Speaker 0: Everyone, and thank you so much for joining us today. Um, hopefully, you got a chance to tune into the keynotes that we had earlier today. Um, we had the now and next, the future of marketing in 2025, and we also had, uh, just now, the marketing cloud, uh, product road map. So, um, just want to get started today. My name is, uh, Christina Anderson. I’ll be moderating today’s session. And before we get started, I have a few housekeeping items to cover. First of all, as you may have been hearing throughout the day, um, yes, this session will be recorded and available on demand after the event, and we will also be following up via email. If you have a question during today’s, uh, presentation, please feel free to use the q and a chat and then also, um, use the chat, get involved, um, share emojis, gifts, we wanna hear from you. And, um, without further ado, I would like to introduce our speaker, um, Durga Prasad Khorothu, um, who who will be sharing his expertise about, um, boosting predictive analytics with Data Cloud and Einstein Studio. Durga, I’ll let you go ahead and take it away.

Speaker 1: Hello, everyone. So let me quickly share my screen first.

Speaker 0: Thank you.

Speaker 1: Yeah. So hello, everyone again. So today, we are going to discuss about boosting predictive analytics with data cloud and Einstein Studio. Just give me a moment. Yeah. I hope you guys can see my screen. Right? Yeah. Okay. Yeah. Okay.

So before we dive in into the demo, a quick introduction about me. I’m Durga Prasad Karothu, a senior Salesforce consultant working with Magnity Technologies with over ten years of IT experience. I’m also a founder and author of Salesforce Chronicles, where I share insights on Salesforce trends and solutions. I have spoken at several Salesforce events, and I am passionate about learning, mentoring, and traveling. This is a bit about me. And a big thanks to our incredible sponsors who has made this event happen.

And let’s jump into our today’s agenda. Today, we are going to discuss about predictive AI and what is data cloud and how data cloud works and what is a model builder and a quick demo, as well as we will be discussing the key aspects.

Let me ask you this. What if you could predict your business outcomes with precision in order to understand the customer in a better way and provide the meaningful insights to the customer? If the businesses are able to achieve this, that would turn a profitable insight to the business. Right? But one of the thing is that 79% of the organizations fail to harness the customer data effectively. This is this was given by one of the survey, not by me again. But the reasons could be anything like why the businesses are not able to utilize the data that was available in the systems. When the businesses have the data in huge volumes, it won’t be that easy to generate the predictions or bringing all the data together from various entities. The reason could be, in order to get the data from various systems into a single entity, it might require huge amount of development effort like integrations need to be done and lot of modeling that needs to be taken care of in the back end to generate the predictions.

Let’s take a quick journey through the AI’s evolution, how AI has evolved. Let’s start with the artificial intelligence. Artificial intelligence gave the computers, like, a power to analyze the data, learn, and make decisions. These are nothing but, like, the smart assistance. Let us say, for example, we have a service AI, which will usually transform the which will usually transfer the cases or the questions that the customer might ask based on the previous cases which are available.

Then let’s see. Machine learning. Machine learning is also comes under the artificial intelligence, but machine learning utilizes the algorithms as well as statistical models in order to generate the predictions. For example, let us say we are using a traffic app that uses a machine learning to forecast the congestion by analyzing the data that the algorithm is receiving from the GPS enabled vehicles. Based on the information that is received, the traffic app will reroute us to avoid traffic jams in the real time, which in turn will be beneficial to the users.

The next one in the AI is the deep learning. Deep learning, again, is a subset of machine learning. Deep learning utilizes the neural networks, and it mimics the human brain kind of structure. For example, in health care specifically, deep learning can process the data from variables to monitor the things like heart rate and helping the doctors to provide some meaningful recommendations to the patients.

Lastly, we have generated AI, and this is the era which we are in. Generative AI is a subset of the deep learning. Again, using the generative AI, we will be able to create some personalized emails or create images. Generative AI doesn’t only understand the data, but it will try to create the content in a new way. For example, marketers can use the generative AI to tailor the emails for each customer, making the interactions feel unique for each and every customer. This is about the evolution of AI.

Let us quickly understand what are the predictive models. Predictive models are nothing but these will generate the output based on the historical data that is available. Predictive model utilizes the statistical models and machine learning algorithms to predict the future outcomes. Predictive models can be classified into two types, and there are various types again. One is regression model, and other one is the binary classification. In regression model, we will be able to predict the number. The number could be like a currency or a count or a likelihood of the percentage. Some of the example use cases for the regression model can be amount of an opportunity, the time it might require to close an opportunity, similarly customer satisfaction for a specific case, all these things. Coming to the binary classification models, we will be able to predict an outcome of the group either in two values. This can be an yes or no, lost or won. Similarly, some of the example use cases for the binary classification could be like, is the account going to be churned or not? Is the opportunity won or lost? This is about the predictive models.

