RTCDP Insights: Harnessing AI for Customer Experience
Explore how Adobe Real-Time Customer Data Platform (RTCDP) addresses challenges by leveraging artificial intelligence (AI) to provide market analysts with enhanced capabilities, build predictive propensity models, and optimize customer experiences through accelerated insights.
Key Discussion Points
- Challenge, Solution, and Benefits of Customer AI
- High-Level Architecture
- Customer AI Models
Transcript
Hi, all. Good morning. Thanks for joining. We’ll be getting started in the next couple of minutes. Today’s session will be focused on real time CDP insights, harnessing AI for customer experience led by Damian Alston. And we are going to wait just a couple of minutes for attendees to fill them in, and then we’ll get started.
Good morning, everybody. Just another minute or so for the last attendees to fill in, and then we’ll get started.
All right. Hello, all. Good morning. Welcome and thank you for joining today’s session focused on real time CDP insights, harnessing AI for customer experience. My name is Reuben, and I work in Adobe’s Ultimate Success Team as Principal Customer Success Manager, where we focus on helping Adobe’s customers get as much value as possible from the Adobe Solutions. I’m going to go ahead and kick off our session today. First and foremost, I want to thank you for your time and attendance here today. Just to note that this session is being recorded and a link to the recording will be sent out to everyone who registered. This live webinar is a listen only format, but it’s very much intended to be interactive in that as we go through the content in today’s session, please feel free to share any questions into the chat, and our Q&A pod from our team will answer as possible there. And in addition, we have reserved time to discuss questions at the end of the session.
We’ll be sharing out the survey at the end of the presentation. Please feel that, please fill that out. It helps us to, shape future sessions.
Okay. I’m joined today by our presenter, Damien Austin, who’s going to cover how Adobe real time CDP is leveraging artificial intelligence. And now to get started. I’ll go ahead and turn things over to Damien. Damian, over to you, please.
Thanks, Reuben. Hi, everybody. Happy. Tuesday. Morning. Afternoon, evening. Depending on where you are, what are we going to cover today? Customer AI over you. The challenges benefit solutions and benefits. A couple of case studies regarding that. General architecture, some data foundations, and then get into a little bit of tactical how it works, and how to set it up. So we’ll go from theoretical to a little bit of hands on, tactical. And hopefully you’ll wake up from the session understanding the value, as well as, some quick capabilities on being able to set up customer AI in your CDP instance.
So what is customer AI or what is customer? As Adobe defines it and, our CDP, offers built in algorithms. It learns the most accurate model for each customer, based on data set, and it applies statistics and classification models. I e.g. boosted tree models, to apply and allow users to understand segmentation around available populations and events and attributes. So that’s a lot of a lot of words. But what does that really mean. Right. Like what are what are the what are the nuggets. So at a high level, customer I delivers high accuracy from a from propensity models.
It delivers influential factors. So it’ll tell you what it’s doing to, create the high accuracy in terms of propensity models. It’s easy to configure. And it then can allow you to seamlessly activate, from a customer experience perspective. Based on the propensity scores.
But what challenges is customer AI trying to address? Well, in the past or even in the present? Write prediction and segmentation and data analysis is fairly hard and requires a lot of time. It requires time and data. So because, our CDP is built around a model of looking at a 360 view view of a customer or a prospect, we’re taking that, and then applying statistical models to it, to help alleviate some of, stresses that come along with creating prediction and prediction models and segmentation.
So it’s necessary, right? The first part is obviously the 360 view of the customer, that support segmentation and data analysis.
Very difficult to do something or create, propensity or prediction. If you don’t have a clear view of who your prospect, our customer is, it’s very difficult to create hyper perfect personalization. In the same case. So the base is CDP, and what it provides, including the data, fishing, the modeling, etc. and we’ll, we’ll touch on some of those topics in a moment. But what’s really core to think about is if you have your 360 view of your prospects and customers put together, right, and solidified, then personalization.
And as well as additional segmentation and prediction becomes something that is, a stair step up. And then obviously customer AI will help you do the predictive part.
So what’s the key differentiator from a customer AI perspective? Deeper customer insights maximizing impact on ROI and time to value. Now ultimately you are shortening hopefully shortening the time to market for new segments.
And for your understanding how the segments will behave. So customer AI does some of the heavy lifting of what, data team or analytics team would do. It’s not going to do everything right. It’s not a panacea. It’s not a silver bullet. However, it should help from a creating new segments perspective and creating deeper insights from a customer or prospect perspective.
So a quick couple case studies. So Dick’s Sporting Goods, has grown loyalty their loyalty program by 20 million members. They’re doing this by adding personalization, by delivering experiences that are specific to their customers and their prospects, and really by leveraging the real time CDP. Right. Or customer data platform, gathering additional insights from each interaction that they’re having with a customer or prospect.
