AI in Action is brought to you by Fast Slow Motion. Our team helps growing businesses AI to work with practical, scalable solutions. To learn more about how we can help you implement AI in your business, visit Fast Slow Motion. Akamai.

Hey everybody. Welcome back to AI in Action. Eric Hirsch here, always joined by fast slow motion director of Zack Terry. And today we’re unpacking one of the most important parts of Salesforce’s AI ecosystem. And that’s Data Cloud or now known as Data 360. So, as Zach and I were joking before the show, if you hear us call it Data Cloud, if you hear us, call it Data 360, we mean the same things.

I forget the slip ups, but, think about this. If Agentforce is the brain of the enterprise Data 360 is the nervous system. It’s how information flows, connects, and gives your AI agents the full picture of your business. So before we dive into that though, and we’re just going to scratched the surface today guys. Just in full transparency.

There’s a lot to this product. But we’re hoping to give you the 10,000 level view. But before we do that, let’s look like we always do a few of the biggest developments shaping the AI landscape this week. Let’s start off with a big an open AI and Amazon $38 billion seven years. What can you tell us about that?

It wouldn’t be another week in AI with a lot of, without a lot of money changing hands, or at least allegedly changing it. Yeah, yeah, yeah, billions, billions and billions. So this is essentially one of the most significant infrastructure deals for open AI outside of their already existing partnership with Microsoft. And it actually led to a 4 to 5% spike in Amazon stock price.

When this announcement was made. But it’s ultimately diversifying OpenAI’s infrastructure, dependencies and positions. AWS is a frontier AI provider in AWS, and of course, has a bunch of these Nvidia CPUs and so, OpenAI is essentially going to be able to leverage a bunch of those CPUs. And that’s really just about additional scale, more redundancy and more flexibility across open AI’s entire compute landscape.

So there’s no winners and losers, but there if there is a loser in this, it’s probably Microsoft. And sure enough, you heard, shortly after this, Microsoft and OpenAI are kind of restructuring things a little bit. They are. Yeah. It’s pretty interesting to see what’s taking shape here. So Microsoft has a stake valued at roughly 135 billion.

And again, when we throw these numbers out I mean that’s like that’s what it is today. I mean, this stuff is just changing all the time. We’ve seen how, you know, the smallest amount of news in AI can cause the entire market to either lift or sink. I mean, it’s pretty wild how much control AI has kind of over the entire market right now.

I’m not going to comment on whether that’s a good or a bad thing, but we know it’s a reality. This restructuring between Microsoft and OpenAI. Microsoft is essentially losing its right of first refusal for compute. OpenAI is gaining more and multi-cloud freedom. And so I think a part of that is the ability for OpenAI to go to AWS and say, hey, let’s let’s strike a deal, let’s, let’s, you know, expand this beyond what we already have.

And so in return, OpenAI is committing to $250 billion in Azure spend, which, you know, again, it’s just it’s more numbers that were thrown out. There’s there’s nothing being signed here. A lot of this is just I think, frankly, a lot of this is just for news and for PR, because a lot of this stuff is happening just with press releases, but not with actual signed agreements.

And so, we’ll, we’ll see if this actually turns into an agreed upon $250 billion spend. I have no idea if that will actually pan out. But yeah, ultimately, we’re seeing a lot of change between these really big players in the space and the terms that they agreed upon. You know, a year ago may not be the terms that they’re working with today.

Definitely feels a little hyperbolic at times, for sure. Here’s a new one. Amazon and Perplexity kind of go into blows over a Gen X shopping. What can you tell us there? You gotta love a good AI feud these days. So yeah, Amazon sent a cease and desist letter to Perplexity over their Comet browsers shopping agent. So we talked about this a while back.

But Comet is perplexities agent to browser, so it has the ability for you to interact directly with Perplexity. You can actually have it go in, do web browsing for you. There’s a whole shopping agent where it can go and make purchases for you. A little scary. So the issue is that Comet actually acts on behalf of users without identifying itself as an AI agent, which is apparently a violation of Amazon’s Terms of Service.

