CI professionals have expertise in data, technology, and analytics — all important ingredients of personalisation. But to make the analytically driven personalisation work, they must :

· Prioritize what to personalize. Marketers have a massive choice of assets to personalize for the customer — web pages, product and service recommendations, email, dynamic prices, discounts and offers, and marketing and advertising messages. Prioritize what to personalize by:

1) distilling the intended outcome, such as improving online shopping experience or improving discount redemption rates, and 2) looking back at A/B tests for clues to positive customer responses that explicitly show how an offer or piece of content worked better than others.

• Estimate customer’s expectation of personalization. Thanks to Amazon-like shopping experiences, customers expect a higher degree of personalization from digital channels than other channels. Before embarking on a significant technology investment for personalization, understand the level of expectations from customers though preference surveys, feedback forms, and even face-to-face interactions.
• Distinguish between known and unknown attributes. Successful personalization depends to a very large extent on using known data about personal preferences and behaviors of existing consumers. But when targeting prospects, firms need to rely heavily on analytics to build proxies for the unknown attributes of prospects. For example, this might mean judging a user’s interests based on recently visited websites or based on other known details that are similar to those of existing customers. With every accepted or rejected offer, prospects are giving more clues about their preferences.
• Understand the level of analytical complexity. The analytics complexity to execute personalization ranges from basic segmentation to real-time self-learning personalization. The level of sophistication in personalization depends on where the firm is in its analytics and technology adoption. For instance, to achieve real-time, self-learning personalization, the
  firm must already have an advanced data infrastructure that seamlessly ties all customer and prospect data together. Work backward from what personalization is supposed to drive and then choose the appropriate analytical methodology.
• Choose the interaction context. The interaction context is the channel, either inbound or outbound, though which the customer or firm initiates a contact. To orchestrate cross-channel personalization, stitch the various channel deployment technologies together by evaluating the integration capabilities of single-channel vendors, such as Adobe Omniture, Baynote, or X Plus One, and multi-channel vendors, such as FICO, IBM Unica, Infor, or SAS. 

http://success.adobe.com/assets/en/downloads/forrester/20460.Forrester_Research_Use_Customer_Analytics_To_Get_Personal.pdf