Salesforce, Inc. (CRM) Earnings Call Transcript & Summary

Good morning, and thank you all for joining us. I'm Valmik Desai. This session marks the fourth in our series of quarterly post-earnings webinars aimed at providing you all with a deep dive on our latest product innovations and strategy. Today, we will deep dive on our Agentic enterprise architecture evolution and innovation. As you heard earlier this week on our earnings call, our 4-system architecture of engagement, agency, work and context it's foundational to how we are helping customers become agentic enterprises. Starting with some housekeeping. Some of our comments today may contain forward-looking statements that are subject to risks, uncertainties and assumptions which could change. Should any of these risks materialize or should our assumptions prove to be incorrect, actual company results or outcomes could differ materially from these forward-looking statements. A description of these risks, uncertainties and assumptions and other factors that could affect our financial results or outcomes is included in our SEC filings, including our most recent report on Forms 10-K, 10-Q and any other SEC filings. Except as required by law, we do not undertake any responsibility to update these forward-looking statements. Today, I'm really excited to host Muralidhar Krishnaprasad or MK, our President and Chief CTO of Product Engineering and Madhav Thattai, our Executive Vice President and GM of Agentforce. We're going to start with a brief presentation and a demo, and then we're going to jump straight into your questions. And I know this is not a shy group, but please do submit your questions in the chat. With that, I'll hand it over to you, MK.

All right. Thank you, Valmik. As you all know, first of all, good morning, and thank you for joining us. As you know, all know, every company wants to become an agentic enterprise. And for us, the definition of an agentic enterprise is where humans and agents drive customer success together. So that you can get better productivity, higher revenue and, of course, more efficiency in your operations. Next slide. But I think the biggest mistake people do is that they just think simply, all it is, is you just need an LLM to do the work. Because raw intelligence is not enterprise work. On the left side, you see the frontier models all ready to go tackle the complex intelligent task. On the right side, you have the enterprise outcomes. Unfortunately, 95% of all these enterprise AI pilots fail because they're not crossing the chasm. Because these LLM can't act their own. They are non-deterministic. They don't -- like you can't rely on their outcomes all the time, and they lack business context. And so next slide, this is really why the agentic enterprise needs more than models. And this is not just us just pontificating but really through our experience over the last several years making so many of our customers successful becoming an agentic enterprise. And the 4 things that we really need, starting from the bottom, is a system of context, which can tell you exactly what your data is, what your customers are, what are they doing. So that you have the business context around the operations you want to do. The next is the system of work where the work actually gets done, whether you're servicing a customer, you're selling to a customer, you're marketing and so on. And then you have a system of agency, which is actually doing that, taking that raw intelligence and being able to orchestrate across the system of work using the context appropriately. And finally, the system of engagement where you're actually talking to that customer on the right channel, whether it's for employees or for customers. And we believe these 4 is really what takes that crosses the chasm from intelligence to business outcomes through an AI-driven model. Next slide. Here, in Salesforce, our 4 layers of context, work, agency and engagement, we believe we have the industry's only unified stack that can make this possible. Starting from the bottom, of course, we can run on any models, whether it's OpenAI, Anthropic, Gemini, Llama, et cetera. Starting from the bottom, our Data 360 provides the foundational system of context. You can bring all your structured data, unstructured data, in many cases, just 0 copy, you can leave the data in their warehouses, but pull them together. And really create that singular customer profile, product profile and others, which can help you give that context of what the users or accounts is doing across your business. And then we have above that is a system of work. We have the most industry's comprehensive thing related everywhere from marketing, sales, service, operations, analytics and so on. And this is where work gets done. This is where all your business workflows are created as well. And as you know, we recently launched our CCaaS and ITSM product line as well to join this. And then you have the system of agency. This is where the Agentforce customer and employee agents are built. This is the 1 that is taking the context and then bringing all of the business workflows that are there in Customer 360 together using the power of the LLM models. Finally, the system of engagement is where we have both our channels with all the channels that we support from WhatsApp, SMS, text, e-mail and others, but more premium channel is with Slack for our employee experiences particularly with Slackbot and our ability to do enterprise search and really bringing in all of the power of the visualizations also into the system of engagement. So this is how comprehensive our stack is, and they all work together in unison. Next slide. Now I'll just briefly deep dive with 2 slides on the context itself and hand over to Madhav. So to do that context, you really need data and this is really where Data 360 comes into play. You can connect to hundreds and hundreds of different sources that's in your enterprise very easily. You can turn any unstructured data, your call transcripts, your nodes, your design documents, whatever they may be into both structured data as well as vectorize them so that agents can actually understand and reason over them. And you also get memory, all these agents get both short-term and long-term memory. So that when you come back after 3 months to talk to your business, it can actually quickly pull back what you have done. And you also want to bring that quality and clarity because data quality becomes important. You're putting the reputation of your business on the line by putting the agents in the front, and this is where the data quality and cleansing comes into play. Next slide. And so putting this together, the way context works as you see on the left side, you have all of this data and all of these applications sitting there with the power of the Data 360, the MuleSoft, Informatica and Tableau, you can actually create those MDM records, the unified profiles, create the context engineering around it where you tie all of this data with the raw data, with the context about what they're doing with the agents and be able to go analyze it and then link it back into the system of work. So this is how the context layer at the bottom ties in with all your data stores in an enterprise with the system of work that we have across and then feed it to the agentic layer. Madhav, over to you.