So Perplexity is refusing to comply, not just refusing to comply, but publicly refusing to comply, arguing that an agent should have the same permissions as a user since it’s essentially acting on their behalf. So kind of interesting to see this squabble going on. I personally, I think that the terms of service that Amazon has makes a lot of sense, like we should have transparency in this ecosystem.

And so I don’t I don’t really understand what the big deal is. I don’t know why Perplexity is putting up, you know, so much, so much of this like being an issue. It just seems like, okay, that makes sense. If we just identify that this is an agent, this whole problem would be solved. So it could be an interesting test for how AI agents are allowed to interact with online platforms.

And then we have more than one agent browser now. I mean, we’ve we’ve got OpenAI browser Atlas as well. And so I wonder if there’s any kind of weirdness going on there. Maybe Atlas is adhering to these terms of service, or maybe it’s not even, you know, doing shopping for users yet, but definitely an interesting feud to watch to see where that goes.

Yeah, huge implications there for any company experimenting with AI driven commerce. Apple and Google are teaming up on Siri. That’s interesting. It is interesting. I’m excited about this as an Apple user, for everyone else out there who’s an Apple user, you’ve probably been maybe disappointed in how Siri can act. And I know when Apple launched Apple Intelligence a year and three months ago, maybe a little bit longer, I was back with the iPhone 15 was launched.

Right. And, the whole idea was Siri was going to have personal contacts and we were going to be able to weave these AI models into everything we were doing and just none of that has ever panned out. Right. And so now there’s there’s been a fascinating photo, Tim Cook and Senator Pichai in a cafe like talking to each other so you can tell their, you know, working out some deal.

But apparently, as reported by Bloomberg, Apple is agreeing to pay Google about $1 billion annually to use Gemini AI model to power a new version of Siri. $1 billion? That sounds like a big number. That is, that is like pocket change for it, honestly. Like it doesn’t sound like that much. I mean, yeah, it’s a really small amount.

So, now it’s apparently an ongoing agreement. So every year Apple will be ponying up $1 billion at least until this agreement ends. I imagine it’s a stopgap for Apple to be able to get performance out of Siri until they have their own in-house model that can essentially rival these other frontier models. But I think I think it’s exciting because I think that will make Siri better.

I’ll be able to use some features that I think Apple has been touting for a while, but it just hasn’t come about. So I’m hoping that this means we will actually get some of those features, and maybe we will get them sooner than we would otherwise get them. Otherwise we’d just be waiting for Apple to complete their model.

So, definitely more deals going on. And I it’s very expensive. Apple’s essentially renting talent from a rival, but they’ve done that before. I mean, Google Maps, they have Google Maps as default on iPhone for years and years and years. And so, it’s not unheard of for these types of deals to be struck between these tech giants.

Yeah, I think I speak for a lot of people out there that we’ve all been a little shocked at how slow Apple has been in the AI race. So I do think it’s interesting. This doesn’t really come as a surprise. They have to step up. So that’s, but you think about who their main competitor is in the smartphone space and it’s like, okay, well, what are the implications there exactly?

Finally, let’s talk about Google’s new, cybersecurity bill. So this is fascinating. So this is, it’s not prompt injection. We’ve talked a lot about prompt injection and the dangers that could face, but this is a brand new type of malware that they have called Prompt Flux. And it is essentially a strain of malware that uses large language models, including Gemini, to rewrite its own code mid execution.

So this is a really big deal because all of our anti-malware tools, they kind of rely on spotting patterns and understanding the code of a given malware. And if, if this is able to rewrite that code and obfuscate what’s going on over and over and over again, it’s going to make it extremely hard to spot. So this is fascinating.

I think this is an eventuality is not surprising that we’re seeing something like this. But I do think the fact that it’s been spotted in the wild, it does exist. We sort of know how it works. It’s pretty crazy to see exactly what’s going on. Right. Or this code is rewriting itself. The purpose of the code remains the same, but how it obfuscates the code is what changes.

So it’s essentially dynamically creating confusion around what that code actually does, which is essentially an entirely new front in cybersecurity. It’s a brand new threat. And so, really pointing out how AI is not just a tool for defense. It’s not just a tool for businesses and individuals, but it’s also an active threat actor. In AI itself, I should say, is not the threat.

