Target Personalization: Getting Started with Recommendations & Category Affinity

Transcript

Hello, everyone. Just going to give it a few more minutes for people to join. Bill, do you think that we can get started? Sure. All right. Awesome. Hi, everyone. Thanks again for joining Bill and I. Today’s session will be focused on getting started with recommendations and category affinity. My name is Leena and I’ve been working with Target for about four years now, which is, I think, a fraction of the time that Bill has spent with the tool. I’m located in Seattle, Washington, and excited to be speaking with you all today to introduce himself. Hi, I’m Bill Rosenthal. I’ve been with Adobe for 13 years as a target specialist and using the product and product from its premium nature days and was one of the beta testers for Rex Classic, the version before the current version. So I go back to the early days. So you all are in good hands. All right. So in terms of today’s agenda, we’ll start with Rex with a Rex overview, talking a little bit about the setup and then we’ll take a look at the UI and then I’ll pass it over to Bill to talk about category affinity, and we’ll do a category affinity demo as well. So to get started with talking about Rex, I think it’s helpful to look at in the context of other target activities as well. Target that covers all the bases necessary when it comes to testing and personalization, making sure your visitor has a personalized and relevant experience within target recommendations helps us recommend specific content to a visitor at scale, allowing your visitor to discover relevant options from your catalog of items or content that they might not have known about otherwise. So in short, recommendations is personalization at scale, and it gives you the ability to intelligently recommend hundreds or thousands or millions of things. This isn’t just products. Rex can be great for videos or content articles or more. The way this works depends essentially on what algorithm you choose. So the algorithm span from behavior based to popularity based content, similarity, and so on. So the recommendation might be for all users that specific user or the item itself, it can also be completely customized. So the room for customization is the strength of Target. So you can apply sequencing, weighting, exclusion rules and more. And we’ll talk about that a bit later, but some strategies to keep in mind when using recommendations. So if you have anything at all that can be structured or fed into a catalog, you can use recommendations. All you need is a common structure or the ability to get a common structure with metadata in that metadata for each item to use. So use Rex if you want to recommend the best products or content across thousands or millions of items, it could be five. It could be 100. Rex can generate a set of results for each visitor based off of the context of their user profile. And lastly, you want to use Rex if you want merchandizer control of your personalization. So not only can you customize the underlying logic of the algorithm, you can also override and take a few slots in the carousel manually and you might be thinking, So when is Rex not the right fit? Rex might not be the best fit if you only have a small handful of offers, content or products to recommend. If your catalog items become irrelevant quickly or items are interacted with only a few times, or if you want to personalize based primarily on user characteristics other than on site in-app behavior. So that would mean if you want to primarily consider a users loyalty segment, web browser or geography, and something like auto target or automated personalization might be the best fit. So before we dive in, just some vocabulary to keep in mind here. So criteria in Rex are the savable and reusable algorithms. A collection is a set of items managed by search term filters. The design of Rex is the Adobe hosted template that delivers the recommendation in sequence. Quite criteria is a set of cascading criteria and users will qualify for the criteria in the order that they are set in the sequence. The entity ID is a unique parameter that is used as the key for an item or a page, and then your feed is a CSB feed of product or page metadata uploaded to Rex to provide some information about a page, including the entity ID. Now let’s talk about how to set up recommendations. So while Rex might seem a little daunting, the workflow is really three simple steps. So first, we think about the audience we’d like to target, then we think about the desired action or behavior you’d like them to take, which helps you decide which algorithm. And then finally you select how you’d like the rack to appear visually. And after finalizing the look and feel of your rack, you’re ready to start building. In the visual experience, Composer taking a bit of a closer look at Rex criteria. This is what the algorithms look like in the UI as you select them to set up recommendations, activities. You can see the rules and logic as well as the configuration. And jumping to the next slide, we have a really diverse set of recommendations, algorithms to play with. These algorithms determine what catalog items are shown to your site visitor and can be customized based on your marketing needs and priorities. Right now, you can select from popularity based, content based, item based, personalized, or build your own algorithm. And depending on which algorithm you choose, you can nudge your visitor towards taking certain actions on your site. For example, let’s look at item based algorithms. If your desired action was to encourage people who purchased a backpack to maybe consider buying hiking shoes or trekking poles, then for your recommendation, you’d want to create a rack that shows items that are often purchased together, such as the people who bought this also bought that criteria. Or let’s say your desired action was to increase the time visitors spent on your site. Then you’d want to create a recommendation that suggests other videos using with people who viewed this also viewed that criteria. So