And using the predictive models, we will be able to do lot of predictions in various areas. Let us see. In the Sales Cloud, we will be able to predict whether the account is going to be churned or not. Similarly, whether the opportunity is going to be won or lost. And also, we can predict or the predict the revenue for that specific account as well. In Service Cloud, we will be able to predict the likelihood whether that particular case is going to be escalated or not. Similarly, what is the timeline for closing that specific case as well. And, also, we can predict the revenue. And, lastly, we marketeers can use the forecast to predict whether that specific email is going to be opened by customer or not based on the historical data again. These are some of the predictions which we can build using the production models.

Let me ask you a quick question. Who has not used the Chat GPT over here till now? Anyone? I think many of us have been using the Chat GPT nowadays for our day to day work. Let us quickly understand the differences between the predictive AI and generative AI. Predictive AI and generative AI are very closely related to one another, but there are differences between how this works and what are the areas this can be used. Predictive AI works based on the historical data that is available to predict the future outcomes. But whereas the generative AI creates the content based on the existing data, and it understands the data and creates the new content. Predictive AI, in order to predict the future outcomes, utilizes the regression analysis, decision trees, and neural networks. And whereas a generative AI utilizes the elements and GANs. Predictive AI enhances the generative AI by predicting the outcome which the generative I, uh, generative AI might need and what is the best possible outcome for such scenario the predictive AI will be able to predict. Whereas generative AI will complement the predictive AI by creating the content based on the information that is available. Some of the examples for predictive AI could be what will be the best possible time to close this case, whether this particular customer is going to purchase this product or not. Whereas for the generative AI, if we wanted to create a specific email to the customer, similarly automate some of the tasks, all these things can be handled using the generative AI.

Now let us look at the islands of disconnected data. Imagine you are a company where we are having the data related to the customers scattered across various systems. The systems can be AWS, GCP, or any other entities. In order to utilize the data available across these systems, it will be a cumbersome process. But we have a data cloud we have the Salesforce data cloud in order to bring these islands of disconnected data together, and Data Cloud helps us to achieve this. All with clicks, we might need sometimes coding as well. I won’t say only clicks is needed, but for some scenarios, we might need to do coding as well.

Let us quickly understand how data cloud works. Data cloud has various phases in it in order to bring the data and utilize the data within the data cloud. In data cloud, we have the connect, harmonize, unify, analyze, and predict, and act. In order to connect a data source, data cloud firstly support various sources like sales cloud, service cloud, similarly, Amazon s three, or any data warehouses, APIs, SDKs, and Biosoft. Once we have the desired connector configured within the data cloud, next, we need to create a data stream within the connect phase. And once a data stream is created, the data will be returned to the data lake object. Data lake objects are nothing, but these are similar to the tables where our data resides. Once we have the data available within the data cloud, our next step and crucial step within the data cloud is harmonize. Why it is important is without harmonizing the data, the data cloud will not be able to generate or predict anything within that specific data. It requires for us to harmonize the data for the data cloud to understand. In order to harmonize whatever data that is available in the data lake object, we need to map it to the data model objects. And once the data is harmonized, we will be able to create the unified profiles And in the analyze and product phase sorry. In the analyze and product phase, we will be able to generate insights on top of that data. Whatever data that was unified and harmonized, we will be able to create some in, uh, analytics using the Tableau or CRM analytics and so on. Similarly, we can run some predictions on top of the data that was harmonized as well. And once we have the desired predictions and insights, we can push the data to various entities or within the Salesforce, either by using the flows, activations, or the data actions which is available as part of the data cloud. This is how a data cloud works.

Now let us quickly understand the Salesforce AI architecture. Salesforce AI architecture is built on top of the Hyperforce platform. This is the Salesforce next gen platform. And on firstly, here, we have the unified metadata framework. Within the unified metadata framework, we have various layers. Again, each layer comprises of its own activities that are going to be happening within that specific layer. The first layer will be the data cloud, which we have seen earlier, where we will be able to bring our data from various sources. And the next layer is the model ecosystem. Within the model ecosystem, we have the Salesforce provided models as well as the large language models. And the next layer is the Einstein trust layer, which is the very crucial layer within the AI architecture because this layer will be taking care of all the privacy and compliance related to the data. Here, we have various mechanisms to sales where Salesforce has implemented in order to take care of the privacy and compliance. And the next layer is the AI tools and builder. This is a place where we will be building our models or building our prompt responses, responses, similarly building our agents. And on top of it, whatever agents which we are going to build will be those can be of autonomous agents or assistive agents. And lastly, we have the apps and embedded AI where this particular AI features can be leveraged in that specific applications. This is about how the AI architecture of Salesforce.