It’s a tool. But the use of that tool for nefarious purposes is essentially what we’re seeing with this. Yeah. Use the word fascinating. I think I would use the word terrifying code that rewrites that dream. Sounds like something out of a James Cameron movie, but we’ll we’ll we’ll move on so that it could be both.

So, AI arms race heating. For most companies, the real question is how do you make your own internal company data work for AI? And that’s where Salesforce data Cloud Data 3060 comes in. But probably a good deal of our listeners may not even know what Salesforce Data 360 is. As I spent a few minutes just kind of bringing us up to speed, it is a complicated product and it’s a big product.

And so you mentioned this at the beginning, but we’ll essentially just scratch the surface of the capabilities today. But one sentence description Data Cloud pulls customer information from all of your different systems into a single place so that you have a complete picture of each customer that updates in real time and can be used across all of your tools.

So that’s a lot of words, but essentially it is. It is. I’m going to use the word 360 because that’s all in Salesforce is branding and marketing now. But they’ve been saying this for years, since the inception of what is now called Data 360, but has gone through several name changes. It gives you a 360 degree view of your customer.

That’s the whole point of it, right? Take all of these disparate data sources. I have information about a customer in my ERP. I have it in my CRM, I have it in Shopify, you know, wherever it could be, and bring all that data together into a central location so that you can understand the full context of what’s going.

So if we look at a timeline of this product, this is actually a homegrown product from Salesforce. It’s not an acquisition. They launched it in 2019. It was called customer 360 audiences. And this was Salesforce’s first attempt at CDP. CDP is customer data platform. This is essentially a space in tech. Operating this idea of, okay, I want to see all my customer data in one place, right?

I want to unify this fragmented customer data across these systems, get better audience segmentation. So there’s kind of a marketing focus. In 2021, it went from being called Salesforce CDP to Marketing Cloud CDP. And they rebranded that because they were really focusing on the marketing use cases here. Because if you I’m creating segments of a bunch of different customers from across a bunch of different systems, that is, of course a very good use case, right?

I want to be able to market with the most up to date information of everything that I have. But then they started to kind of see the benefits that this could have outside of marketing. And in 22 they rebranded to Salesforce Genie, which is really funny because this is about the shortest name that I think we’ve ever seen, the shortest lifespan for a name.

It lasted less than a year, right? But yeah, they were shifting essentially from a marketing only tool to a more general purpose tool supporting sales, service, sort of everything in Salesforce. And then in 23, it was officially rebranded to Data Cloud. They dropped the Genie name. They started to weave in some generative AI capabilities as of last year with the launch of Agentforce, and it became Salesforce’s fastest growing, organically built product.

And then as of October of 25, it’s now called Data 360. So again, kind of aligning with the whole brand around the 360 view of the customer. It was also renamed as a part of the Agentforce rebranding, which is now called Agentforce 360. And it’s shifting from data management to data intelligence. And so the idea is this is sort of the central hub that can feed AI agents data.

It can assist with workflows, it can make real time decisions instead of just storing the information. And so bottom line is it started as mostly a marketing tool. It has evolved into the data foundation powering Salesforce’s entire AI and automation strategy. And it’s an essential infrastructure for Agentforce in how it operates on the Salesforce platform. Let’s go a little deeper.

Let’s talk about some key concepts people need to know about Data 360. And let’s start out with context engineering. What are we talking about when we say that. So if you think about what we described, which is a 360 degree view of the customer, the whole purpose of that all encompassing view is to understand the full context. Right?

And that’s the same idea with context engineering. So when we started working with AI systems and specifically generative AI systems, there was a whole conversation around, hey, the next role in tech is going to be prompt engineer. This is going to be like the next big thing. And then it only took about a year for that to become.

Well, I don’t think that’s really going to be the role anymore. So prompt engineer kind of went away. And instead we have context engineering. So prompt engineering was really about making sure you were wording an individual prompt in the right way to get the results that you want. Well, now these models are much better at processing this. And so a specific prompt containing the ranked right context is less important then the AI just having native access to that context.

And so context engineering is essentially the practice of designing and optimizing your environment, the environment that your AI has access to with the information that the system can use to operate within. So it system prompts knowledge bases, access to tools, guidelines around when to use those tools and how to use those tools. And so it’s essentially equipping your AI agents with a complete, connected view of every customer and every process and giving it the right context for, you know, each use case and each process that is going to be working with.

