With Braze features like AI Item Recommendations, Predictive Churn, and Catalogs, marketers can create campaigns that drive re-engagement and conversions through truly personalized messaging.

In this episode of Thread the Needle, Tatum Lynch, Solution Architect at Stitch, covers the key components of setting up successful product recommendations, including:

• Laying the foundation: Ensure clean purchase data, a robust product catalog, and alignment on personalization goals.
• Building the engine: Use Braze’s AI models — like personalized, trending, or most popular — to power dynamic product recommendations.
• Activating across channels: Combine AI-driven insights with churn predictions to trigger personalized messaging that re-engages at-risk users.

‍

‍

Transcript:

Hi there. My name is Tatum Lynch and I am a Solution Architect here at Stitch. And today I'm going to be talking about how to approach personalized product recommendations, in Braze everything from initial discovery questions to solution design, and finally, how to activate this across channels.

Personalization is always the end goal to drive high engagement and Braze makes it very achievable with the right building blocks. So we get a lot of these questions from our clients. What does it take to be able to get to this level? And specifically for product recommendations, we need three items.We need a product catalog within Braze, and we're also going to leverage two of Braze's AI tools — Predictive Churn and AI Item Recommendations, which we're going to talk more about here in just a few minutes. 

Before we can even get started we really have to lay down the foundation, do some discovery, and really just understand what's your brand's end goal. What are you hoping for these campaigns to look like? So before we can even get started, we have four different buckets that we really need to nail down. 

First of those being data foundation. Not only do you need a product catalog, but we're also going to need some purchase data or event data within Braze. And so does your team have that? That's going to be something that's going to help drive the AI Item Recommendations model. So it's really important that you have that. And if you do, how much of a backlog do you have?

Of course, as you can imagine, the more historical data that Braze has, the stronger your model is going to be. So that's something to definitely consider. If you have that already in Braze, fantastic. And if you don't, do you have the capability to be able to do any sort of historical backlog and then have that as an ongoing source as well?

Second is going to be your personalization strategy. There are many different AI Item Recommendations models that you can use, and so really just understand with your team, which one you want to use. To start off with, we're going to talk about a few different types that Braze has, which is awesome, so we're going to be diving into those. 

And then also, what sort of personalization details do you want to include in your campaigns? So of course you're probably going to want to include that when you're recommending these products. You're probably going to want the product name. And also you're probably going to maybe even want to have a link that goes to that product page. 

It's understanding details like that to help you understand, okay, this is the type of information we need to build out the product catalog. Because what we're essentially going to be doing is making calls to that product catalog to say, “Hey, this person, we want to recommend this specific product. Please give me that product name and the link to that product.” And maybe even potentially if you have specific copy or a product image that you want to pull in as well. Those are all things that you can do with this solution. But you need to have these conversations to understand, okay, these are the personalization details we want. We need to make sure we have that built into the product catalog. 

Third is going to be channel read readiness. Which channels does your team want to use? Are these already active? in Braze for you? Do you have your creative templates in place? And also just understanding what's your desired frequency for recommendation as well.

And then finally, as I've mentioned, that product catalog, do you have it within Braze? That’s something that's very important. Is it up to date? Do you have a flow within the process that you're always making sure that the product catalog within Braze, if there are ever any changes that these are also being reflected in the platform?

Alright, so let's go ahead and dive into those three buckets that I mentioned at the beginning of this, just so we can understand what exactly are AI Item Recommendations, predictive churn, how can we use it, and then put everything together. So what are AI Item Recommendations? As you can imagine, Braze has been using a machine learning model that allows brands to be able to build this recommendation engine within Braze.

And so like I mentioned, they have different models that you can build. These are three, I would say of their most common. First of those being AI Personalized. This is where you can predict what a user is most likely to be interested in based on what they've shown interest in previously. 

Of course, self-explanatory, Most Popular. This is just where you're recommending items that users engage with the most. So this can be things like most popular items, most popular liked items. If you have any sort of custom event that indicates someone has liked an item, most viewed items and popular items in user's carts, right? So you're now understanding why we need to have either purchase data or event data within Braze. It's so Braze knows, “Hey, okay, a lot of people are buying this specific product or a lot of people are viewing this specific product.” We need that data to be able to build the engine. 

And last but not least, we have Most Recent. So as you can imagine, this is just recommending items that a user engaged with most recently. So this can be things like recently clicked items, recently liked, things along those lines. So let's go ahead, let's get into the Braze platform and just see exactly how we can build a recommendation engine.

 Okay, so I'm within the platform. If you're on your homepage right now, what you're going to do is go to analytics and you should see under predictions, AI Item Recommendations. And what we're going to be able to do here is if you go up to create prediction, go to AI Item Recommendations. This is where you're going to outline all of your details.

