Getting Started with Adobe Sensei within Adobe Target

When starting out with artificial intelligence and machine learning-driven optimization and personalization at scale, you might think you need to build a program that looks like something straight out of a science fiction thriller. It might seem like a daunting or even impossible task. But it’s not that difficult in reality, and Adobe Sensei within Adobe Target makes it more accessible than ever to get started.

Adobe Sensei is the technology that powers intelligent features across all Adobe products. Within Adobe Target, Adobe Sensei can help you find insights faster and deliver the right experiences to the right customers at the right time.

Let's say you have an upcoming product launch, pre-order event, or enrollment period where you have a short timeframe to maximize a key metric. You might run an A/B test against several variations to see which one drove the most clicks, orders, or sign-ups— which is great because you're testing! You're feeling good because you're being a diligent optimizer; once the test window closes, you check the results and you find out that one of your experiences out-performed the others. You have a winner! Then you think, wait... I had one winning experience, but most of the traffic in the test went to the lower-performing experiences. Wouldn't it have been nice to have captured more clicks, orders, or sign-ups while the test was running? This is where Auto-Allocate could have optimized your traffic during the test and captured the most possible gains with statistical significance.

How does it work?

Auto-Allocate will conduct rounds of scoring to measure the experiences against each other and then send more traffic to the better performing experiences for you! No fancy setup required—just choose Auto-Allocate as your allocation method when setting up your A/B test. For more detailed info on how the Auto-Allocate algorithm works, click here.

To turn any A/B test into an Auto-Allocate activity, simply select Auto-Allocate in the "Targeting" step during the activity setup.

To get the best experience for every visitor on every visit, we're going to need some help from more advanced machine learning capabilities. This is where Auto-Target shines.

Auto-Target allows you to still build your activity as an A/B test, where you might have a single page or multi-page experience, but instead of trying to find the best overall winner, we call on Adobe Sensei to find a winner for every visitor on every visit.

What does that mean?

It means you can build experiences designed for different personas or users at different stages of the customer journey. Instead of painstakingly building the audiences to manually target them with what you think is the best experience, you allow Adobe Sensei to evaluate a user’s profile and make a real-time decision on what would be the best experience for them.

Here's an example:

Let’s say you have several categories of products or services but you only have one homepage to display them. You might end up with a long page that attempts to show everything to everyone all at once, or worse, a carousel that rotates between all the different categories of products or services. You end up with a one-site-fits-all approach and your success is reliant on hope. We hope that the user will find the thing that interests them on this gigantic page full of content. We hope this is the right order of products or services within the carousel. This is not a winning strategy.

Instead, leverage the incredible power of machine learning to determine and deliver the best experience to every visitor. This allows you to create more focused, more intentional messaging and creative knowing that Auto-Target is going to match the right visitors to the right experiences.

How does it work?

When you activate an Auto-Target test, Adobe Sensei will build a model for every experience in your activity. These models represent the patterns and predictions of which user attributes (or the combination of attributes) are used to determine users' likelihood of converting on your goal metric. These models then inform Target of which experience to show every visitor on every visit.

Typically, this is where people will ask: what data is being used to drive these models and can I see it? Out of the box, Adobe Sensei will leverage the same data that is used to create audiences in Target as well as shared Experience Cloud Audiences; where things get really interesting is when you inform the models of which data to leverage. Time to value really speeds up when customers feed Target with their own brand data.

Further, Target will provide Insights Reports that will inform you which attributes are important to the models for the activity, as well as the performance of each experience in the activity by segments that Adobe Sensei is finding and targeting. These reports not only help you decide how you may want to tune the models, but also discover which experiences are “speaking to” specific segments and how you might leverage those insights in other targeting opportunities.

To turn any A/B test into an Auto-Target activity, simply select Auto-Target in the "Targeting" step during the activity setup.

More details on Auto-Target and how it works can be found here.

To deliver the right offer or combination of offers for every visitor on every visit, we're going to once again turn to the advanced machine learning capabilities of Adobe Sensei within an Automated Personalization activity.

Automated Personalization leverages the same powerful algorithms as Auto-Target; but, where Auto-Target is set up like an A/B Test, Automated Personalization is set up like a multivariate test.

What does that mean?

It means you can build variations of offers or even combinations of offers in multiple locations and leverage Adobe Sensei to evaluate a user’s profile, make a real-time decision, and then deliver the best experience for them.

Here's an example:

Let's say you have a highly visible spot on your homepage (like a hero banner) where you want to present the best possible offer to each visitor. The hero banner might have an image, a headline, and a call-to-action. You might also have 6 different images you can use, 3 different headlines that call out different value propositions, and 2 different calls-to-action, all of which might appeal to different segments of your visitors. With Automated Personalization, you can input these variations into your activity and, since it is a multivariate setup, Target will create the combinations for you! In our example we would have 36 possible experiences (6x3x2) that Adobe Sensei will choose from to match to each visitor.