So Data Cloud really comes into play here because it has the ability to gain all of that context from all the systems and store it in a central location, but it can also unify that data. And we’ll talk about that a little bit later. But, Data Cloud is able to take all those disparate data sources and essentially connect it together so that, if I’ve got a bunch of different foreign keys and different databases, I can use something common to a customer and join that all together so that I have a single view.

And then Agentforce is able to leverage Data Cloud. In addition to Salesforce. So kind of a confusing concept, but Data Cloud and Salesforce are two separate products and they work really well together. And so Agentforce has access to both your structured CRM data as well as your structured and unstructured data, cloud data, which gives it a ton of context if you’ve set it up correctly, and if you brought in a bunch of information from other systems.

So when it really comes down to agents being able to see everything that matters, not just isolated records in the CRM, but it’s also really important that the data is clean. And so that way you’re bringing in relevant information so that you can drive quality output with the AI. Super powerful. When you think about it, a customer reaching out the system, already knowing everything about them, history, preferences, previous conversations, current situations, humans aren’t going, aren’t trying to hunt around for data and put a picture together.

The system’s doing it for them. Let’s move on. That was great. Semantic search and vector database is this is teaching AI to understand. This is where things get really powerful. This I think is a huge value driver. If you are looking to implement Agentforce and you’re looking to take advantage of unstructured data and the ability to understand the meaning of a question and apply that to the information that you have in the system.

And when we say semantic, so semantic refers to the meaning and intent of words rather than just the literal meaning of the word itself. So for example, if we’re talking about cars, automobiles, vehicles, trucks, these are all distinct words with distinct meanings, but categorically in certain contexts they can more or less mean the same thing. And so that’s the idea behind semantics.

So if an AI agent gets a question about, let’s say revenue growth and you have a bunch of information in your database, but it’s not necessarily the exact words revenue growth, it can look for information about, let’s say, increasing sales. And it has the same sort of semantic meaning, even though it’s completely different words. So to power this you have to have a set of tools, a vector database being one of the primary tools that allows you to store information in what’s called embeddings.

And so a vector database stores numerical representations. That’s that’s the embedding. The numerical representation of information words, tokens, phrases meaning. So the AI systems can really quickly find and use the most relevant data when trying to answer questions or generating content. This is essentially one of the core tenets around how these systems actually work, because in order for a large language model to even be able to understand the meaning behind what I am asking it or telling it to do, and then go out and actually find the right information in trillions of data points to come back with something that is actually meaningful and relevant.

It’s it’s a lot of math, essentially, but it’s comparing. The way I kind of think about it is you have addresses for words or meaning, and I ask a question and that question has a bunch of different addresses around where the words are and what they mean. And then we’re going to go out and find the closest address inside of the database.

And it’s doing that a bunch of times. Right. It’s not it’s not just as simple as like, well, this word lives here and this word lives here, but that’s that’s sort of the best analogy that I can think of. It’s very complicated. It’s like it’s more than a three dimensional space, but the whole idea is we’re understanding the meaning of words as opposed to just the literal context of that word.

And that allows AI to go and search through these databases and pull back relevant content. So Data Cloud has a vector database, and they have actually made it very easy to use. And you can even on Salesforce foundations without even necessarily purchasing a lot of Data Cloud credits. You can have access to the Data cloud vector database. You can upload your unstructured data like PDF documents or call transcripts.

It could be knowledge articles from your knowledge base, whatever that is. And that will all get turned into embeddings on the Data Cloud vector database, so that you can then connect that to your agent for agents. And then it’s able to go and actually retrieve that information. And this is also where we get into retrieval augmented generation or Rag, where you’re essentially injecting the context at the time that the prompt is run.

And so instead of me having to explicitly list out all of that context, I can take a query, understand the meaning of that query, and then I can have the retriever go and actually find the relevant information dynamically. And essentially it saves a lot of time and it makes it more accurate because I’m not having to front load the prompt with all of the information that I need, I’m able to go and dynamically retrieve just the relevant information at that point in time, and it’s grounded just on the data that you’ve provided it.