Thank you so much, MK, I'm glad to be with all of you today. So MK talked about the system of context. I want to now touch on the system of agency and what we built with Agentforce. I think you all saw this in our earnings, we talked about the performance of Agentforce in the last 15 months. We're now at an $800 million run rate up pretty significantly year-over-year. And we went from about 3,000 customers to now over 23,000 customers across those 29,000 deals that you see. It's been incredibly encouraging to see customers from all over the world, in different industries, using Agentforce for many different use cases, some of which we're going to talk to you about today. But a couple of the things that are really, really important as we think about this. First of all, we see Agentforce uplifting our license products where customers are now using them for premium experiences for employees with our Agentforce for Sales, Agentforce for Service, Agentforce 1 Editions products. The customer is getting a lot of value on those employee use cases. And then, of course, there's the consumption business where these are fully autonomous agents that are facing customers. You've obviously seen the great examples with Williams-Sonoma and others, where we have these agents that are doing kind of end-to-end life cycle work for our customers. And then also really encouraging to see that some of those customers now are really at a deep stage of maturity. And so they are coming back for more credits and more expansive use cases as they start to drive this across their life cycle. So incredibly exciting performance for the product, and we're really just getting started. Let's go to the next slide. Now I want to spend a minute on this. We introduced this new metric this week, and I know I've talked to some of you already about what it means. But MK made what I think is the most profound and important statement, which is, it is not sufficient to just measure intelligence. And we've been talking about intelligence as measured by tokens for a couple of years now ever since this capability launched. And it's really important to understand what's happening at the infrastructure layer, just like when we talk about the cloud, we want to understand what's happening at the compute layer, at the storage layer, at the networking layer, and we really see tokens as a part of that infrastructure layer. But what we care about and what our customers care about is turning that intelligence into work. So we introduced this new metric agentic work unit, which is really comprising all of the work that is being done on our platform with these agentic systems. Now that could be a decision made by an agent to respond or make a decision and reason through a particular task. It could be the task of actually performing a record update or triggering a workflow or maybe an API in a third-party system, maybe running a piece of code. And so we now systematically measure everything that's happening in Agentforce, in Slack, in MuleSoft and we say, okay, let's really understand how are customers getting real work done because that real work is what really tells us that the utility of the product is valuable to the customer. The more they use the product, the more we know it's having an impact on their business. And so it's early. We're trying to understand a lot about this metric. We can now look at this metric by product line. We can look at it by customer. We can understand which use case they're using it for. So really, really exciting. And we think that the moment has come to move from a simple measure of intelligence, which is an input into what is the output and the actual work getting done. So if you go to the next slide, and here are some examples. These are some of our incredible customers. We talked about some of these at earnings as well that are really doing a lot of work at scale. And you can see the variety of different use cases, and that's really what jumps out on this slide. We have customers like Agibank recently went public in Latin America. They are using this agent externally facing their customers. So every time a customer comes back to ask about loan status to understand what their next steps are, they are directly interacting with this agent. A great customer like Bouygues in Europe, they have their agent that they call Iris, which is helping their employees. And it's helping their employees with fairly complex tasks across all of the things that their employees need to do to really drive that internal productivity. ADP is a great customer. They're actually using an agent internal to their company that's helping them with HR workflows. So this isn't even in just sales and marketing and so on. This is actually expanding what they think Agentforce can be used for. And then, of course, General Motors is using task-based agentic automation. So this is -- I have very specific tasks that I need to go accomplish. These tasks are in the flow of work. They are where my employees are working in Salesforce and Slack and that automation is driving a lot of productivity. So that's really been encouraging because we see customers that are experimenting with a variety of use cases, pure employee productivity. Employee assistance with more complex agents. And then, of course, really the important use cases, which is we now feel confident enough in this agentic technology that we're going to have our customers interact with this agent. So that's been really, really encouraging to see and a lot more to come with all of these customers that are really starting to add more use cases. Let's go to the next slide. But this is really the bottom line here. There has been a tremendous amount of energy around building. We can Vibe code, we can accelerate the deployment of software, and we are significant users on MK's team and across all of our engineering teams, where we believe that this technology is accelerating our ability to build, and that is remarkable. However, there isn't an enterprise in the world that wants to Vibe operate. The build is the first part of this journey, and we have really worked over the last couple of decades with our customers to help them run their business. Now to run your business, after you build, you need to understand how do you want to test the capability? Is it working? How do you evaluate it? How do you observe it? Is it driving your KPIs? Is it driving your outcomes? How do we continuously optimize. And that is where the fact that we are so deeply embedded with our customers, the fact that their most valuable work, which is how they interact with their own customers, happens on Salesforce, really gives us a deep understanding of how we take the best of this intelligence technology, turn it into remarkable experiences, but then importantly, help businesses operate and manage that capability. But I'm going to stop there. Let's go to the next slide because rather than just talk theory, we actually want to just take a few minutes and just show you how the product works end-to-end. And for that, we're going to bring in the incredible Gabe Sumner to walk us through a quick demo of what this experience actually looks like.