Now let us quickly understand the model builder, which we will be going to use as part of building our prediction models. And model builder is a very useful tool where we doesn’t need to write any code in order to create a prediction model. Usually, in order to build a model, we might we might need to be a data scientist or having our we need to have some background with respect to to the machine learning algorithms, then only we will be able to create a model. But in this case, we can just create a model just with clicks where we don’t need to write a single line of code. Whatever model which we are going to build using the model builder, we can design a model as well as train that specific model and deploy that predictive model as well. We can bring the desired dataset using the data cloud using the data cloud. And also, as per our business business requirements, we can use either the regression or the classification techniques in order to achieve that specific scenario. And the model builder will provide us the predictive insights with clicks just without any quote.

In model builder, we will be able to create a model from scratch as well as we will be able to bring the model from external systems. Let us see how we will be able to bring the model from external system. When we say from the scratch, Salesforce has, by default, some of the models that are available in its ecosystem. What in case? If we have some of the models that are available within other systems like AWS SageMaker or Google, Vertex, or anything. Salesforce data cloud has a concept called bring your own model by use using by which we will be able to bring our models into the data cloud. Here, we need to configure the model within the Einstein Studio, which will establish a connection with that specific model and utilizes the data that is available within the data cloud to generate the predictions. And the predictions can be used as part of the Salesforce, or we can send the data to the external systems as well. That is possible using a bring your own model concept.

Next, whatever prediction models which we are going to build, Salesforce suggests an Avocado framework. I know this is not an Avocado, but it looks like an egg. So Avocado framework has various components like dataset, segment, example set, and prediction set. Each and every component, we will be able to understand what exactly we need to do. For suppose, let us say we wanted to understand whether the customer is going to churn for the subscriptions which they have paid or purchased from the business. In order to understand or predict the churn for that specific customers, first, we need to identify the dataset. In this case, the object is customers. Similarly, our segment here is to achieve the whether the customers who has stayed with us for more than six months and going to churn. That’s it. That is that will be our requirement which we wanted to create a segment on. That will be the segment in this case. And each and every prediction model will require an example set, and these examples set can be in positive examples or negative examples. The positive examples can be the customer canceled subscription or the customers who maintain the subscription. Similarly, the prediction set will be the dataset which was not having the exact values and where we are going to predict on top of that data. Using the Avocado framework, we will be able to understand what we are trying to achieve and what exactly will be our desired output. And our kind of framework will be very helpful whenever we are using the prediction models.

Each and every prediction model will require the nuts and bolts for it to work. For every model, we’ll expect the input. The input will be variables and observations. The variables are nothing but, like, the fields which we usually create in the Salesforce. The observations are nothing but the records. These will be passed as an input to the model, and the output will be the prediction, whatever which we are trying to achieve. Either that can be a regression output or binary classification output.

Now let’s see it all in action. Let us assume we have a fictional company called Techron Group where the data related to the Techron Group company is residing at various sources, and they are having the information related to some of the ecommerce as well as the e bikes. Now we wanted to leverage the data cloud to bring the data into the data cloud and generate some predictions on top of it.

Let us quickly switch to the salesforceorg. Once we have logged into the salesforceorg, click on the app launcher and look for the data cloud app. In the data cloud app, we’ll be having various tabs. First, let us click on the gear icon and click on the data cloud setup. Here is the place where we will be configuring our connectors. Click on the other connectors and click on new. Here, you will be able to see the list of connectors that are available. Each and every connector will require certain authentication information that we need to provide in order to establish a connection with that specific date connector. Once we provide the authentication details, the connection will be established. For time being, I have already configured one connector with respect to to the AWS. Similarly, we will be able to create connectors for the Salesforce orgs as well.

And once our connector’s created, next, we need to switch back to the data cloud app and click on the data streams tab. Here, we have a new button. Click on the new. Whatever connectors were configured and authenticated, then we will be able to see the list of tiles that are available for them. As we have configured an Amazon s three connector, we are seeing the tile available. Click on the relevant connector which you wanted to fetch the data from. Click on next. If you are having multiple logs, you will be able to see the org from which you wanted to get the data into the data cloud. And post that, select the objects. And the object from which you wanted to get the data, click on next. Here, we will be able to see all the list of fields that are available within this specific object, and either we can select or remove the fields. Once we click on next, we will be able to select what exactly the category which we wanted to define that specific data stream, and we wanted to get the data into the data cloud. Once we have defined everything, let me click on deploy. Okay. Once a data stream is created, a relevant data lake object will be created. So let us go back to the data stream while it is running on.