And so it will provide more accurate answers as opposed to giving it every possible piece of information up front, in which case it may find more irrelevant results. So this to me is one of the most powerful. I use cases for Data Cloud because this vector database comes out of the box with Data Cloud. Yes, you pay for it.

There are credits that are consumed when you store files in the vector database and when you query the vector database, but overall, it’s much more affordable than trying to build this entire infrastructure yourself. And this is one of the most complex parts about integrating AI into a business is if you do want to use semantic search, if you don’t have a tool like Data Cloud, or you’re not investing in some other tools that integrate all that together, you’re looking at a pretty big engineering challenge to build all of those capabilities yourself.

It’s it’s that evolution, right from chat bots that are just guessing to agents that really have a deep understanding of your business. Exactly. Cool. Let’s talk relationship mapping. This is the agent’s internal neural network. Tell us about that. So if you think about a Salesforce record like a contact, and that contact has a bunch of information about the contact itself, you know, maybe you’ve got contact information, a phone number, email address, first name, last name, things like that.

But then we also have relational information. So that contact probably belongs to an account in our system. That’s the business that they work for, the business that they own, you know whatever that relationship is. And so we’ve got this relational database that says this contact belongs to this account. And then that account could have related deals, and then those deals could have related products, and then those products could be associated with quotes and with orders and things like that.

Right. So we have all this relation relational mapping that goes on inside the database. So Data Cloud kind of takes that idea and says okay well I know I’ve got my CRM with all of its data and all of the relational information there, but how do I take something from an external system and understand that this particular order from this other system actually belongs to this customer?

And so when I look at that customer, I can see those orders. And that’s where this sort of relationship mapping capability comes in. And so it allows us to perform essentially some transforms on the data that comes in from these other systems. And it allows us to create those relationships in a way that gives us visibility into how all of those things are connected.

And so if I’ve got contacts to accounts, to orders or products, to cases, to knowledge, we can kind of tie all that together, even if they’re coming from a bunch of different systems. And it’s a really short way of describing it. It’s pretty complicated to actually go out and do it, but Data Club gives you all the tools that you need in order to connect it together.

So you’re essentially just identifying what are the commonalities that I can use between these systems to tie everything together, sometimes as primary keys and other systems. Sometimes it’s something like an email address that I’m going to have in all my systems. Or it could be a combination of things. Maybe I have an email address in two out of my five systems, and then the other three systems.

I have a phone number and I can use a combination of those two things to say, okay, the email address and the phone number both belong to this one customer. So let’s stitched that all together so that I can kind of see everything and understand that relationship. That’s really where this whole relationship mapping comes into play. Because without those relationships, we don’t really understand the contextual meaning beyond just I have an order that belongs to some customer.

It really makes sense when I have a customer that has multiple orders. And let’s say that they’re calling in to an agent for a voice service agent, or they’re chatting in and that agent is able to understand, hey, Zach Scott, ten previous orders with us. One of them is still outstanding. So he’s probably asking about this particular order that hasn’t yet been fulfilled.

So that’s really bringing all that context in. And so instead of me having to tell that a, hey, this is my order number and like it’s hasn’t been delivered yet and blah, blah, blah, it can essentially get all that context, even if it’s in another system, because Data Cloud has got it all stored together and it’s all tied together with that relationship.

And it gives that AI agent. And it can be a human agent too, right? You just think about it as even if I’m a human being looking at this, it’s giving me the ability to understand what’s going on without having to go and make a bunch of additional queries or a bunch of additional clicks. But in this, this use case of AI, it allows that AI to be a little bit more proactive instead of reactive, because it’s going to have the right context.

And so if we give the agent the right instructions to understand, hey, whenever you have a customer that has an open order and they’re asking about an order, they’re probably asking about that open order. So go ahead and retrieve that information and provide that to the customer. And that can, you know, take a whole lot of other, it can essentially be a lot of other ways to do things like that.

It’s not just orders, but, ultimately it’s about maintaining those relationships, maintaining that context, and being able to proactively understand everything that’s going on without having to reach out manually to different systems. And I think that’s going to be key to the adoption of agents and customer service. Is that transition from transactional to really intelligent to when I dial in, when I engage with these agents, it really feels like they have an understanding of who I am.