it really depends on your end goal. But starting there, you’ll easily be able to determine which algorithm to use. So at the foundation you have the algorithms that we just reviewed and then you can also layer on refinements specific to your business, like only showing items in X price range or never showing out-of-stock items. And it gives you total control over the inputs and outputs of what you’ll be recommending. And then as you layer on your domain expertise and business rules, in the end it outputs a highly customized card like this. So again, what’s great about Target is that you can test multiple algorithms against each other to see which drives higher engagement and revenue. And we’ll take a look at this a little bit later. Okay. So in order to start using recs, we need to do the following three things on the back end, which are one, teach target about our content or products. This is where you’ll create a catalog where you’re pass in information about all of the items that you want to recommend. Target offers several integration options for creating your catalog, but the simplest and most frequently used method is to send a CSV file on a daily or weekly basis from your content management system. But you can also pass information on the data layer from your page using Adobe Target’s JavaScript library. So leverage our APIs to pass information directly from your source system or use our Adobe Analytics integration. If you’re already passing catalog data to analytics. And sometimes you might wish to use multiple options together. For example, passing in most day to daily via a CSP and then passing inventory updates more frequently via an API. Next, you’ll need to capture user behavior by adding tags or leveraging analytics to track views and purchases. This data will really serve as the driver for your recommendations. And then finally, besides user behavior, you need to pass along to target the specific contexts where recommendations are being shown. So information about the page in the user profile target will use this information to make personalized recommendations, and that’s done by sending that entity ID for each item viewed when a user clicks on that item. Now it’s time to see some recs in action. So we’ll start with a quick demo popping here into the UI. You can see I’m in my target sandbox and here in the recommendations tab is where all of that information lives. So we start by creating a feed and uploading it and then that feed uploads all of these items into our catalog where they’re available for us to search. This criteria section is where you’ll be able to find all of those algorithms. Sometimes it has an easier time loading than others. Let’s see. And as you can see here, there’s a ton of out of the box algorithms to get started with, and you’re also able to create your own criteria by clicking this button here. And then we also have a bunch of out of the box designs to get you started with the look and feel of your EC. So once you’ve nailed down your design and criteria, it’s time to build your recommendation. So heading over to the activities page, we can click create activity recommendations, Make sure that you’re working in the visual experience. Composer and then I’ll be creating a rec using this experience league page, recommending relevant articles. So just going to insert the activity URL here and then the visual experience composer launches your site here and you’re able to select where you like to place your rec. I want to place it above this. Discover your learning path so you can say which page. Select your criteria I want to do most viewed content across the site. It’s I’m selecting my design and you can add a front or back end promotion here, but I’ll just hit save and then Target will place your recommendation as well as some examples from your catalog. So here are some help articles that it’s place. And then we hit next and we’re able to select what audiences we’d like to use. And how many of those are people you’d like to show your recommendation to? So here we have 10% that aren’t getting the rec and 90% are, which is a great place to start in my opinion. Then simply add your goals and settings and you’re done. It’s a pretty straightforward workflow and really easy to get started with. So heading back to the presentation and let’s take a look at some interesting use cases. So taking a look at some use cases for X, the first being we can test recommendations offers inside an AB activity as well. So that includes auto allocate, auto target, inexperience, targeting activities. And this functionality opens up entirely new capabilities such as testing and targeting recommendations and non recommendations, content within the same activity, easily experimenting with the placement of recs on the page, including the order of multiple recommendations and automatically pushing traffic to the best performing recommendations using auto allocate, as well as dynamically assigning visitors to tailored recommendations based on their profile using auto target. And just to speak to this auto target use case a little bit more, here’s where you can select, let’s say, three different criteria for different visitors. An auto target would automatically leverage the correct one for that visitor based off of their visitor profile. So an example being, let’s say a visitor recently purchased a bike. They could be shown a carousel of bike elements using the bought this but that criteria whereas a new visitor could be shown the most popular items so really helpful use case here. And then similarly you can test recommendations for different audiences in a single experience targeting activity. So if you have an idea of which audience you’d like to be shown which algorithm, then you can leverage that in an experience targeting activity like serving most popular to new visitors and recently viewed to returning visitors. Now a great example of how this would work in the real world. A major telco provider had some aggressively quarterly business goals for revenue and was looking at how they could improve and maximize ROIC