Wildcard Content In AI-Driven Personalization

Wildcard Content In AI-Driven Personalization

What is wildcard content?

By Josh Bowers and Denise Parris

Wildcard content is content created and distributed to be different.

Wildcard content has long been a tool of content marketing managers to course correct when a content strategy needs a reboot or as an opportunity to take risks trying different tones, concepts, or mediums.

When using AI powered personalization what is considered wildcard content and how wildcard content is used must evolve. To Pavilion’s AI wildcard content is content selected to test a user’s interests and preferences, or because the content is different than the rest of the content presented to the user. What is considered “wildcard” is dependent on the user, and wildcard content will be used more frequently than what content marketers are accustomed too.

How is Wildcard Content Used in AI Driven Personalization?

Pavilion uses wildcard content to:

Avoid positive feedback loops.

Positive feedback loops are when a user engages with a specific type of content and because of that the AI engine only shows the user similar content. The result is high churn rates as users lose interest or assumes the organization has nothing else to offer.

Moving beyond the user’s top ranked content provides opportunity for the user to express their full spectrum of interests and avoids positive feedback loops. Enabling the organization to understand the user holistically and keeping the content mix engaging.

Break up the content mix.

We first realized the value of breaking up a content mix with wild card content in 2014. Our start in AI powered personalization applied our NLP capabilities to identify outdoor sporting events posted online. We had a feature on our original product which allowed users to “Create an Alert” and be notified through email of any event happing within a geographic area for the outdoor pursuit of their choice. When determining content to display, we found intentionally breaking up the presented content with non-user related interest items (i.e., wild card content) resulted in decreased list churn and increased engagement. Since then we have improved our methodology for delivering personalized wild card content providing increasing engagement in multiple industries.

How to select wildcard content for AI Driven Personalization:

For Pavilion, what it means for content to be wildcard is dependent on the user. Content which would fill a wildcard slot for one user may be directly tied to another user’s interests. AI driven personalization engines rank order content for each user with a probability distribution. Probability distributions are expressed as a score—the closer the score is to 1 the higher the confidence in the recommendation.  

Take the following example: if our content library had 9 content pieces available, and we were ranking them for a user named Bob, content pieces 4, 5, and 6 would be used as wildcard, but could be engagement content if rank ordered for a different user.

Pavilion groups content into the engagement zone, wildcard zone, and the digging deep zone. Zones are based on the relative scoring of content for the user. As Pavilion scores content according to the taxonomy defined in Pavilion’s semantic layer—the score reflects the user’s interest in the topic, concepts, ideas, and themes presented in the content.

Pavilion selects wildcard content using any combination of: content featuring topics the user has had the least exposure, content closely related to that which is already selected, content which is NOT closely related to that which is already selected—all accounting for the content score and zone. Content repetition is an engagement killer so Pavilion ensures all content selected is new to the user.

How much wild card content should be used?

There is no good answer to this question, but we have found the channel to be the single biggest factor. For email engagements, where Pavilion tends to have narrowly defined engagement and wildcard zones, we see the best results with 16% of the presented content being wildcard. Display on websites and apps, we see less sensitivity, but the percentage really depends on how the zones are defined.

Everything is Built on the Taxonomy:

While wildcard content plays an important role in AI driven personalization it is reliant upon the quality of the taxonomy upon which you are building linked data. If you are starting down this path, we would strongly encourage you to read our post on building a taxonomy for AI driven marketing and reach out if you would like to discuss.