Delivering the Promises of AI Driven Personalization

Delivering the Promises of AI Driven Personalization

By Josh Bowers and Denise Parris Co-Founders

Across the internet the two dominate models for AI driven personalization are Product (like Amazon) and Content (like NetFlix). But neither model for recommendation makes meaningful connections between products and content, and both struggle to infer interest cross categories. Despite the advances in AI solutions, why do they fail to create a holistic user and customer experience?

Largely, because they are built by people who are trained at crunching numbers, and as a result drive their recommendations on variables which can easily be counted. While numbers and prediction are powerful, they ignore the context. To provide a holistic and personalized customer experience requires injecting human expertise, the ability to infer, and the understanding of how things, objects, content, and sentiments are connected or unconnected. The Pavilion Intelligence team are technologists and experts in qualitative research recognized for advancing methodologies for the systematic review of content and thematic identification and mapping.

Pavilion Intelligence specializes in defining, creating, and designing taxonomies, metadata, and automatic classification that identify the themes, topics, and ideas derived from content and products to deliver on the promises of AI driven personalization. Enabling organizations to re-act to each individual user requires:

  1.  Informing Pavilion’s AI or your AI with a Taxonomy of interests, themes, and concepts which populate the content and product catalog,
  2. mapping the associations between these concepts such that Pavilion’s AI or your AI can infer content or products of interest to the user, and
  3. having an engagement engine which can track what the user has been presented, what they interacted with, and adjust its curation for the customer in real-time.

Pavilion Intelligence provides organizations the ability to address all three aspects with a combination of service and technology offerings. In fusing our expertise in qualitative research with the advancements of technology and data science, we have developed a four-phase methodology to drive meaning into AI solutions by defining and measuring the topics, themes, and ideas with which a customer interacts as they browse content and products, enabling more natural and accurate recommendations. We offer each phase as either a service or product offering. We describe the Pavilion Intelligence’s four phase methodology below.

Phase 1: Definition of a Taxonomy   

Pavilion’s method begins with defining a taxonomic structure intended to inform AI driven personalization. This starts with a Theme Thinking Workshop where in partnership with the customer’s team a Pavilion research lead makes the process of uncovering themes and subthemes upon which the creation of the taxonomy will be built tangible and often tactile.

Following the onsite workshop, the research lead in partnership with the Pavilion research team validates, refines, and defines the themes through conducting a systematic content review. Applying the methodology developed by Pavilion Co-Founder Dr. Parris, whose research on systematic literature review has been widely cited in academia. 

The concept definitions stored in the taxonomy and the ability to programmatically use them enables Pavilion to structure unstructured data like product descriptions, blog posts, influencer content, videos, and images into a content library.

Phase 2: Configuration of the Content Library

Leveraging the taxonomy enables the configuration of your content library. The first step defining the assignment criteria, and the second is implementation of the content pipelines. The assignment criteria is the standard which will be applied to determine when a content piece will be assigned to a concept from the taxonomy. This is specific to both the content type (video, text, image) and the content source. Content pipelines are the methods through which new and existing content will be automatically collected. Depending on the content source these methods can range from integration through an API, a direct RSS feed, or a crawler bot.

The taxonomy and automation implemented in phase 2 forms a semantic layer programmatically indexing all the available content (including products) for user engagement. The content library (a centralized structured representation of the content) automatically retrieves and assigns new content as it is created. Creating an evergreen—always relevant and updated—content library allowing the organization to address the enormous volumes of content required for AI driven personalization.  

Phase 3: Defining Connections with a Knowledge Graph

While the taxonomy and automation, which create the content library, enable content to be acquired and structured. It is in phase three where Pavilion enables recommendations of products AND content or cross category recommendations. Here the research team maps the intellectual connections between topics, themes, and ideas (labels) into a knowledge graph. A knowledge graph is a way to represent basic relations between labels. Thus, in this phase Pavilion applies the taxonomy which names, defines, and classifies labels and apply human intelligence to make meaningful connections between distinct labels by applying a mixed method qualitative and quantitative approach informed from Theme Thinking Workshop, the discovery from the taxonomy definition, and data insight the research team’s investigation of the customer data.

Phase 1, 2, and 3 combined create semantic layers which identify and label the topics, themes, and ideas across your product portfolio and digital content library.

Pavilion's method for creating semantic layer. Applying human intelligence to create taxonomy, using the taxonomy to structure unstructured data. Applying human intelligence to map associations within the taxonomy in a knowledge graph.

Phase 4:  User Engagement

While the taxonomy provides the concept definitions, the content library the material, and the knowledge graph the connections within the content it is Pavilion’s engagement engine which selects content for each individual user using the knowledge graph’s associations.

  • Selection is refined on a 1-1 basis per the topics, themes, and ideas of interest to the user by rank ordering the available material for each user in real time.
  • Content impressions are tracked on a 1-1 level ensuring with each interaction the user is presented with the most individually engaging fresh content avoiding content repetition. 
  • Content is systematically selected to assess a user’s interests and preferences, requiring Pavilion to select content and products representing distinct concepts and track both what the user as seen and what they engage with informing the users relative preferences (the user’s preference for one concept over another).
Pavilion's model for generating 1-1 engagement. Converting unstructured data into structured data through taxonomy definitions. Mapping into a knowledge graph, and then informing 1-1 engagement with machine learning.

Pavilion Results:

Creating a holistic understanding of the user delivers results well beyond improved “user experience”. While Pavilion is just starting out already it has generated increases in customer purchases by 23% in six months in retail settings and delivered more qualified sales leads in a B2B setting in two months than all other digital marketing activities and the sales team combined. Pavilion has both the technology and methodology to enable organizational success.

Pavilion helps technology and marketing leaders go beyond the transaction to create a virtual two-way dialog with their customers to increase their lifetime value.