with recommendations. A key differentiator for Adobe Target is the ability to embed recommendations algorithms within an AB test activity, which we refer to as the REX as an offer feature. This organization needed to determine the highest possible conversion rate for recommending smartphone to customers and was already using target recommendations for showcasing a set number of products on the product page. After deploying the REX as an offer feature, the team used three separate experiences within their AB test, first with no recommendations. Second, with the most viewed REX algorithm, and third with the top selling algorithm. An impact was achieved here with powerful results. It ends up being that there was a decrease in the original most viewed algorithm and almost an 8% conversion rate lift for the top selling smartphones. Here’s another great success story shared by the senior manager of Nvidia’s web strategy. So the Nvidia team leverages Rex to create personalized experiences for visitors on the home page. They use multiple audiences and criteria, adding filtering options for visitors to customize their experience and the results from this content were staggering. There are 182,000 more home page clicks per year and 150,000 fewer bounces per year, and this nearly doubled engagement for over 3000 pages on their website. So all in all, Rex is a pretty powerful tool. And with that being said, I’ll pass things over to Bill for our category Affinity Overview and demo. Okay. Well, let’s start with grabbing the screen share and I’m going to throw this link in the chat so you can play along. Also. Okay, So this is a mocked up version of category affinity. And what category affinity is meant to do is to recommend a category of either products, articles or any type of content you want to sort out. And this is not going to recommend a specific product, but is going to recommend a category or grouping. So in retail, well, we would often break it down to a very high level product in recommendations. We may go far more granular as far as how we organize the products, but in category affinity, we need to be able to support this by building our content and realistically, if we’re talking hundreds or thousands of categories, you’re never going to be able to accomplish building out that level of content, which is why we typically would keep it to the higher level, you know, like top level of product navigation or if you’re in a lead gen industry, we might be picking up on a person’s role within their business based on the types of things that they’ve clicked on or a, you know, type of content or, you know, really whatever it is we’re trying to push back. So in this demo, what we’re going to do is we click a few products and we assign points behind the scene. So I’m going to start with bananas because I actually can pick bananas from my patio. And immediately it says, While you love bananas, well, I also kind of like grapes. So no grapes have scored the next highest amount of points because bananas were first. That was my priority. And as I add additional clicks, it begins to prioritize both on the recency of this click and how many times this click has happened. So if you’ll notice, I’ve done hotdogs and grapes each once. They both are still behind bananas because of the prioritizing. But now if I click grapes again, it’s going to slide up to the top. The bananas are going to go back down. So as they move from section to section, it’s going to start prioritizing things differently. So the you can see where, yes, we can build out content to support these ten items and you can use like a into your hero on the home page or maybe a secondary content placement on the home page to push people toward this type of content. So in the case of lead gen, if I’ve specified I’m a purchasing agent for whatever it is your business is knowing, you can then start pushing content toward me that supports my line of decisioning. As far as support articles and other information I need to know to leverage all of this within the tool. It is simply setting up an audience and we can create the audience. And this is going to be done under visitor profile and it’s already pre-built. It should already be listed. There would be category affinity and we list both favorite and first. They’re actually the same thing. It’s just we used to do favorite and then realized going from favorite to second felt weird, but not having favorite and only having first felt weird. So we just duplicated it. But we can pick somebody’s favorite category and just the equals and we are going to do a static value and we will just say bananas and then we give that a name and save it. And this can now be used in any activity on that site because of this audience being predefined. So if we wanted to apply this, we would build an experience targeting active 80 add one audience for each of our affinities that we felt were valuable enough to personalize. So in this case, are ten fruits and vegetables and just go ahead and apply the content for those to implement. This is a single tag and what we need to collect is a category I the tag and it is passed just as user act category I.D. and you can also pass multiple items with a comma if necessary. Generally, most clients don’t do that, but it is certainly possible. And in retrieving this, if you wanted to see an array, you can pull back user category affinities in a profile script and it’ll give you the list. But generally people are looking to see the the first couple in particular. If you’re curious about the math behind the scoring, we do show that in the documentation so you can kind of follow along and understand why things are getting reprioritized as you click on more and more products or articles. Any questions on how you might leverage category affinity? Okay. Well, I will go ahead and pass it back to Leanna. Leanna, are there any parts of Rex that we should do an additional live demo on that you feel are value? That’s a good question, Bill. Because I’d be happy to point one out if you’d like. Yes, that would be awesome. Okay. So in the last year