So for time being, we have already see we have already configured the data streams relevant for the connectors, which we wanted to get the data into the data cloud. Whenever a data stream is created, a relevant data lake object will be created. In order to harmonize the data that was brought into the data cloud, we need to create mappings with data model objects. The left side one will be the data lake object and the right side one will be the data model object. The data model objects can be the standard data model objects or custom data model objects. Data Cloud by default offers the standard data model objects. If that doesn’t suffice the need, we can create our custom data model object. And once we create a mapping with them, the data will be harmonized.

Once we have the harmonized data, we can go to the Einstein Studio. This is place where we will be building our models. Now let us click on add predictive model. Here we have four options in order to choose either to create a model from scratch or get the model from external system, which is available like Amazon SalesMaker or Google Cloud Vertex AI or Databricks model. Let me select the create model from scratch. Click on next. And here, we need to select the dataset which we wanted to make the prediction on. Right now, we wanted to predict the quantity of the orders which the customers might place in future. Click on next. Once we click on next, either we can select all records or filtered set of records. I’m just going to use all records as I don’t want to add any conditional conditions for the dataset. Click on next. And here we need to select what was the field that we wanted to predict. I wanted to predict the order quantity and I wanted to maximize. Either we wanted to reduce or increase the prediction, we can choose the relevant field. Click on next. In the next screen, we need to select the variables. Here, we have two options. If you enable this, the Einstein will choose the fields, or you can disable this and select the relevant fields. Right now, I am going to use the Einstein selected fields. Similarly, it will be the same for the algorithm. Either we can select the relevant algorithm or let the Einstein choose. Currently, Salesforce supports three algorithms. GLM is the more kind of a simple and where we will be able to use it for the simple relationships and simple kind of data. GPM and XGBoost, we will be able to use them when we are having the complex data and we need addition trace. Once we click on the next button, we will be able to see what is the data and records and what are all the things which we have selected in order to create this model. One of the important consideration is that whenever we are selecting a dataset, make sure that the dataset has at least 400 records available in it in order to kick start or train the prediction model. You can click on the save. I’m just going to Give a name to the model. And once a name is provided to the model, it will take certain amount of time for the model to train.

In the meanwhile, let us go through a model that was created previously. So whenever a model is created and the training is done, we will be able to see the training metrics that are generated for that specific model. Each and every model that we have created will have the relevant metrics, and we will be able to understand whether that particular model is able to understand our dataset and generate the predictions appropriately or not based on the metrics that are available. For suppose, if it is of regression, we will be having the predicted versus actual value, which is nothing but what was the actual value and what is the predicted value. Similarly, r squares, uh, r square score, it will be from zero to one. The higher the value, the better is will be the model’s prediction capability, and we can understand the top predictors as well. For this, in order to understand these metrics, we need to have a statistical we need to revisit our statistical course again, but Salesforce provides some help text. Going over there, we will be able to understand what exactly is the metric for and which metric we need to consider. And once we have validated the model training quality, our next step is to create a job. Here, we will be able to create a job.

Speaker 0: Hi, Durga. As you’re going through the next step, I wanna give you a heads up that we have a little less than sixty seconds left. So if you’re able to, um, highlight the last step, and then, um, we’ll go into, uh, the next session. So go go ahead. Thank you.

Speaker 1: Yeah. So we can we can create a bat. We can create a job for whatever predictions which are generated in order to write it to the data model object. We will be able to see the output in the data explorer. Go to the datamodellog.com data model object and select the relevant data model object. Whenever a production job is run, the relevant data model object will be created. This is where we will be able to see the predictions that are available over here. Whatever data that we have predicted, we will be able to write it back to the Salesforce using the Salesforce flows, and as well as we can use the query API as well in order to use the information which we have generated using the prediction models. Specifically, this will help the businesses to drive the data driven decisions as well as improve conversions. And, also, we will be able to use the data, which is available as part of the data cloud in order to improve the business. That’s it, everyone.

Speaker 0: Thank you, Durga. We greatly appreciate it. Um, alright. That is all the time that we have today. Um, if you have any questions for Durga, please feel free to reach out to him directly under the messages tab in the event chat. And, um, we hope that you enjoy the rest of your Marjorie Mann conference. Thank you again.

Speaker 1: Thanks, everyone. Hope you have enjoyed the session.