I’m not having to answer a bunch of voice prompts and things like that. So you can really see how that can be super powerful for an enterprise. Yeah, absolutely. So now we’ve got the data together. We’ve got the relationships, Matt. We’ve got the context engineering. Once we’ve got the foundation, we can start surfacing real insight, turning that data into real decisions.

Yeah. And this comes down to a feature called Calculated Insights in Data Cloud. And this is essentially derived metrics and aggregations that Data Cloud can compute from connected sources. And so when you connect something in Data Cloud it typically goes into something called a data model object. There’s also data lake objects. And then there’s a bunch of unstructured information that you can gather as well.

But if I’m taking all this data and putting it into a structured source, then we can actually go out and perform calculations on that data. And so if we had that same use case, a customer with orders and order history, we could calculate things like lifetime value or churn risk based on similar customers and similar order patterns. We can look at engagement scores.

If we’re talking about marketing, we could even bring in a bunch of marketing data and unify that data so that we understand that all that engagement that’s been going on is actually specific to a given customer, and then we can use calculated insights to go and actually create, results from that data. And so it’s not just all right, Zach visited 100 web pages this past month, but it could be a calculation around, hey, do we draw some correlation between the frequency of website visits and the lifetime value of the customer?

And then we could create a score and then Data Cloud could dynamically actually calculate that so that we can use that. And then I can use that as well. And so agents can act using insights. So not just the data about the customer but also the insights that have been calculated about the customer so that it can essentially have much more valuable context around it.

And it’s not having to actually perform those calculations itself, because we know that AI is not necessarily great at math unless it’s calling another tool to actually perform that math. So that’s kind of what we’re doing. We’re calculating it and then giving the AI access to those calculations so that it can interpret the results and answer those questions for us.

So it really is turning that data from just raw data into actual action and analysis that we can use to understand more about a customer or about a given scenario. Let’s talk identity resolution one customer, one truth. This might be the single most practical capability inside a data. Yes. So if we go back and talk about relationship mapping, right.

And I’m going to be bringing in a bunch of data from other systems. Well, I’ve got multiple basically copies of customers in all these other systems. Right. And so if I just kind of bring that data in, let’s, let’s look at it from an integration perspective. So if I was going to integrate all that data into Salesforce from multiple systems.

And so I got five different versions of a customer, I’ve got my Salesforce version, they’ve got my ERP version, I’ve got any of my sort of sales, my e-commerce, whatever that is. Those are all different versions of, of the same customer. So if I were to integrate all of that together, I have got to do a lot of deduplication, understanding that this is the same person.

And so when I bring this data over, I’m actually going to be updating a record. And instead of creating a new record and things like that. So it’s the same idea, except we’re not actually creating records, we are just tying a bunch of data together and so identity resolution allows us to use rules in the background to say, okay, if a customer has the same email address and phone number or a combination thereof, it’s the same customer.

So let’s merge that into a single identity so that we are not creating multiple disparate data sources or multiple copies of the same customer. And that applies not just to the customer itself, but also to all that data about the customer. So let’s say I am tracking orders in Salesforce, and I have another system that’s bringing those orders in as well.

Well, we can use this resolution to prevent those duplicates, but it really is mostly about the person making sure that we’re looking at the same person and understanding everything going on with all the activities that person is engaging with. So it’s combining data across sales, service, marketing, external systems, and really gives you one source of truth for a customer as opposed to having to resolve multiple identities.

It’s doing that resolution for you, Zach, we’ve talked about a lot today. We’re coming up short on time. Let’s hit on the final point here. Real time data. Why is this important and what do we need to know about. So this is a feature of Data Cloud. That and it’s not just real time. It’s it’s subsecond. So if you are using real time data streams with data Cloud, it is extremely fast.

And a lot of that has to do with the zero copy data model, which means we’re not actually copying data from these other sources in Salesforce. We’re just visualizing it. And we can act on it and we can understand what’s going on, but we’re not replicating it. Right. That allows it to be very fast and low latency. But this essentially allows any update to come directly into Data Cloud very quickly.

