Architecting Personalization: Part I

May 29, 2015

An examination of the high-level technology and processes needed to create a personalized experience and how to incorporate it into an analytics/conversion rate optimization program.

Editor’s note: We’re going to be hosting a roundtable at Opticon in June. Come on by and meet our author!

Personalization. Everybody wants it.

As marketers are able to collect more and more behavioral data about its consumers, personalizing a site or campaign experience has become an essential piece of any marketing strategy. In particular, personalization can greatly impact how effectively a marketer reaches a consumer and significantly contributes to the optimization of conversion rates. Further, recent studies indicate that a well-executed, personalized experience is expected by consumers. A study by Janrain found that 74 percent of users become frustrated with websites when content, offers, ads, promotions, etc. appears to have nothing to do with their interests. A different study from Accenture highlights customers’ willingness to forgive some of the perceived negative aspects of site tracking in order to receive a more personalized experience. 85 percent of those surveyed indicated they were aware that such tracking goes on but understood that tracking enables companies to present offers and content that matches their interests. This is despite the fact that 86 percent of those surveyed said they were concerned about websites tracking their online shopping behavior. These responses are indicative of the need for marketers to understand the personalization landscape and develop an effective personalization strategy, to avoid significant hits to a company’s perception and ultimate conversion rates.

We’re going to show you how it’s done.

Because the coordination of personalization and optimization is such a critical part of a marketer’s overall strategy, we’re going to dive deep into this topic with a comprehensive 3 part series of posts.

Part I of this post will examine the high-level models and processes that are necessary in order to deliver a personalized customer experience.

Part II will review and discuss many of the specific services and technologies used by marketers to both optimize and personalize their site.

Part III will review some of the methods and challenges for measuring the effectiveness of personalized content and incorporating personalization into one’s analytics and CRO strategies. This part will conclude the series with some best practices and suggestions to better understand the impact of your personalized marketing efforts.

Let’s begin the first installment, by shedding some light on what people mean by personalization.

Defining Personalization

A quick survey of some of the experts reveals a variety of definitions that address different aspects of personalization. Forrester broadly defines it as “the ability to provide content and services that are tailored to individuals based on knowledge about their preferences and behavior” [“Smart Personalization,” Forrester Report by Paul Hagen, 1999]. Others define it as “the capability to customize communication based on knowledge preferences and behaviors at the time of interaction” [Jill Dyche, CRM Handbook, Addison-Wesley, 2002] or as “a way to generate customer loyalty by building a meaningful one-to-one relationship, understanding the needs of each individual, and helping satisfy a goal that efficiently and knowledgeably addresses each individual’s need in a given context” [Doug Riecken, “Personalized Views of Personalization,” Communications of the ACM, 43(8), 2000].

A Five-Part Process

If we consider the various aspects described in each of these definitions collectively, we can define personalization as a process that is made up of 5 main components:

  1. The individuals/consumers.
  2. The content that is altered or varied based on certain criteria.
  3. Observations or data about certain behavior or preferences
  4. The provider or technology that is managing and serving the content to the consumer
  5. The goals that are driving the entire process.

With Three Methods

There are 3 main types of personalization: Provider-centric, Consumer-centric, and Market-Centric.  Each type is characterized by the different interactions between the 5 components of personalization discussed above.

The Provider-Centric Type

In provider-centric personalization, each provider uses its own personalization engine or algorithm to segment consumers and deliver content, such as personalized Web pages and links to the consumer. This is the most common type of personalization and has been largely popularized by companies like, Netflix and Pandora. The 2 main goals in this approach try to find a balance between creating the best site experience possible for the consumer while maximizing the profits for the providers in the most efficient and financially viable way possible.

The Consumer-Centric Type

The second method, consumer-centric, assumes each consumer has its own personalization engine/algorithm that connects to providers based on this knowledge. This method is often referred to as an e-butler service and the goals and benefit of the personalization are entirely consumer based.

The Market-Centric Type

The third approach, market-centric, provides personalized services for a marketplace within a certain industry or sector. An example of this would be a personalized portal where customized services offered by corporate partners are matched to the individual needs of their consumers.