Yes. Thank you so much, Madhav. So what are we looking at here? What we're looking at is the type of agent that we are helping our customers create, which is essentially an agent that represents the brand. But as Madhav and MK just talked about, for it to do that reliably, it's got to be grounded in those 4 systems: engagement, agency, work and context. And let's just start with context. So if I just give this agent an instruction, I need to reschedule my flight. We know that to do this job reliably, to answer that question, the agent has got to be grounded in that enterprise context that Madhav just -- or MK talked about. It's got to know who the customer is. It's got to know what their flight was. It has to know what an appropriate replacement flight might be. But customers also want to take these agents a step further. We want the agents to take action on the brand's behalf. And that means you need to give the agent some level of agency to respond flexibly across a whole spectrum of requests that the agent might get while always adhering to your business rules. Now if we're going to transform all those customer experiences with agents, we need these agents to engage across all the different channels where customers are, which is going to include mobile, but extends all the way out to voice and your 800 number with those CCaaS systems that MK talked about earlier. And if the agent needs to escalate to a human, it means the agents have to work where your employees work, which means supporting seamless handoffs between the agent to employees inside the work systems where they are using to work and collaborate. And as MK and Madhav just talked about, we are the only agentic platform that brings together these 4 systems. And let's take a little bit of a deeper dive to explore exactly how that happens. So this is Agentforce Builder. This is where our customers define their agents and agents are defined with topics. You can think of these as essentially the jobs the agent is allowed to do. And behind each of these topics is essentially a little subagent that is comprised of both instructions and actions. And if we were to go into instructions, like these are the guardrails that you create that guide the agent's behavior. And because we've introduced this new scripting language, Agentforce Script, we are able to blend this kind of like deterministic logic, your business rules with natural language kind of instructions. And that is what gives Agentforce this kind of the agency to adapt while always kind of adhering to your business rules. Now there's -- also these agents are comprised of actions. And you can kind of think of these as the tools that we've equipped Agentforce with to actually do all of this work. And what would that work be? Well, it can be things like retrieving data from Salesforce or outside Salesforce with Data Cloud. It can also be executing your business workflows. It could also be even using those MCP tools from kind of the open ecosystem that we participate in...

Let's pause here for a second.

Yes, sure.