Similarly, Panera Bread is doing this something very similar, in terms of being able to increase overall subscriptions in their loyalty program, and then they’re able to obviously add additional segments, add additional personalization, and drive deeper insights into their customers, and their customer behaviors. Leveraging CDP.
So if you’re key P customer. This is probably familiar to you. And if you’re not, it may make sense to become familiar with, these concepts. These core concepts are.
Key in terms of, thinking about specific, how you how you look at customers. So you have event data and profile data, and then you have business rules and configuration. So event data is any interaction that a customer prospect is having with your brand or, or your organization.
The profile data is. More, specific data around your customers and prospects. Right. So think of that as you used to be called demographic geo data, membership data, income range, etc., and then business configurations. How does your business view what happens in the future. Can you do you have forecast windows? Do you have specific rules around eligible populations whom you can speak to, whom you can market to, etc.? And then what’s the outcome? The outcome is breaking down your business, segmentation, your customer experiences, hopefully enhancing those and personalizing those.
And then overall business planning. So or merchandizing calendars or marketing calendars in general can be affected. In terms of what your customer or prospect behavior looks like.
So how does customer AI sort of blend into this? Well, customer AI lives inside of app it is available to everyone that has a CDP.
And it configures it’s configured based off of the data that you have and your customer data platform. So you’re able to run models and develop insights against the data that exists in your customer data platform. You can also export the models, and prediction into additional platforms if you need to. So you can download Raw scores and push them into a different database or a different client.
You can layer additional, reporting suites on top of, your, your data set, or you can do everything inside of the Adobe suite. So leveraging, Adobe Journey Optimizer, customer journey analytics, is a capability as well, but you’re not pinned in to the Adobe ecosystem. Although obviously, the those platforms work well together.
So what’s core, in terms of pulling all of this together? Well, more data enables better prediction. So you’re able to leverage the app APIs and 50 plus data connectors to connect your data to get the 360 view of your customers and prospects, which is sort of the foundation of doing any predictive modeling, right? Understanding what the actual customer prospect looks like.
So that’s key.
The real time CDP helps you collect that data, which is why customer AI sits on top of that.
And again, we’ve already sort of talked about pulling the data together. Just think about your modeling. Right. The first party known customer data, second party data and partner data can come together in the CDP. And that’s what you’re leveraging to create your models.
There are a couple zdm schemas that are supported.
The Adobe Analytics schema, the customer experience event data, and the Adobe Audience manager Zdm schemas are all native. And you can pull in a number of those data sets. The key ultimately is connecting those data sets and having a solid model right of your data prior to trying to model or use customer AI.
So you have also the concepts of data completeness. So data completeness. What data is contained in your data sets and what data is available for each data set.
Is important in terms of the fidelity of your model.
But again, ultimately what you’re looking to do is connect your data sets together to create the single profile of the user, by which you’ll be able to then model segments, and get predictive scoring, based off of those models.
So let’s talk really quickly about use cases before we get into some of the more tactical aspects of how to get customer AI set up. And sort of its general simplicity in terms of the model. So what is customer I really focusing on basically conversions and terms. So when you’re building a customer model, typically you’re going to give the system an understanding of what does a conversion actually look like in your use case. And also what is churn actually mean in your use case. So that it can review the data sets that you have and give you an understanding of, how it’s predicting the behaviors that will result in a conversion, as well as the behaviors that result in a churn. It’s not just behaviors. It’s also the overall look and feel. Right, like it’s doing lookalike modeling as well. From a profile perspective, but mainly right.
It’s driving at what’s the conversion? What can it predict what will happen from a term perspective. So this use case financial services, there are a couple of different ways that a financial institution might consider something as a conversion. Right. Viewing particular products that might be a signal, submitting an inquiry form that might be considered a conversion. Opening a retirement account may also be considered a conversion, and then adding additional share of wallet wealth management services, maybe considered conversions, closing accounts or not completing opening accounts? Maybe consider churn and you can apply the model, and you can apply customer AI to get a sense of which types of customers based off of behaviors and profiles will, complete either of these two activities, i.e some conversion or churn.
Retail example very similar. There are multiple conversion paths from a retail perspective. Signing up for newsletters, searching for relevant products, completing purchases as well as churn examples which is open cart or abandonment of cart rather. So if you are able to predict based off of previous behaviors and profile, then you can tailor the experience and tailor the segmentation, for a specific customer or prospect in order to lean them towards, actual conversion and away from churn, or maybe even potentially win back. In the case of retail, if you have one of these churn examples.
So theoretically, we’ve talked about customer AI helping from a conversion perspective, creating models for that churn, prevention, creating models for that. How does it actually work and what are the tactical steps to get customer AI set up? Well, first we want to think about how custom AI works on top of our CDP, customer and rely on the individual behavior data. In order to help create the scoring and modeling for propensity. It’s flexible. It can take in multiple data sources, including your third party data sources, as well as any data source that exists across your Adobe platform stack. So Adobe Analytics audience manager, and then any consumer experience data events that you may have captured. So key is ingesting data and hopefully lots of it.