Visualizing Personalization Methods

An often-cited academic overview on the topic of personalization provides an excellent visualization of these three methods. Here, the providers and consumers are represented by open white boxes and the personalization engines/algorithms are in illustrated with grey boxes.


(from Adomavicius, Gediminas and Tuzhilin, Alexander: “Personalization Technologies: A Process-Oriented Perspective”, University of Minnesota and New York University)

Different Methods, Common Approach

Despite the differences described above, each of these three models follow the same overall process for personalizing content: goal setting, understanding the consumer, content delivery, and measurement/evaluation.

Step 1: Goal Setting

Goal setting is probably the most important stage in the personalization processes it impacts each subsequent step along the way. This is especially true for the most common, provider-centric model of personalization that most marketers follow. It is critical at this point for the marketer to first ask what is the end goal of the personalization program? What KPI should be impacted through this type of optimization? Once one answers these questions, determining the content to be personalized, the target audience, and what KPIs will be used to measure success should be a lot clearer.

For example, the goal of personalization at Netflix is to “maximize member satisfaction and month-to-month subscription retention.” The content they are optimizing is the video selection presented to the consumer, and success is measured via a number of KPIs including cancelled subscriptions, interactive sessions resulting in a playback, played minutes per user, video completes, browsing time, and return visits. (Pancrazio Auteri “10 Lessons Learned from Netflix”).

Step 2: Understand the Consumer

Once the goals and target audience have been defined, the marketer needs to understand the consumer in a way that will impact the marketer’s KPIs. Much of this process revolves around how to collect and analyze data about the consumer in order to make relevant, actionable content recommendations. First, the data for these models is first collected in one of two ways:

  • Intrusive:  This method asks customers along the way questions (i.e. 5 star ranking systems) to help add data to better fit the recommendations
  • Non-intrusive: Here, the behavioral data of the consumer is collected in a strictly observational (browsing behavior, past purchases, etc.) way. Consumer preferences are then inferred based on observations such as repeated activities, search terms, etc.

Then, there are two main ways/types of algorithms used to process the data, categorize or segment the consumer, and arrive at recommendations:

  • Heuristic-based techniques create recommendations based on previous transactions or ratings made by the consumer. Using Netflix again as an example, a heuristic-based recommendation engine would recommend movies that mostly closely match the tastes indicated in the movies the consumer has already watched or ranked.
  • Model-based techniques arrive at recommendations by using observed behavior (collected either intrusively or non-intrusively) to calculate a probability that a consumer will like a certain piece of content. Adding to the above movie example, a model will use the movies the consumer has already seen to estimate the probability that he/she will like each of the unseen movies available.

Step 3: Content Delivery

Finally, the consumer is presented with content or recommendations. There are usually two main types and each of these recommendations can be arrived at using the heuristic or model-based techniques described above:

  • Content-based: the consumer is recommended products or shown content similar to products or content he/she has viewed, purchased, or interacted with in the past.
  • Collaborative: the consumer is presented content/products that people with similar tastes and preferences have liked in the past.  Collaborative find the consumers most likely peers and bases recommendations on the most liked content/products from that group.
  • Hybrid: recommendations that draw from both content and collaborative models

Step 4: Measurement

This is a topic that requires its own post, so we’ll be covering it in part III of this series.

Putting it All Together

The chart below shows how these processes are structured as a feedback loop. As marketers learn more about the consumers and how the content presented impacts behavior, goals will need to be adjusted to align with the new insights.


The definition, models, and data analysis techniques described above provide a very top-level summary of the landscape of the technologies and processes used in personalizing content. Sections II and III of this series will attempt to dive deeper and provide more information about the specific technologies used in personalization as well as the measurement techniques and challenges incorporating personalization into an overall optimization (CRO) program.

Have questions about CRO and personalization? Reach out to us, we’re happy to help!

Hey There!

Thanks for reaching out.

Job Title
What're you interested in?

We'll be in touch shortly.

Until then, why not browse some of our work?

See our work