recommendation system. If we find a subject for whom we cannot find any similar profile in our database we come across the issue of an abnormal subject. The main added value of collaborative filtering lies in implicit similarities (experience, knowledge, interests, evaluations and preferences) among participating subjects.
Both major techniques - collaborative filtering and content-based filtering - have their disadvantages. However, in combination they form a remarkably harmonious system , where the disadvantages of one system are the advantages of the other and vice versa (e.g. latency issue of collaborative filtering is solved through content-based filtering and diversification issue of content-based filtering is solved through collaborative filtering)."

Práce obsahuje tabulky a schémata.