This is where you can put the name of the recommendation, a description as well. And this is where we get into the nitty gritty details. So this is where you're going to indicate what type of model you want to build. Just like we talked about Most Popular, AI personalized, Most Recent, and it looks like they also have a trending model as well. So if there's a recent surge in user engagement with a particular product, this could be one that you want to use. 

So I'm going to go ahead and just click Most Popular. And then what you're going to be able to do as well is you're going to be able to select a catalog. In this dropdown, what you're going to have is that product catalog that you have built. And so once you've selected that, going to go ahead and just click products. If you also have a specific catalog selection that you specifically want to use, if you're not familiar with catalog selections, just simply think of them as a filter, just a way to be able to filter the huge product catalog. So if you only want to maybe recommend a specific set of products, you can add a filter or a selection into your catalog and use that to specifically build your recommendation engine. So you do have that flexibility as well if you would like to. 

But otherwise, the last step is just choosing how you want to track those events. So like I said, we just need to be able to know, are we tracking this based on products people have purchased, or are we going off of a different kind of custom event? So I'm going to go with purchase, and then this is where you get to select your property name. So what I want to do, what we're just selecting here, most of the time it's going to be your product ID. Simply what this is just saying is it lets Braze know, “Hey, when purchase events are coming into the platform, I'm going to be specifically looking at the product IDs that are coming through.” That's going to help me build my engine to realize this specific product ID is one that a lot of people are purchasing right now, so I'm going to use this to build the model. So most of the time it's going to be the product ID that you select there. Then from there, that's it. That's all you have to do to create your recommendation. 

Now, I'm going to go back here. I’m not going to save this and just show you this demo item recommendation of one. It's actually been built, so few different analytics. Here we have precision. It's going to be something that's really important. So this is a percentage of time the model correctly guessed. 

The next item, a user purchase. So always something good to know. And coverage is what percentage of available items in the catalog are recommended to at least one user. So valuable information for you to know. 

This also outlines the different recommendation types. Just for your all's awareness like we talked about, if you don't have, maybe a lot of purchase data within Braze right now, that's okay. But of course, the more data you have, the more essentially energy you're going to have to be able to build that model. And so what Braze does, if a user doesn't have a lot of purchase data on their profile, they will fall back to this Most Popular recommendation type. So just be aware of that if you ever see, “Hey, I set this to be AI personalized. Why are some of my users getting this Most Popular model instead?”

Then you can even see here which items have been recommended to users as well. And then it just gives you an overview, right? Like I said, this type is AI Personalized. For this one in particular, and this is something that you do have the option to choose, you can exclude items from users that they've previously interacted with. So if you want to make sure you're not recommending an item to someone that they've previously purchased, you have the flexibility to say, “Hey, I don't want to do that.”

This is how you build the model, right? So just want to make you aware of exactly how to be able to see what this is, how this is being built in Braze, and how easy this is to do. So now let's go ahead and dive into and ahead. Don't save, we'll go back to our tab here and now let's go ahead and dive into predictive churn.

So what is predictive churn? What we're able to do is that your business is able to define and select an audience that you want to monitor and just keep track of, “Hey, have any of these users churned?” Again, very similarly, Braze is building a machine learning model and they assign each user a churn risk score from zero to a hundred.

And that's just based on patterns from past user behavior. Of course, the higher the score, the more likely a user is. to churn. And so what does your team need to be able to do before you can go ahead and create a prediction of which users are going to churn? 

You first have to define what your team thinks of as churn related to your user's lifecycle. So this can be something that a user, an action they take, or an action that a user does not take. So this can be things like if you have a custom event coming into Braze called, trial expired that indicates, “Hey, this user, they allowed their trial to expire.They didn't go ahead and continue with us.” That's an action essentially that they took within your platform, they performed the “trial expired” event. Something maybe that you could say a user did not do. 

Something that I would say is widely used is session data, right? So if you haven't seen anyone come onto your website or your app, and log a session within Braze for a certain amount of time. You can indicate and say, “Hey, that user has churned.” So then you also can have the flexibility to define your audience. So you can either build this model based off of all your users or off a specific segment of users. And then also you can update the frequency to understand exactly how often you want to update your prediction.

 Alright, so let's go ahead and get back into the platform. And very similarly we're going to be back in analytics. You can see here we have predictive churn. So if I go to again, create prediction, churn prediction. That's okay if we don't have enough active users, but just something to be aware of. 

So what we're going to do here is we're just going to simply type in a name. Then we're going to move on to, like I said, we have to define what churn means to you. So I'm just going to do something where the user has not started a session. Now you can see here this is very similar to the Segment UI. So you can do things with and or or operators as well. So a lot of different flexibility when building this audience.