This is really important. What Gabe is showing you is how the agent gets built. And when you think about how the agent gets built, he said some really, really critical things, and these are kind of the core and most important innovations that we brought in. Number one, tying in that context directly so that the agent is leveraging all of that enterprise data that is connected into Salesforce. You don't have to build extra data connections. You don't have to move data around, it is seamlessly integrated into that data. So that was one point that he made, that is really, really critical. The second thing that's really important is one of our biggest learnings in the agentic enterprise has been this need for deterministic control. These aren't just agents that are answering simple questions. These are agents that are executing on work. And so how do you make sure you are tying that deterministic ability into the agent build is really, really critical. But really importantly, those business workflows sit in Salesforce today. And MK brought this up before when he called it the system of work. Tying it back into the work and the process that is happening is what allows customers to not have to recreate process in order to roll out these agents. And that's a really important way in which these agentic deployments can go a lot faster and a lot easier. Go ahead, Gabe.

Yes. Absolutely correct. And then as we move forward, we've talked about all the different engagement channels. So this would be where you extend Agentforce to all those channels where your employees and customers are. And then if we go down to here to the very bottom, and I'm sorry, Zoom is in my way here. But you have data, which allows it to be grounded in that unstructured data like those knowledge articles. But honestly, the way to kind of like see all of this come together, I find it just like go test the agent and see how Agentforce actually goes and does a job. So this is the simulator inside of Agentforce Builder that we can use to kind of see how the agents work or how Agentforce works. And I'm just going to ask you to find a flight from Seattle to JFK. And so like what is happening right now is Agentforce is kind of using that agency that we granted it to flexibly understand the intent to make a plan to ground itself in all that right enterprise context, to execute the actions and adherence with our business rules until finally, it surfaces the right answer. And not just like surfaces the right answer, but explains the reasoning that it used to produce that answer. And we've given our customers the tools to really dig deep into all of this and just really look at all the individual details that were used to create this reasoning. And Madhav talked a little bit about the Vibe coding experience, like we're trying to bring that Vibe coding experience into the enterprise. So if you notice problems with any of this execution, you can just ask Agentforce to fix it. And instead of like it generating a bunch of ad hoc code that you have to then maintain yourself, it's generating kind of metadata that sits on top of an enterprise-grade platform. And that's how we get -- help our customers get their agents ready for deployment. But of course, 1 test, 1 test, never enough for these nondeterministic solutions, which is why we also have testing center, which uses AI to generate hundreds, if not thousands, of AI-generated test scenarios. So that we can just test a whole spectrum of things that we -- different ways that customers might ask the agent to do different things to make sure that the agent is performing as you expect, across all of those different simulations. And then long after deployment, we're giving our customers the tools to understand how their agents are doing against the backdrop of all their business KPIs. So they understand the value that agents are bringing to their business. And we're also not only helping them understand how agents are doing, but also what the agents are doing. So you can drill down into this, get visibility across different topics and even highlight things that the agent could be doing better. And zoom all the way down into individual interactions to really see how an agent is doing an individual job, look at everything that's being done to do that job, interrogate it, troubleshoot it and that becomes that kind of flywheel that just makes the agent better and better and better. So this is how we're bringing together those 4 systems at MK and Madhav have just talked about, to convert that raw intelligence of these frontier models in the trusted enterprise work. Madhav, I'll give it back to you.

