New Adobe Target Features to Drive Adoption of and Gain More Value from AI
Image source: Adobe Stock / by sdecoret.
by Jason Hickey
posted on 02-20-2019
AI is critical for helping brands optimize rich customer experiences and personalize at scale, analyzing data in real-time and over time, and delivering faster results and deeper insights. Drew Burns gave you a small glimpse of some exciting new features and enhancements we’re adding to Adobe Target, the world’s leading Experience Optimization solution. In this post, I will go deeper on two announcements that enhance our personalization capabilities powered by Adobe Sensei. The first makes it easier than ever for you to see the performance or revenue impact of your AI-driven personalization efforts and to gain buy-in for those efforts. The second gives you a dynamic marketer-controlled algorithm to deliver your visitors one-to-one product, media, article, or content recommendations.
Show the ROI of your Adobe Sensei-driven personalization the way you need
Buy-in for your optimization and personalization program hinges on your ability to show and clearly communicate ROI. It’s also how you get stakeholder buy-in for trying out new products and features.
We’ve previously given you a peek behind the curtain of AI personalization in Adobe Target. In that post, we discussed how Adobe Target uses a rich policy — an ensemble algorithm consisting of Random Forest, patented Genetic Algorithms, Thompson Sampling, and Epsilon Greedy approaches — to build, score, and interpret machine learning models. This policy balances exploitation against exploration to deliver lift and positive ROI while continuing to discover and adjust to changes in user behavior to build better models. Exploration in this context is of a random control group. Yet random, while useful for machines to learn in a closed system, is not conducive to fully measuring the net effects of AI against something else entirely.
That’s why we developed a new enhancement that helps you more clearly track and show the ROI you drive when you personalize using our Adobe Sensei AI features, Auto-Target and Automated Personalization. We’re doing that by letting you select the control experience against which to compare your AI results. That control can be any experience — a static one with no personalization, a dynamic one with rules-based targeting, or even one that leverages non-Adobe intelligence, such as that developed by your own data science team.
Choose the experience against which you wish to compare your AI-driven personalization.
Recommend products and content by matching them with weighted visitor attributes
Recommendations has a broader array of personalization use cases beyond the traditional product suggestions, so we’re always exploring new ways that can help you leverage it for even greater benefit. Adobe Target now offers Weighted Relevance Recommendations, a feature that gives you full control over the content and product recommendations your visitors see based on the visitor attributes you believe are most relevant, weighted by an importance score.
Easily specify and score visitor attributes for Weighted Relevance Recommendations
Here’s an example to show how it works:
A job listing website knows that the most important attributes in order of importance that lead to a visitor finding the ideal job are area of expertise, location, and years of experience. When a job seeker looks for possible jobs and hits the home screen of the mobile job app, Recommendations looks at attributes in their progressive visitor profile based on values they enter in a form, their on-site search terms, their onsite behavior, geolocational data, CRM data, and so on. Recommendations would then display job listings that best meet the visitor’s expressed or implied interest.
In this case, the marketer might create these rules that weight the importance of those attributes relative to each other on a scale of 1-10:
Rule 1. Job seeker area of expertise exactly matches job area of expertise gets a score of 10.
Rule 2. Job seeker’s current or desired city falls within job location region gets a score of 7.
Rule 3. Years of experience of job seeker falls within or above range of years of experience required for job gets a score of 3.
One thing that you can note immediately above is that there are no static values listed in either the requirements of the job or the attributes of a visitor’s profile.
When a graphic designer from Portland, Oregon with seven years of experience in graphic design searches for a job, here’s how three jobs would get scored and the order in which they’d get displayed to the job seeker by Recommendations:
- Senior Graphic Designer for Seattle requiring 5 to 10 years of experience. Score: 10 + 7 + 3 = 20
- Art Director in Seattle requiring 10 years of experience. Score: 10 + 7 + 0 = 17
- Entry level graphic designer job in Atlanta, requiring 0-1 years of experience. Score: 10 + 0 + 3 = 13
You can see how this capability enables truly dynamic recommendations and merchandising by matching the attributes of a visitor’s unique visitor profile to available products or content that most closely match. Weighted Relevance doesn’t require you to use any underlying recommendations algorithms like “Top Sellers” or “People Who Viewed This Bought That” — it’s completely customized by your rules and scores. You can also use this feature with all the inclusion rules, criteria sequences, and promotion capabilities that you’re currently leveraging with your other Recommendations activities.
Redesigned Recommendations workflow unlocks new use cases
You can now include Recommendations inside A/B Testing and Experience Targeting activities. This workflow redesign lets you integrate this valuable capability into your core experience optimization strategy and opens entirely new ways to use Recommendations with each of these key Adobe Target capabilities.
Redesigned Recommendations workflow shows using Recommendations in an Experience Targeting activity.
Some of the new use cases that this workflow redesign unlocks are:
- Test and target recommendations and non-recommendations content within the same activity and even in the same experience
- Easily experiment with placement of recommendations on the page, including the order of multiple recommendations trays or modules
- Automatically push traffic to the best-performing recommendations experience using the multi-armed bandit capabilities of Auto-Allocate
- Dynamically assign visitors to tailored recommendations experiences based on their profile using the AI-powered personalization in Auto-Target
Look out for a post on this feature that showcases additional and amazing new use cases for this workflow.
Don’t forget these recently released reports
The winter holiday season can be a bit hectic, so we wanted to make sure that you took a moment to revisit two reports in Adobe Target that might have slipped under the radar when we released them in November. These Personalization Insights reports are designed to help you better understand how the Adobe Sensei AI-driven personalization used by Adobe Target ticks.
The Important Attributes report surfaces the attributes and traits from all your data sources that most impacted the decisions the machine learning model made. Those sources can be first-, second-, or third-party sources; Adobe Target, Adobe Audience Manager, or Adobe Analytics; and historical or real-time.
Important Attributes report in Adobe Target.
The Automated Segments report shows new segments the model viewed as valuable and created by combining various attributes and traits from any of those sources.
Automated Segments report in Adobe Target.
Learn more about these Personalization Insights reports in this blog post, which explains how MAGIX, the technology that surfaces these insights, works.
Want to know what else we’re announcing?
Check out these additional announcements of the new features and enhancements in Adobe Target:
Topics: Digital Transformation, Personalization