Featured in Collection 04 Adobe Summit
New Adobe Target Enhancements Unlock the Power of AI in Personalization

New Adobe Target Enhancements Unlock the Power of AI in Personalization

Last week, my colleague Drew Burns shared stats on the impressive growth we are seeing in experience optimization programs in the past year, and he also mentioned that throughout the week of Adobe Summit in Las Vegas, we’d be announcing several new features and enhancements we’re adding to the Adobe Target solution. Well Summit has arrived, and the start of the big reveal is now. In this first of three posts, I want to tell you about new features in Target that will allow you to unlock enhanced transparency and customization of Adobe Sensei-powered AI for personalization in Target. These include:

  • New Personalization Insights reports
  • Propensity score model comparisons
  • Real-time customization to customer-owned models in Recommendations

Address the black box issue of AI with the new Personalization Insights report

Today, when marketers leverage Adobe Sensei-powered AI capabilities in Adobe Target as their “machine learning sidekick,” they see powerful results. The algorithm’s analysis of all profile attributes of each individual visitor over time produces exponential conversion lift, along with a deep analysis – equivalent to the results of hundreds of concurrent tests – in real-time and self-optimizing over time.

Our new Personalization Insights reports highlight more of the deep analysis that the algorithm is doing to deliver personalized offers and experiences to each individual. It shows what visitor attributes were most influential in the model Adobe Target built, how it grouped customers together into the audience segments it used, and how it decided what offers would resonate most with those audience segments. This does a lot of the heavy-lifting analysis for the practitioner, delivering to them results on these insights that they can share with stakeholders and use for more effective personalization and analysis in future activities.

Adobe Target Product Manager Shannon Hamilton explains, “There’s a discussion across the industry that algorithms used by AI are very accurate and valuable, but they are also highly complex and not human readable.” She continues, “To solve for that, we’ve created a patented algorithm that sits on top of our AI-based personalization used in Auto-Target and Automated Personalization. That algorithm, which is cutting edge data science, generates human understandable insights from the output of the AI-driven activities.”

Personalization Insights is the new report type in Adobe Target that provides transparency  into the algorithm’s decisioning and provides these human readable insights. It reveals the most influential visitor attributes, audience segments the algorithm formed, and which offers or experiences the algorithm delivered to those audience segments. For example, the “Model Attributes Ranking” section of the report could reveal the algorithm determined that loyalty program status was the most important attribute in forming audience segments. The marketer might then look at how the model grouped together visitors and see an audience that consists of platinum- and gold-level status customers who clearly preferred one offer. That could lead to inferring that customers with platinum and gold status respond similarly, so the marketer could create an experience specifically for them. By looking at the type of offer or experience that resonated best with customers with platinum and gold status, the marketer might also create similar offers and experiences for those customers elsewhere.

Personalization Insights will launch in beta this spring, and as Shannon notes, “It shows our commitment to making data science approachable for non-data scientists.”

Leverage your own data science with propensity score model comparisons

Many Adobe Target customers know that machine learning, AI and data science are critical to their future success. They’ve invested in building data science teams who develop data models that incorporate domain- and company-specific knowledge. But, as Shannon says, “They needed a way to take this valuable output of their research environment and apply it practically to the business by operationalizing it in Target.”

For example, let’s say a furniture retailer has several product lines, two of which are tables and beds. They’ve developed a model that determines and scores the likelihood, or propensity, of a visitor purchasing a table or a bed in the next year. A visitor may have a score of 2 for purchasing a bed and a score of 4 for a table. As the visitor visits a page personalized with Target, the retailer would like to compare those propensity scores and deliver an experience related to the one with the higher propensity to purchase—in this case, the table.

Propensity score model comparisons is a new feature in Adobe Target that will allow marketers, product owners, developers, and data scientists to bring their own data models with custom propensity scores into Adobe Target and factor them into the solution’s AI personalization algorithms, rules-based personalization, or anywhere customers use audiences in the solution today—for example, as entry criteria for any Target activity type. As a result, Target will be able to compare multiple propensity scores for a visitor on the fly, place that visitor into an audience based on the highest propensity score, and deliver the most relevant experience.

Shannon says, “This new feature reflects our overall strategy of continuously providing new opportunities for our customers to use their own in-house data science in Adobe Target.” This new feature will be released in June.

Real-time customization to customer-owned models in Recommendations

This final feature, which marketers can access through the Custom criteria option in Recommendations activities of Adobe Target Premium, represents another way we’re giving brands flexibility to leverage their data science investment in Target.

With Custom criteria, brands can pass the results of their own data models to Adobe Target and apply them to Recommendations activities in Target. They output a file that reflects their data science, like propensity score models, and upload it for use in the recommendations algorithms of Target. Target then uses its Edge network for lightning fast delivery of the recommendations. With this new feature, we’re essentially letting them bring their own algorithm into Target by way of our Custom criteria option and then leverage the well-oiled machinery of the solution’s content delivery infrastructure.

For example, an online movie streaming company could regularly run algorithms that output a file with the specific movies to recommend when a visitor views a particular movie. The file basically provides a curated list of recommendations. The company uploads the file into Target and then uses it in the Custom criteria option of Recommendations.

In addition, we’re now letting brands execute real-time filtering rules, weightings, and customizations on top of custom criteria in Adobe Target Premium for Recommendations activities. The value of this enhancement? You can create an awesome algorithm, but at runtime you may find that something changed—maybe the customer’s preference or an item is out of stock. At runtime, we can now account for those types of changes in the recommendations Target delivers.

Using Custom criteria, the movie streaming brand can apply the results of a propensity model that scores the customer based on his or her propensity score for romance, comedy, horror, action, documentary, or drama categories of movies. If that customer’s affinity changes from action to comedy, at run time, the company can now deliver recommendations for comedies.

Similarly, a sports apparel retailer could have a proprietary algorithm that recommends specific items related to a particular shoe model—perhaps complimentary shoe laces, shirts, shorts, hat, and so on. The retailer can use the results of that algorithm as Custom criteria, and then set up a filter that checks inventory at runtime. If the recommended shoelaces are out of stock at runtime, they’re excluded from the recommendations.

This enhancement to using customer-owned models via the custom criteria algorithm gives marketers the power to combine their off-line calculations with real-time filtering. We’re pleased to let them know that this enhancement will be generally available as part of the March product release.

Making it easier to leverage algorithms and investments in data science

Like many companies, we see the important role that AI and data science will play in the future of digital marketing. We also see it as our job to make it easier to leverage the data science investments both we and our customers make by adding features that let them unlock the power of AI in their personalization efforts.

Stay tuned to the Personalization blog—tomorrow, we’ll be revealing exciting new features that will enable Target users to extend the functionality and accessibility of Adobe Target.

Recommended Articles