Thank you so much, Gabe. Really appreciate it. So let's go back to the slides because we want to show you a little bit of how we think about the stack. But we know this question is going to come up. So we thought we would just address this head on. You've probably seen here in the last couple of weeks, a lot of announcements from the frontier model companies about their enterprise agentic stack. So let me start with saying this, these are companies that we deeply, deeply partner with, both OpenAI, Anthropic, the other model companies, the hyperscalers. We are very close partners in really building this technology and bringing it to customers. And secondly, these companies are also customers of ours. And so we also work really, really closely together to make sure that they are being successful, as they're building these remarkable companies with this remarkable technology that is genuinely changing the world and changing the enterprise. But there's some important takeaways here when you think about what the frontier companies are saying about the enterprise business. So if you go to the next slide, we kind of broke this down a little bit to map it to how we think about our stack. So let's just start with frontier. We are in deep agreement with OpenAI that context really matters. As Gabe showed you, the workflows really matter, where your employees are working and how they are working is really, really critical. You need a system of agents across all of these important surfaces in order to go deliver this capability out to the enterprise. And so that is really good validation of the things that we have said for a couple of years. The bottom line being the model layer alone, which, by the way, isn't really represented anywhere on the stack. But the model layer alone is insufficient to make sure that, that intelligence is being turned into work. And in fact, we also were encouraged to see that OpenAI is partnering with many of the big system integrators that we work very closely with as well. Accenture, Deloitte and McKinsey and others because that implementation is really critical. Turning this into operation is really critical. And so this is a deep and important validation of our understanding of how these things actually get turned into real work. And then Anthropic, of course, has also been talking about their enterprise strategy. They had an event earlier this week where they talked about their strategy. And you kind of see how these experiences that Gabe talked about, the employee experience that we think of in Slack. These prompt and automation workflows that we've now been doing for a couple of years with our customers. And then, of course, these agents that are going to be building truly autonomous capabilities are really, really critical. Now it's interesting in one of the main demonstrations that they showed, the experience began in Slack, a group of people were talking about how do they improve a decision that they needed to make. But the immediate next step was someone needed to leave Slack and move to a different UX in order to go execute a task. Then they had to take that task, bring it back to Slack in order for the decision to get made. Now we don't believe that's the right experience for users. We think all of those things have to happen in one place. And of course, we're deeply partnering with Anthropic to make sure that customers do not have to leave their flow of work in order to bring all of that intelligence and that capability. And so a deep validation of the things that I think we've been saying for a while as we have built this comprehensive stack that MK talked to you about before. So let's go to the next slide. This is now a view of our system of agency. We touched on each of these things. So I'm not going to really get into -- repeat the detail that we just talked about but a lot of deep innovation at every layer. And this innovation is not limited to Agentforce. Gabe showed you the operating layer today, where we optimize, we analyze. Well that layer is built with Tableau. When we think about the orchestration layer, an agentic system where agents are talking to each other. We are very closely tied with our MuleSoft technology to extend that across the enterprise. Of course, at the context layer, we're making sure that we're tying in with Data Cloud, with Informatica for all of the reasons that MK said to really surface up all of those capabilities. And when you think about the experience layer, this is the work with our applications teams to really make sure this intelligence is surface for employees. It's working with Slack, so you're in the system of engagement. It's working with all our channels like voice and chat and WhatsApp to make sure these agents are surfaced in those channels. And so it's a really critical strategy for us to bring these unified capabilities to our customers. But really importantly, we are unified, but not locked in. We know we live in a heterogeneous environment that customers in the enterprise will want to connect a lot of technologies at every single layer. And so the open side of this equation really, really matters. Openness at the data layer, the ability to orchestrate across all of these systems. The ability to share telemetry and data, so that customers can do the analysis that they need and then the ability to really connect across all the interfaces. It could be Microsoft, it could be Apple. It could be whatever interface a customer really wants to use and make sure that our agents are able to surface up in those interfaces. And so that's really the strategy and what we've done. If you go to the next slide. But really importantly, a strategy and a set of slides and a set of demos and blog posts are really not the business that we are in. The business that we are in is to be really, really deep in the trenches with our customers, really making sure this technology is making them successful. And it's been a real privilege to work with some of the biggest companies in the world. As I said, across industries and across use cases, so we can really, really understand how this technology can create value in a number of different scenarios. I won't drain the slide, I'm happy to talk about some customer examples in Q&A. But each of these customers has helped us drive a critical innovation in the product. Working with Williams-Sonoma really helped us understand how do we tie that context layer together to create an incredibly rich experience. And if you haven't played with it, you should go to Williams-Sonoma's website. This is an agent that's helping you not just discover Williams-Sonoma's products, but really create experiences in your life. And it starts from that in order to go in and then make product recommendations and actually think about what products customers want and then customer service. But tying all that together requires a tremendous amount of context. Adecco is an incredible customer of ours. In fact, they just yesterday launched a brand-new voice experience for their candidates. And Adecco uses us to do what is their most important business process, which is qualifying candidates in order to match them to jobs. And that process is a perfect example of why LLMs and determinism have to come together. You want to use LLMs to create this rich experience with the candidates. It's flexible. It can ask them questions. It's empathetic to where these humans are. But a qualification process has 30 steps. You want to make sure every single step is followed. If you just handed that off to an LLM, it is not going to execute consistently and accurately across all of those things. Equinox. Equinox really helped push the boundary on how do you create rich experiences with this technology. This isn't just a bot that you put on a website. You want to have rich information. You want to have rich content. You want to have interactability. You want to have flexibility in how the agent actually plays out. And so it's really been incredible to partner with these customers as they're pushing the boundaries of how this agentic technology is going to be used. So what I want to do next is hand it back to MK. He's going to give us a little bit of view of where is this technology going? What does the future really look like. MK, over to you.