So that way, whether it’s a human or an AI agent, we know that we’re always interacting with the most up to date version of a customer. So let’s say that I am on a phone, a phone call with a customer, and a minute ago they actually made an order. And it’s unrelated to the call that they’ve actually called in about what’s important to have that context.

And if I didn’t have real time information, I might not see that until the next day. Right. And so to understand all that information immediately, that enables us to do much more, and especially in the context of having an AI do this for you, if they have that full context, they’ll be able to answer questions much more accurately.

It gives you fresh data, which leads to more credible AI, right? Because it understands exactly what’s been going on. But one important note here is you don’t have to do real time, because the real time data streaming is pretty expensive compared to the alternative, which would be batch data ingestion. And so when I’m connecting data sources to Data Cloud, I can choose how I want to connect.

And if it is a data source that supports real time streaming, that’s that’s called streaming ingestion. And that means that I can get updates immediately as they happen just constantly. Right. The implication there is that it’s more expensive. So you’re going to consume more credits if you do it. That way. And so it’s really important to think about when you’re designing a Data Cloud instance, what requires real time versus what suffices for batch.

So can I do this once a day. Can I do this once an hour? Can I do this once a week? It really kind of depends on what that data is. So be thinking about how critical is it that we get this data in real time versus can this be something that is just regularly updated at a at a certain interval, right.

And there’s a good chance you will save money and credit consumption by using batch ingestion. Over using streaming ingestion, you only want to use that streaming capability for scenarios that actually require real time data from other system. Zach, we always like to leave our listeners with something actionable they can do this week. What’s your advice? What can listeners do right now to dig into this a little further?

Well, I’ve said this before and I’ll say it again, but look into Salesforce Foundations because if you’re a customer that’s not on Data Cloud, you’re not on agent Force foundations is a great way to get an introduction into both of those tools. So foundations includes free credits for both Data Cloud and for Agentforce. And so if you wanted to kind of test the waters and get started with everything we’ve been talking about today, foundations can do that at no cost to you.

If you are on an Enterprise Edition org and above. So that’s my first recommendation. And then let’s say you’re you’re thinking about implementing Data Cloud, or you’re thinking about how to layer AI into maybe an existing Data Cloud implementation. Audit your data sources. So understand which data sources and what data from those sources can add important context for an AI agent.

Identify the unstructured data sources that you might have in your business knowledge base PDF documents. Or maybe it’s something that’s living in somebody’s head and they haven’t documented it yet. Think about that type of information that could be leveraged by agents and a vector database on Data Cloud to enable semantic search. So think about that and then think about mapping key relationships across your systems.

So do you have multiple systems. And do you have a single view of the customer. And if the answer is no then Data Cloud could be a great opportunity to tie all of that together so that you do have a single view of your customer. And then finally, prioritize quality over quantity. And so this goes back to the garbage in, garbage out axiom of AI is really only as good as the data that you have.

This also goes into Data Cloud, which is only ingest data that adds value. So if you have a bunch of data, but only 10% of that data actually brings value to your I use cases or to whatever you have in your CRM. Only ingest that data. Don’t worry about data that’s irrelevant or data that’s not going to bring value.

Focus on the data that is truly needed and that will that will give you two benefits. One, you’ll get better results from your AI because it’s only using relevant data. It doesn’t have to work as hard to go and find the right information out of a sea of irrelevant data. You’re already working with relevant data, and then b you will save money because you will consume less from an ingestion standpoint.

When you’re bringing that into Data Cloud, focus just on the data that adds value. Wrapping it up, Zach Salesforce Data 360 is that plumbing that powers the agentic enterprise? It’s how Salesforce turns disconnected systems into a single intelligent network that your AI can actually use. So, if you want to start small, focus on your foundation. Clean data, connecting systems, meaningful context.

That’s what fuels everything else. Exactly. You want to start with those essentials and keep in mind that AI essentially is only going to work as well as the data that it’s connected to. We know a lot of this can sound complex. It can sound overwhelming at times. That’s why we’re here, guys. Fast, slow motion. If you want help understanding how Data 360 can impact your business, mapping out how it fits into your business, visit us fast slow motion.com/i would love to have a conversation with you.

We’ll see you next time on AI in Action. Thanks, Zach. Thanks.

Yeah.