or so, we have added a new criteria that is a little bit different than what we’d done in the past. So let’s go into the criteria building interface here. Most people end using popularity and item based criteria the most often. So popularity is just going to be most viewed or top sellers if it’s a retailer and that info was picked up from the call that’s made on the product page or article page and in a retail scenario, also at the checkout, those two inputs form the majority of our criteria, like within a view affinity. So something item based, it’s based on views. So people had viewed this viewed that would only look at the input from the product detail page viewed this bought that’s looking at both the product detail and the checkout bought this, but that is looking only at the checkout and orders with multiple items on it. So each of these may take different amount of times to generate data. So views, we don’t need a whole lot of time because everybody looks at products, but if you have a 3% conversion rate, this is actually going to be smaller than 3% because not everybody buys more than one item on your orders. So make sure to think about that as you set times. But the more interesting ones to me are the user based and in recommended for you. I want you to take a look at how when I pick item based and I picked a view, we have a recommendation key. This is where we’re going to specify the item. We’re going to base this criterion. So it’s usually the current item, but it can be picked up from behavior or other methods. And when we do user based instead that he’s going to get relabeled a little, it’s no longer about an input to the criteria, but this is only used for filtering and the filtering is applied further down. And we’ll take a look at that in a second. But what this is going to do is it’s effectively going to go through the visitors behavioral data and create a multi input recommendation based on multiple products previously purchased products potentially. And it goes ahead and runs criteria against that, against all of those and ranks and scores and sorts them out accordingly to return the most appropriate items. So it is a I would think it was a multi skewed input that you pick up from user behavior. We also allow you to do this in a slightly different manner and instead of user based you can do CART based in this. We’re still doing multi SKU, but you get to decide to pass them to us and it’s passed, as Kurt adds usually from the shopping cart and a logical and then that gives you the control over what are the included items. But from there it is still very much like recommended for use versus being a multi skewed input like that. The one thing I don’t like is that we called this Kurt based. It was developed thinking specifically about shopping carts, but I would prefer you think of this as the multi skewed input because it really can be any list of SKUs that you wanted to pass. So theoretically, if you had a wish list on your site, you could take that list of SKUs instead of Kurt SKUs and pass those through. And we would recommend items based on items in there Wish list. You could also do it from a search results page and take like the first five items, pass those SKUs in and create recommendations on the top search result items. So there’s more flexibility to this than immediately meets the eye. And the other thing I want to cover real quick is sequences. So when we’re thinking of the value of a visitor and the data, we know about them, if a visitor has been moving around our site and we know quite a bit about them, we can be very specific about how we deal with recommendations. I like to visualize this like a wedding cake and the smallest, most important layers at the top, and it gets more and more broad as we get to the base. So our first criteria might be aimed at returning visitors and using previously purchased items as an input. Our repeat traffic is just making up a number 10%, so this might siphon off 10% of our traffic. Obviously that’s not enough. So we need a plan B, So beyond that, we may set up something that is for returning visitors based on our assumption of them having some behavioral data and criteria. Three might be designed for people that we know we don’t have behavioral data for. So now if they fall into this, they get that. If they fall under criteria two, they’ll get that criteria three, people get that. But this also works in scenarios where we didn’t have enough depth to fill the the template. So if you had a carousel that was 16 deep and her first criteria could only give you four items, we’re going to see if we can pull 12 other criteria too. If we only had eight there, we’re going to grab three other criteria. Three. If that came back with a zero, we keep going down the list until we fill our template. We still have the possibility to design our options of do we show backups or not? Do we decide to just give up past criteria? Five And if we only had 15 out of 16 in our template filled, return it that way. So you do still have additional flexibility, but it gives you the ability to prioritize by putting that high value cake at the top, which is for the bride and groom only, or the second layer, which is for your next most important, the people sitting at the head table with the bride and the groom and category three, four and five for the rest of us that just get regular slices of cake. So I hope the analogy makes sense. Criteria sequences are one of the most powerful things that we’ve got in Rex as far as new additions over the last few years. And it’s something that our best clients leverage almost exclusively to make sure that they’ve got the depth of recommended items that they’re after. One of the primary uses is in dealing with categories in particular. So if you’re collecting category data at a high level, a secondary level and even more granular level, you could start out by matching at the most granular loosen it up as you go, ensuring you’re in a full year template. Any questions? Got Thanks. All right. I’m also looking at Q&A and it looks like nobody’s posted any question lines.