Thanks, Madhav. I think what you saw in even Gabe's demo is a hint of where agents are going to be calling other agents, the super-agent, as we call it, right. Because we want -- every customer of ours wants a single brand agent that coordinates work across everything else because when you go to salesforce.com, you're not going there just with the intent of just asking a service question. You might want to go to sales. You might want to go to marketing and so on. The Williams-Sonoma is a good example, where you may be asking about the soup chef agent about some recipes and then move on to going purchasing things. And so super-agent becomes an important thing. We believe 2026 is the year that every company and the brand will choose a super-agent and be ready for it. And so on the consumer side, with Agentforce. We have the super agents that can go talking to other agents or let other agents call into us. Next slide, but even more interesting on the employee side is where we believe Slack is going to play a huge role in the system of engagement and super agents for employees. Next slide. And that comes with Slackbot. If you have not tried Slackbot, you must try it today. And Slackbot is a perfect example of how the context that's in Slack, where all our work gets done. It uses the power then of our AI tools to be able to go bring that right thing at your fingertips. And it's able to go orchestrate not just across all your data in your enterprise, but across all your agents as well to get your task done. And as you can see here, we've already seen huge success with Slackbot. In fact, we heard from many, many customers, where if their Slackbot was turned off, they literally said we can't work anymore. So it's become such an important ingredient just a few months since it's been released. So that is how we believe both at the consumer side with our Agentforce for super agents, on the employee side, with our Slackbot and Slack super agents, we're going to really bring the power of agents talking to agents to your fingertips. I think that's the end, correct? Yes. So with that, thank you all, and we're ready for your questions.

Awesome. Thank you so much, Madhav and MK. We are going to jump into Q&A now. As a reminder, please do submit your questions via the Q&A feature in the Zoom channel. To start here, we have a question that we've been getting pretty regularly from investors this week. People are really excited to see the Agentforce momentum, the $800 million ARR, the strong growth. But people really want to understand what the long-term ARR ceiling or long-term growth rate assumptions are when we think about our path to our fiscal year '30 revenue target. So maybe I could start and then pass it over to Madhav. I think first, just what we're really excited about with that $63 billion. We have incorporated Informatica, of course, but it also takes into account the net new AOV performance we've had over the last few quarters. And when we gave you the update at Investor Day on how we see the slope of the curve into the back half of this coming fiscal year, that's what really gave us a lot of conviction, is we're seeing this broader adoption motion that's Agentforce, but also let me go deeper on my core apps. Let me go deeper with my products with Salesforce and really driving a broader set of adoption than just Agentforce over here and the rest of the business over here. So maybe Madhav, you can help bring that to life in the conversation here.

Yes, I think it's important for us for everyone to get an understanding of how we go to market and what the monetization strategy looks like. So when you think about Agentforce, you can think of three key ways in which we bring this capability to market. Number one, as Val mentioned, we sell products that create premium experiences for employees. So this is a license business. This is an upsell uplift business. And we have products that are add-ons to our cloud. So we call them Agentforce for Sales, Agentforce for Service. We have all of these products for the industries as well. And then we've got the most premium Agentforce 1 Editions, and those are 50% to 70% uplift on a per seat basis. And these products are really ensuring that in employee scenarios we're driving that productivity. We're driving the agency that customers really have for all of those kind of employee use cases. So that's a really significant part of the business. Half of the ARR really comes from that license uplift business. And we're continuing to see expansion and growth, a remarkable trajectory. We just launched that product in the middle of last year and remarkable momentum there. That's one type. The second type that's really critical is the consumption business. And so what Gabe showed you is an example of an external-facing agent that is doing a lot of complex tasks. In many cases, those tasks cut across marketing and commerce and service and these are agents that are truly autonomous, creating new types of experiences and so we sell that as a consumption business. And that business also has grown really significantly, a significant ARR on that business. And that gives customers the ability to really build these flexible multiuse case agents. There's also a lot of value. You imagine an agent that is now acting as the front door, as MK said, as we build towards these super agents future. That is a remarkably valuable experience from a customer perspective and something that we expect we will also be able to generate a lot of value from. So that's the second model. And then the third model for some of our customers that really go end-to-end with us is this new Agentforce enterprise-level agreement. And those agreements are really looking holistically across a customer. They're thinking about the licenses and the seats. They're thinking about what are the consumption agentic use cases. How is this customer going to use data cloud? How do we integrate across their enterprise with MuleSoft? How do we bring the power of our analytics with Tableau. And so a comprehensive really at the transformation level and that we introduced not that long ago, and we have a lot of customers and great momentum on that front as well. So that's really how you should think about the 3 monetization models. And as Val said, expect uplift in all of those things. So today, we report Agentforce as kind of a sum of those things, and we will continue to do that. But we expect Agentforce to really penetrate and uplift all of those businesses. And so far, we're really seeing a lot of expansion in each of those buying models.