Then once the data is ingested you move on to configuring. So mapping the insights or the behaviors against what it is you’re looking to model. So think about how it is that you’re going to.
To classify something as a conversion or something as leading to a conversion. So that we can get to what the customer probability to purchase a product or upgrade is. Lastly, once the model is created, you can let the model run, and then it’ll give you a prediction score. You can take that prediction score and then action off of that. Create creating new campaigns, creating additional behaviors. Create additional experiences for customers and prospects.
The modeling and training can take up to 24 hours in some cases with larger data sets. It may take a bit longer than that, but ultimately, modeling and predicting is something that should be thought of as a continuously optimized flywheel. So we’ll talk about that in a second. Once the predictions are completed, you can then action off of them.
So how do you set it up? Create a new model. Select the data set that you want to be included in that model. Define the goals around the model. Run the model. Create the segment. Activate the segment. So those six steps, are fairly straightforward.
And in the tool there’s mapping to help you, simplify them. Right. So if you go into the tool, you can go into the configuration area for customer. And you can define the prediction goal, the eligible population, the custom events, the profile attributes that you want to consider.
And then you can define the define the schedule.
Defining the schedule is important. And redefining the schedule and retraining is also important.
Once the model is configured and completed, the scoring can take up to 24 hours. I think I’ve just mentioned that, it can also take a little bit longer. Subsequent training will run automatically, and score the user’s configuration. In the training window, you’ll want to. Retrain the model from an optimization perspective consistently.
This allows the model to get better over time. Right. Just like any sort of optimization. And quick tip overall best practice is the model, retrain cadence is weekly for quarterly retraining or monthly for, if you want to retrain every six months. Sorry, I think I said that wrong. What I mean is, if you want weekly scoring, you should retrain it quarterly.
And if you want monthly scoring, you sure you’re training? Retrain it every six months. So that is really dependent on, what your marketing cycles look like and what your merchandizing cycles look like. And should be determined by your overall marketing team.
So what does the output look like ultimately of, Of customer AI, and of this process. You get essentially four things as an output from customer, customer, AI.
Post configuration and post training and run get the score, which is relative likelihood of a customer will achieve the predicted goal probability. So probability of achieving that goal percentile. So ranking inside of a segment for achieving that goal for a particular population. And then the influential factors into influential factors are the system basically telling you here’s what I considered in order to come up with this model. So all four of these pieces of information, are helpful on their own. And then obviously together, it can help to drive additional insights into what’s happening. So a little bit of, Data, analyst team unto itself. Right. If you don’t have the resources for a full blown data analyst team, this can help. And if you just want a quick hit, to understand what’s happening in a particular campaign, this system will help.
What are the reports look like? Well, the model output. Right. Is the score summary and the distribution and the influential factors. It also gives you some sense of what the historical performance looks like. Ultimately, doing prediction, has to do with understanding historical performance. So this gives you a good view into, what things look like and, and how the system got to the prediction, and the modeling that it’s providing.
The insights page gives you the latest scores. So as the scores, propensity scores are delivered, all the scores generated by the model are categorized into three buckets the high, medium, and low propensity, with reasons that help you understand why the user belongs to each bucket.
The latest score is tab gives you an understanding of. The historical and influential factors for the propensity bucket.
The breakdown of each values for the influential factors, and understand what behaviors and attributes contribute positively to the predicted outcome. So you get a better understanding of which activities, behaviors, and demographic elements, can help you target individual customers or prospects, for the specific experiences you want them to have.
The historical performance tab breaks down previous performance.
Track outcomes and is also a automated way of testing a model’s prediction against realized outcomes.
New model evaluation tab and lift chart. So live charts are a good measure of the model quality. This helps the user measure the improvements of using predictive models over random targeting. The lift is the rate of the outcome in the bin, divided by the overall rate of a positive outcome across the entire eligible population. Higher lift values of the first few deciles mean that the model is good for identifying users that are likely to take an action or interest. So this allows you to target the populations and the segments that come out of the models. For the highest performing, sets and deciles.
So what does that mean? Ultimately? How to use customer AI scores and audiences. Marketers can use high, medium, and low propensity segments from customer AI to build additional and smaller segments for targeting.
If you’re looking, depending on what you’re evaluating, low propensity, medium propensity, and high propensity can help you actually target behaviors.
For, particular campaigns or particular experiences. They can create subsets of audiences, with specific scoring. So that’s additional segmentation. And then creating create experiences around those specific scores and subsets, to enhance performance of campaigning.
