Skip Navigation



Semantic Segmentation to Improve Personalization

3/16/2018 | Chris Baribeau

We all know that in the Age of Experience, personalization is no longer an option, it is now required to acquire, engage, and grow customers. However, many personalization efforts are falling short. In this paper, we will help explain the most common pitfalls of personalization and describe a new approach to improve the usage of your data to deliver more relevant experiences. Said differently, we will show you how to generate a higher return on the millions of marketing dollars that you are already spending to drive visitors to the web.

According to Forrester, 62% of consumers have chosen, recommended, or paid more for a brand that provides a personalized service or experience.

Personalization done right requires the balance of the 3C’s: Customer, Content, and Context. This means collecting, storing, adapting, and using data from many different sources; and there is a lot of it, and it comes in many different forms, both structured and unstructured.

Here are some examples:

  • Customer – behavior, purchase history, demographics, social, segment, etc.
  • Content – Format (video, blog, text, whitepaper), category / tags, length, style, language / tone
  • Context – location, time, reason for interaction, life stage / event, sentiment, intent, etc.
  • Where most companies are falling short is that they are only using one of these types of data in their personalization decisions. For example, most personalization focuses only on the customer or content. Targeting decisions often look like:
    • Customers who look like X
    • Customers who have / have not purchased a product
    • Customers who clicked on this will like that

Where is the targeting based on the characteristics of the content I like or more importantly, the context in which my current interaction is occurring? Why am I acting? Eliminating contextual information crucially misses an individual’s reasoning for their actions.

Furthermore, most personalization is based on a decades old model of A/B testing, which essentially means that at any given time, you are presenting content that up to 49% of your visitors don’t want. Why can’t I get content based on my unique profile, the kinds of content that you know that I usually engage with personally, and that is relevant to why I am there right now?

Why can’t marketers solve this problem? We all know what needs to be done. We have these all in one marketing cloud solutions available. Everyone has a big data initiative (or initiatives…) and marketers are increasingly technologically savvy.

The answer lies in your data. In most cases, it is not a lack of it as, most companies have some sort of warehouse, datamart, or business intelligence tool that contain more data than ever before. However, most commercial marketing and data applications are built on a paradigm that makes it extremely difficult, time consuming, and expensive to work with and effectively use your data. For example, and this is as technical as this paper will get in showing these constraints:

  • You have to first understand the data model making it very difficult for non-IT professionals to know how to use it. How many of you have seen extremely complicated data model diagrams and wondered how you are supposed to do your job???
  • Modeling, preparing, and integrating crucially needed unstructured data is very time consuming and costly as it is the hardest type of data to model and format in traditional analytical applications. This may result in some data being dropped that does not conform to the data model.
  • To add valuable new data sources (like data that has information about the content or context), requires a physical change to the database, and in most cases, these changes have to occur in multiple places. This means a call to IT, more time than you have, and even more complexity to now manage. Marketers have been circumventing this challenging by activating separate channel SaaS applications and managing them on their own. This approach solves their immediate need, but compounds the problem by creating new silos, which increase maintenance time and cost.
  • The only relationships and insights that can be derived are the ones that are explicitly defined when the data model is set up. What this means is that if you don’t know, at the time of set up, all the hidden relationships that are revealed when data from multiple sources are aggregated together, then they are not discoverable without technical changes to the data. Isn’t that the point of bringing all your data together in the first place? To discover unknown insights about your customers, the content that they like, and the context for which they are there?
  • These are not characteristics that align with the way marketers are measured. The approach is called relational modeling and since the 1970s, it has been the industry standard for storing and managing data.

So what is the solution? How can marketers have the flexibility to quickly add new data sources as they need them, uncover new insights that they didn’t know existed, and incorporate customer, content, and most importantly, context into their personalization strategy?

The answer is a new paradigm called semantics. This paper won’t get into the technical details of how it works, but we will describe its benefits. According to Allen Taylor, author of Semantics for Dummies:

  • Semantics is a powerful and flexible way of integrating and modeling data so that users can have more context to their data than ever before
  • It is not constrained by the rigidity of traditional relational models and it is more flexible and very powerful for seeing relationships in the data and discovering new insights in the data. This reduces the time and cost of introducing new data
  • Enabled to be more easily visualized, which often leads to users uncovering new patterns they didn’t see before
  • Applications are made smarter and users can simply find and explore what they’re looking for faster and easier

How does it work? The non-technical answer is that data (structured and unstructured) is loaded as it is and then augmented with descriptive attributes, properties, and relationships about their data (called ontologies for the technical audience) to provide context to the data.

The major difference is that these descriptive attributes data can be added, generated, or configured without having to transform and define new data structures. This means that marketers can continuously enrich their profiles by adding more and more data about the customer, content, and context quickly and easily.

Semantics is used by Homeland Security to fight terrorism. If it is good enough to keep us safe by detecting obscure patterns across agencies, people, geographies, and billions of disparate data points, then why can’t we use it to determine the right content to present to me during each interaction?

Once the data is available, it is easy to understand and marketers can ask any questions that they want, about their combined data set. This is the same approach used to power web searches. How easy is it to search and find information online, navigating through an endless web of content linked together? Unfortunately, this approach has not been available in marketing technologies. You can still get your results in these other tools, but it will take a lot more work, time, and development, which is time you don’t have.

Let’s now look at an example of how this relates back to our three C’s, using a retail example:

John is a high value customer who buys the premium brand of formal men’s clothing 2X a year. Here is what we know about him.


  • > $500 spend per year
  • Top product categories: Formal shirts, dress pants, belts, dress shoes
  • Address: Massachusetts
  • Common device: Desktop
  • Common time: Evening
  • Frequency: 2X per year – Spring and Fall
  • Family: yes


  • Preferred content type: Items with multiple pictures and zoom
  • Most responsive language / Theme: Formal, sophistication
  • Navigation: direct search
  • Content people like John engage with / discuss on social media:
    • Golf and Tennis
    • Sports cars
    • Fancy restaurants
    • Social status

However, during this particular visit, he is on a mobile device in a different location.


  • Digital intent – vacation, travel
  • Social – Travel tips, restaurant recommendations in the Caribbean
  • Device: Mobile
  • Location: Caribbean
  • Time: Day, February

So what should we serve him?

  • 15% Offer to buy summer shirts and shorts

Because we can collect, store, and relate all of the customer, content, and context information together, we know that John is likely on vacation and probably not thinking about business clothes. We don’t want to present content related to the newest shoe or pants items, it is out of context. Also, since he is on a mobile device, he probably doesn’t have a lot of time and we need to get his interest and capture his attention very quickly. So we present him content for the new summer line of men’s clothes, with a creative theme of a family at an outdoor restaurant on a golf course, and a special offer to buy now (while he is in the moment).

This wouldn’t be feasible unless we could easily store, link, and derive insights on all of the customer, context, and content information we have about John and then act on it in real time. Without this contextual information, the retailer would have most likely presented him an offer based on his previous purchase patterns and it would have fallen short.

So how do you start using semantics? The best news is that you don’t have to discard your current systems, even your databases. In fact, semantics is a way to enhance the value of your existing investments. Tahzoo’s semantic based solution, powered by MarkLogic, can plug into your existing marketing infrastructure and allow you to start combining your customer, content, and context data to drive better personalization.

Our services

One experience. One team. The perfect solution.