So the next question here is one that I think kind of ties very closely to this question. And when we think about the longer term, obviously, we had a record quarter with RPO, $72.4 billion in Q4 of FY '26. But how do we think about the trade-off, as you mentioned, between seat licenses, agentic enterprise license agreements and the flexible consumption credits, especially in the context of potentially reducing seats, right? I think everyone saw some of the news that came out yesterday. So how do you think about that? How do you think about the growth in agentic work units with that as the backdrop.

Yes. And maybe, Val, you could start with some of what we shared at Investor Day on how we think about the expansion on a per customer basis because I think that will be...

Yes. I think just to remind everyone, when we think about the overall wallet share expansion opportunity that we have with our customer base. If we are able to address the agentic enterprise opportunity with these customers, it's not just an incremental add-on type agreement, right? When we're going live with products like Agentforce Data 360, we're not only getting this motion of customers more willing to go deeper on their current applications whether it's in Service Cloud, they want to add on field service because they want to plug in, to tie in the agentic experience to be able to plug in across that use case. Or they're willing to go from just a Sales Cloud service cloud experience to add on Tableau or Slack because now they have confidence in the ability for these agentic capabilities to lift all of those products and the underlying context to get a lot of value from all these different touch points. That for us, when we've seen customers who have gone live early, we've actually seen a significant uplift that goes beyond 20%, 30% growth. It goes 2 to 3 to 4x of an overall spend expansion. So when we're trying to address this opportunity, we're really trying to address 3 things. First, of course, we want consumption. We want to ensure that we get high-value consumption. Agentic work units is a really good measure of that, and Madhav, you can touch on that. The second is we want to have a lot of value in the seats that they have today. We still see seats expanding right, across sales, service, Slack. There's still growth that we're seeing today. Longer term, that might change. But I think a motion that's been really exciting for us is these bundled SKUs, Agentforce 1 Editions, the more premium seats where they're actually increasing their ARPU and to translate that to customer terms, they're getting more value out of that existing seat and are willing to pay more for that seat base. And then the third piece is really around that data in context, right? We just brought Informatica into the fold. It's been a great first quarter for us with that asset. But now with Informatica, MuleSoft, Data 360, Tableau, all these assets together really give us a broad scope to be able to address the context, data needs that our customers have. And that gives us a lot of confidence in that 3 to 4x kind of expansion opportunity that we shared at.

So let's build on what Val just said. I mean, 7 of our top 10 deals saw this kind of motion. This isn't the same as I have 1 more app I'm going to add, and that's going to give me 10%, 20% uplift. This is now, I'm going to reimagine what my enterprise experience is going to be. And so that is a 2x and 3x and so on. And so when we really think about that, the right question that we've been really thinking about and that we talked about a little bit at earnings as well as every one of our customers is going to be on this journey of becoming an agentic enterprise. The really key question for us is, how do we make sure that Salesforce is clearly and obviously the partner for our customers as they think about completely changing their customer experience. Now we've had very encouraging growth on that front. We started with 3,000 Agentforce customers. We are now up to 23,000 Agentforce customers. And so the rate at which we are penetrating our customer base is really significant. We've added a lot of new customers along the way as well. And so that penetration of, let's go help our customers transform into these agentic enterprises is really the key. That's on the employee side, that's on the data side, and that's on these very significant orchestrated super agents as well and those types of use cases that we're driving. To measure the work of those agents is really where we think about agentic work units, and MK, maybe this is something that you could help talk about a little bit as well. We really want to move from just measuring pure input consumption like tokens into output. So MK would love your thoughts on the Agentic work units and how you think about that for our customers.