So that’s a bit about how we tactically get there from a configuration perspective and what customer AI delivers. From a data perspective, privacy and governance, is also key.
Customer AI, privacy and governance, or at least the tool has privacy and governance enabled, with audit logs, policy access, platform privacy services, field access controls, and automated data sets. There’s a little bit more information available in this deck, for additional resources across customer, overview and enablement videos.
I think at this point, we can move on to some of the questions I saw a bunch of them come through.
And, Ruben, I’m assuming you’re you’re categorizing them and we can kind of discuss some really quick. Yeah, absolutely. Before we do that, and before we get into the Q&A portion of our event, I think we should have perhaps just lost the poll real quick. Sure. Let’s do that. Get that off. So as we go.
There you go. Yeah. So if you can fill that out as we move into the Q&A session, I’ll definitely have a quick look here. We have an overview, of the questions here in the chat. That’s one from Pavan. Will the system decide which model to use based on the data set, or is only the symbol tree model available in real time CDP? I think this was related to a slide you showed earlier.
Damon. Yeah. So there’s, only the singular model is available. It’ll it’ll produce, it’ll produce predictions based off of that model.
And you can incorporate that model basically, based off segmentation, etc., or include that model into your segmentation, etc…
Right. Thank you. Here’s another question from Jason. How do you attribute approve ROI for this customer AI approach? So you if you have benchmarked, revenue as an example, or if you have benchmark activities, you can then compare your benchmarks against additional segmentation or these additional models. So you can imagine taking a subset of a segment and changing their experience in an attempt to get them to convert at a higher rate, and then do a comparison against your historical benchmark. So, you should be able to to easily determine if the predictive model is actually, driving additional conversions, just based off of, based off of the additional marketing or, the additional events that are happening. And over time. Right. Which is why the, the.
Which is why, optimizing the models is key, right? Or retraining the models is key. You should see additional benefit or additional scoring happening. So the model should adjust itself over time, to help improve ROI as well.
Great. Thank you. Damon. There’s another question here. Is customer AI available to all real time CTP customers, or is it a separate skill? It is not a separate skill. It’s available to all our CDP customers right. Let’s see. Here’s another one. Any other data source limits? There are data source limits. There’s a maximum limit of 20 data sources.
In order to, in order to, to leverage customer AI. So, yes, there is a data source limit.
But 20 data sources is, is, a large limit. And additionally, what’s key is focusing on making sure that the, your, your schemas and models represent an individual 360 customer view.
Right. Another one. But visualization tools can be integrated with custom AI to effectively communicate insights to stakeholders. Can you can you speak to that, Damon? Yeah. So I think there is a slide that mentioned, you can leverage CJA, right? As a, visualization tool. If you have other third party visualization tools, you can also, ship the model data into those tools and leverage those tools as well.
Right. Let me just double check the Q&A portion. So I’ll just say.
How does customer AI help in predicting customer lifetime value and tailor marketing efforts accordingly? So, customer AI is focused on conversion rate and churn, sort of the two basic, event types. So if you’re preventing churn, you are inherently increasing customer lifetime value. So that’s one of the ways, right, by decreasing churn.
In another way, you can leverage lookalike models of a high value customers or customers to have, you know, the larger lifetime value.
And create models around those customers, to then create experiences to drive people, to become those types of customers.
Great. Damon I think that’s it. On the Q&A side. Let me just double check the chat here again to see if there are any unanswered questions.
There was a question. Can your presentation be available after this presentation? Yes, we will share, the presentation and a recording of today’s session with, everybody who signed up. So look out for an email in your inbox on that.
All right. I think we want to wrap up. Thanks again to all of you for taking the time to join the session today. We hope you and your company, will join us again on future webinars. Thanks, everybody, and have a great day.
Thanks, everybody.
Overview
The webinar focused on Adobe’s Real-time Customer Data Platform and its Customer AI capabilities, led by Damian Alston. The session covered how Customer AI leverages built-in algorithms to create accurate models for each customer, applying statistical and classification models to predict behaviors such as conversion and churn. Key benefits include high accuracy in propensity models, ease of configuration, and seamless activation for personalized customer experiences. Case studies from Dick’s Sporting Goods and Panera Bread demonstrated the practical applications and benefits of using Adobe’s real-time CDP, such as significant growth in loyalty programs and enhanced personalization efforts. The webinar also detailed the implementation process, which involves creating a new model, selecting data sets, defining goals, running the model, and activating segments. The importance of having a comprehensive 360-degree view of the customer and continuously optimizing the model for better predictions was emphasized. Additionally, the session touched on data integration, privacy, and governance, highlighting that Customer AI is available to all real-time CDP customers and supports multiple data sources and visualization tools. The Q&A session addressed various questions about model usage, ROI attribution, data source limits, and visualization tools.
recommendation-more-help
abac5052-c195-43a0-840d-39eac28f4780