Advantage of the New Adobe Target: Automation

Let’s continue our journey with Rick Patterson, our optimization manager at a fictitious expectant parents’ website, in his optimization maturity saga using the new Adobe Target. Each stage in this maturity process maps to a key area of development that we have focused on during our major monthly releases over the past few years to redevelop Adobe Target for scalable and sustainable optimization growth and success.

In part one, Rick easily implemented Adobe Target across all of his channels and executed high-value testing with our standard guided workflow, leveraging built-in best practices for optimizing his strategy. In part two, he leveraged the unified progressive profile of the Adobe Marketing Cloud, easily augmenting his data with his customer relationship management (CRM) and other profile data sources to build sophisticated audiences and execute the most effective optimization and personalization approach moving forward. In part three, he was able to mature beyond a single element on a single page to run effective personalization across all of his channels. This includes acquisition optimization in display advertising, email, social media and partner sites connected to multiple pages on his website or mobile applications. He can even optimize content across the cutting-edge “ Internet of Things,” Internet-enabled devices and screens found at in-store kiosks, ATMs, smart appliances, cars and other places. Adobe Target is the only solution that allows Rick to extend in to these offsite areas for connected personalized engagements.

Rick now has an effective process for having over a hundred testing and targeting activities at relative stages of the development process — from ideation to build and execution to reporting and recommendations. But Rick is only one person with a few part-time resources he can leverage at different stages of his process. He realizes the potential value of having analysis and insights easily at his fingertips. In fact, having that analysis readily available could make him even more productive by showing him where, what, and to whom he should be testing and targeting specific content and experiences to in order to achieve his key business objectives. This is where the optimized recommendations and automated personalization features built into Adobe Target can help accelerate Rick’s results and maturity through machine-learning analysis.

The automated algorithms included in Adobe Target are built on industry and statistical standards that have been proven over time and tested for years by our data scientists at Adobe Technology Labs. They are the equivalent of running hundreds of tests at a given location and can highlight where the biggest bang for your efforts resides. They also self-optimize continually and can adjust how content, design or experiences are displayed to an individual based on up-to-the-minute data analyses.

How? Our random forest algorithm gathers all of the profile data of each individual who arrives at the automated personalization algorithm location. The more data it receives, the more accurate analyses it can provide. It then evaluates each variable of the profile based on what it knows is most predictive from other visitors who have passed through this location. It targets based on the most predictive variables of the profile. For instance, specific geo-location data, such as the state the person is located in, and behavioral variables, such as “buys baby food,” will define what content Adobe Target will display for the highest propensity for converting based upon whatever success metrics are defined for the algorithm.

By using automated personalization, Rick ensures that all profile variables are being evaluated without having to manually run and evaluate multiple tests. He assures that each individual sees the best and most relevant version of his content. He can also gauge relative performance of the algorithm over a random or set control, the individual performance of each content variation, and he can generate a report that shows the relative predictability of all variables being evaluated — a report that actually shows him what to focus on for the greatest return! Automated personalization is by far the most successful feature we provide in Adobe Target. Most clients see a 20-100-percent lift in whatever the algorithm is optimizing to.

In addition to random forest and a residual variance model that optimizes effectively over longer periods of time, Rick also has access to a customer-lifetime value algorithm. This looks at the full visitor profile of each individual and evaluates it based on past profiles and purchasing patterns to deliver the best variation of content with a high-propensity for generating multiple or repeat conversions. It’s a discounted likelihood metric, due to the fact that the propensity for a second and third conversion is less likely; however, there is a greater return from those conversions, and most often exponential value in terms of customer loyalty, so optimizing them (especially in terms of frequent convertors) can be immensely valuable.

These algorithms are effective when Rick is just starting out and wants to uncover the best elements and audiences to refine his experiences through optimization. They’re also effective further on in the maturity process when he wants to introduce new experiences and ensure that multiple divisions can all effectively target their content to the right audiences — with far greater scientific confidence. He can test multiple content variations, experiences, or even layouts of his pages to see if boosting particular content to different audiences is more effective.

If Rick feels confident that one variation of a manual test might succeed within an A/B/n or multivariate test, he can select an auto-allocation feature during his test set-up rather than putting in a percentage split of his traffic. This diverts traffic to a variation that is leading the other variations for the possibility of completing a test earlier than expected — without sacrificing rigor in results like other predictive methods in the market. It will confidently show the winning variation’s performance to Rick, so he can take action with that, but it should be noted that selecting this method will not necessarily show the other variations’ performances in the test, as he would see if he let the test fully run to statistical confidence without auto-allocating traffic.

Another valuable guided approach to automated personalization is our optimized recommendations algorithm. Just like automated personalization and manual testing and targeting, Adobe Target‘s recommendations follow the same three-step workflow. Rick can visually select a location to add or replace recommendations. He’s then given suggested best practices on which algorithm to use based on the page and his industry (insights our team has learned over a decade of working with leaders in the space). He can use any algorithm he chooses or even test algorithms against one another. These algorithms are standard logic that he can then customize based on any metadata, profile data, or analytics he wishes to leverage. What should the algorithm base its logic on: the current item, category, or last action? Should it be looking at popularity or some level of correlation? Rick can easily select these options from a guided workflow and write any inclusion or exclusion rule based on metadata from the feed. He can select the design for the recommendations, and visually quality assure (QA) and edit it to ensure it’s displaying exactly what he wants. Recommendations can be immensely powerful in any industry to provide the customer with the next articles, collateral, products, or other content that they most want to see and engage with, for improving time onsite, conversion, average order and lifetime value.

Stay tuned for my final installment of this saga where we’ll see how Rick can move the needle in the critical area of mobile and mobile app optimization.