And then this is where you can indicate how many days a user has to meet this churn definition in order to be qualified. So for example, if I want to say, “Hey, users who haven't gone to our site in the last seven days, we've noticed that they typically fall off.” So that's where I'm going to go ahead and just say — it's been seven days since they've started a session. I want them to be included in this model. And this just gives you a little bit of an overview of exactly how many past churners there have been based off of this definition, et cetera. 

So if we go to prediction audience, I'm just going to go ahead and leave it as all users. Then this is where you can indicate your update schedule. So again, this is just something where maybe you want to update this weekly. Maybe this is something that you want to do monthly. It's really just based off of whatever is best for your brand, and then this is where you get to build your prediction.

So once everything is actually set and ready to go, if we go back here I want to show you, once you have built your model, what sort of data are you going to be seeing on your side. So as you can see here in this image, you're going to see of the users that you included, who are the people that have a low risk, they have a low churn score, and then we go all the way up from medium to high risk.

And this bar indicates to you exactly how many of these users fall into each category. Listed here is also a prediction quality, so it's always really good to pay attention to that, to make sure the model is actually producing solid results for you. This can just be dependent on how many users and how many, how much data you have going on right now within the platform.

So if you have a prediction quality that may be low, maybe just reevaluate the audience that you defined and potentially also, how you're defining churn. Are you looking at an event that maybe people don't perform as often, so the model doesn't have enough data? There are a few things that you can do to manipulate and make sure you're building a solid model.

And then these estimated results, of course. Always with everything with ai, right? Nothing is always perfect. So Braze is of course, transparent with that and just wants you to know exactly how many of the expected churners will be targeted based off of, “Hey, if you're doing high risk, how many of those users are actually potentially people who have churned?”

And they also sense down here how many expected non churners will be targeted. That's what I said before, right? AI's not perfect within the model. There could be people that essentially were thinking they could churn, but actually they may not. So Braze is just transparent with that and just lets you know exactly how many of the people, that are more than likely going to be churning, are going to be included within this audience.

All right, so now we've built both of our models, which is great. How do we stitch everything together? So what you're able to do, regardless of what channel you are using within Braze. We're very familiar with the composer editor and something that's really great to have on the left hand side that you'll find is this personalization button that's underscored here when you click on that.

Braze makes it super simple and easy for you to be able to start pulling in these recommended products, and to specifically target users who have churned, right. You can actually see here. If I go back to the slide, you can see, “Hey, once I have actually built my churn prediction, I want to create a segment or create a campaign based off of this audience.” So immediately, once you click that button, you can stitch everything together and start pulling in based off of your AI Item Recommendations and say, “Hey, these are users who have churned. I really need to add some personalization to get them back into my app, to get them back to my website. What, what sort of personalization tools do I have with me?”

Well, you can use your AI Item Recommendations now, right? So once you click this personalization button, you can see here, all you'll have to do is select the personalization type as item recommendation. You just select the item recommendation name. This is going to be your catalog name, right? So that'll be simple, super simple and easy. Once you select that from the dropdown, then you click the number of predicted items. So say you want to pull in for this user, “Hey, these are our three most popular product items that we have. Check 'em out.” 

You would indicate, “Hey, I want to pull in three predicted items and then I want this information to display.” This is what I was talking about earlier, right? What personalization details do you want to include? Do you want to include the name? Do we want to link to that product page either on our site or deep link into our app? Do we have a specific product image that we want to pull in? That is all of the information that you have to pull into this information to display dropdown. So just be aware of that. 

And then what's amazing, after you have indicated all of these boxes, this Liquid snippet Braze is already done. All the coding that you need, all you have to do is copy this Liquid snippet and simply just add it into your campaign. Wherever you want to pull in the product name, you can just paste the Liquid, whatever Braze gave you. Super simple and easy. 

All right, so what are the next steps? This is so cool, right? There are so many things that you can do within Braze. It's not going to take you a lot of time, as long as you just have the right strong foundation and truly understand what your brand's goals are for these types of campaigns, right?

So, like I said, start with the right data and discovery. Use these tools to your advantage. Leverage catalogs and AI to drive your personalization strategy. And then you can, you know, the sky's the limit from here, activate this cross channel. You know, you can have this in one canvas, you can use this in multiple places, right? So you can have a strong impact here. 

Thank you so much, so excited for all of you to take this on. Do just want to have a last call out here. Of course, Stitch, we want to work with you. So if there is especially a lot that you have going on, and this is the type of work that you want to do, Stitch is here to help. Please reach out to us and if anything, please at least, subscribe to our Cutting Edge monthly newsletter. It's amazing. It has a lot of tips and just MarTech trends just to keep you all on your toes, right. This is a fast moving environment right now. So definitely please sign up. You can see it here where you can sign up here at the bottom.

Thank you so much. Really appreciate it and good luck.