Another example could be content tiles. You might have a series of 3 or 4 tiles on a page but have hundreds of potential pieces of content to show to your visitors. Instead of swapping out content manually for those 3 or 4 tiles every few weeks, waiting to measure their effectiveness, and then living with the one-site-fits-all approach, you can have Adobe Sensei do the heavy lifting and find the right combination of content tiles for you!

How does it work?

Just like with Auto-Target, when you activate an Automated Personalization activity, Adobe Sensei will build a model for every experience in your activity. These models represent the patterns and predictions of which user attributes (or the combination of attributes) are used to determine users' likelihood of converting on your goal metric. Target then uses those models, sees the visitor's profile, and matches the user to the best possible experience.

Just like with Auto-Target, Automated Personalization activities will provide Insights Reports that will inform you which attributes are important to the models for the activity, as well as the performance of each experience in the activity by segments that Adobe Sensei is finding and targeting. These reports not only help you decide how you may want to tune the models, but also in discovery of which experiences are “speaking to” specific segments and how you might leverage those insights in other targeting.

If they use the same algorithms, you might be asking, what makes Automated Personalization different from Auto-Target?

Aside from the multivariate-style setup mentioned above, another big differentiator is the ability to do offer-level targeting. You might think wait, isn't the machine-learning supposed to do all the targeting magic on its own? It certainly can, but there are several use cases where intervention might be necessary.

Let's say you have 6 different content tiles that you want to deliver via machine-learning, but 2 of those tiles are really meant for existing customers that you do not want to show to prospects. With Automated Personalization you can simply target those 2 offers to existing customers and Target will not show them to anyone other than those identified as existing customers. This allows you to keep your setup the same and not have multiple activities targeted to different audiences.

To get started, simply choose Automated Personalization when creating an activity and leverage the familiar 3-step activity setup in Target.

More details on Automated Personalization and how it works can be found here.

If you've been anywhere on the internet in the last decade, you've likely seen phrases such as 'you may also like' or 'people also bought' or 'recently viewed' followed by a series of products, services, or content. This is because it's a very effective way to increase conversions or engagement by displaying content that is relevant to the user. These kinds of experiences are powered by a recommendations engine. Adobe Target's Recommendations activity type leverages Adobe Sensei with marketer-friendly, customizable algorithms to deliver data-driven recommendations to users on a 1:1 basis in real-time.

What does that mean?

It means you can easily create an activity that could have an infinite number of experiences that would otherwise be impossible to manage manually.

Here's an example:

Let's say you have a large number of products and you want to recommend the right set of products to the customer based on the last product they viewed. Or you may want to only show them products from the same category as the category they are currently viewing. You might also want to match the recommended products to something in the visitor's profile like a specific preference or affinity for a brand or size or color, etc. These are all very common and easily achievable use cases with Target Recommendations.

How does it work?

Recommendations looks at items or products and how users view/purchase/engage with them, as well as the metadata of the items or products, to build relationships between them. These relationships are the building blocks for making data-driven recommendations.

What do I need to make it work?

Target Recommendations works off three main things:

1. Product Catalog (Entities)
2. Algorithms (Criteria)
3. Delivery template (Design)

The product catalog is comprised of what Target calls Entities. Entities are the items that you want to recommend and can be anything that is classifiable with an ID, such as product SKUs, document IDs, or video IDs; the key is that each item in the catalog has a unique ID. Along with the ID, several other pieces of information are passed and associated in the catalog like name, category, color, price, size, description, etc. This is the information that Target will use to populate the recommendation to the end user.

The algorithms in Target Recommendations are called Criteria. This is the fun part! Criteria are the rules that you get to create to make the recommendations happen. You might be thinking: 'no chance I can create an algorithm,' but it's a very marketer-friendly interface with lots of drop-down options to guide you.

Once you create your custom Criteria, you can save it to use in multiple activities. Target gets you started with several common use case algorithms that you can customize as much as you want from there.

The delivery template in Target Recommendations is called the Design. The Design is another re-usable element within Recommendations that defines how recommendations appear to the end user. For example, you might have a recommendation for related articles that lists 4 articles with a link to go to each one. The Design is an HTML template that would enable the display of those related articles and would be customized to match the look and feel of your site.

Can Recommendations be used for more than products?

YES! Beyond the more common retail product recommendations, you can leverage Target Recommendations to recommend articles, videos, downloads, travel experiences—essentially anything you can classify with an ID!

How do I create a Recommendations activity?

Once you have your Entities populated, your Criteria in mind, and your Design ready, you can use the familiar 3-step workflow to create a Recommendations activity. There are a ton of features within Recommendations like testing Criteria against each other, timed promotions of specific items, running specific Criteria in specific slots of your recommendations, and many more. To get started, select Recommendations as your activity type or you can insert a Recommendation within an A/B test or Experience Targeting activity as well.

More details on Recommendations and how it works can be found here.