Yes. Sounds good. First, let me add one more thing to what you guys said, and it kind of drives into the agentic work unit. If you take that sales engagement STR agent, what it's able to do today is really go bring in more customers that we would probably never have even looked at before, leads and opportunities. Like, for example, even within Salesforce, like almost 90% of the folks who come to our website and others is too small for anyone to actually go have a meaningful conversation with them. Now with our sales agent it actually starts engaging with them, knows what they're doing and then it brings them up either to schedule a meeting with the salesperson or to actually close the deal. Now that is an example where it's actually going to cost more seats to -- like more humans to be used to kind of create more sales, grow the top of the top line. Now coming back, one of the challenges we saw was just looking at raw token usage doesn't give you the full picture. Because that's just -- okay, we are going have $1 billion, somebody else is going to have a $1 billion and so on. With this new agentic work unit, what we're able to say is it's not just about the tokens. It's also about the work that is happening within that business. So in the case of sales, it would include the workflow that's needed to go call and create the lead as an example. In the case of service, it's the work done to actually go understand that user's intent and then go close the case or like create incidents and so on and so forth. So each 1 of those, the agentic work unit will capture that math of what is the actual work done in that system of context or in the system of work. So that we can give a more comprehensive thing of how AI is actually helping your enterprise.

And this is really important because we believe the Agentic business needs a canonical metric where we are showing that valuable work is getting done. When we measure AW use, that is not internal use of agentic work units. That's not trial use of agentic work units. That's not even the use of agentic work units when customers are testing, which is super valuable, by the way. But we want to hold ourselves and this industry to the standard of is real work happening in production. And I think having a canonical metric that can measure that is really, really important for us to understand how we're creating value with our customers.

To Madhav's point, we welcome the rest of the industry to join us in this effort so that we can actually have some common metric across the industry to represent how work is actually getting down.

Awesome. Our next question, I'm going to combine 2 questions into one. But Matt VanVliet from Cantor Fitzgerald is asking about super agents. Are they going to have different monetization elements to encourage uptake. And then I'm also going to pull in a question that's tied to super agents and multi-agent orchestration. Specifically, if you're bringing in a third-party agent, how are you thinking about the data ontology? Do you have a layer that would help you understand and map the data across different systems. So maybe that second part can go to MK and the first part can go to Madhav.

Yes, absolutely. I mentioned the different monetization models that we had. I think another really important point here is, we have really learned over the last year just as we've innovated on the product, and you saw some of the incredible capabilities that we've built already and some of the things that are coming. We've also really innovated on the go-to-market and the pricing motion. And I think where we are is making sure that customers have flexibility and making sure that we're meeting customers where they are, where they are in the journey. If what they want is a model where they are still thinking about productivity in their employee use cases. Great, let's make sure that we have offerings, and we have licenses that they can buy that they can leverage all this capability. If they want to move to these more complex agents that they're building, including super agents in the future, let's move them to a more consumption model. So they can really understand the real specific work that, that agent is accomplishing, and the monetization model is tied to that. That same argument is going to apply in the case of super agents. When you think about super agents, there are going to be actions that are executed across these agents, either within Agentforce or into external systems. We will absolutely think about how we monetize those actions as we do today. They're also going to be pulling in things at the data layer. And that data layer will all be monetized with the way we think about our credits and our data layer monetization. It's very likely in those super-agent systems. They are going to be tying across the enterprise. So they have governance, so they have observability. So they have control, which means our MuleSoft capabilities will also be involved. And you've probably noticed over time, we're really moving towards a model where all of our consumptive products are on our single Flex credit system. It's fungible. It's movable across all these capabilities, and we make it as easy as possible for customers to say, "I need the agent capability. I need the data capability, I need the analytics capability, I need the Fabric and the Governance across my enterprise. I can get all that with one mechanism which I want to buy. And that's really what the super agents are going to represent. They don't represent just Agentforce. They really represent that full unified stack that MK talked about to give customers the ability to take advantage of it. MK, do you want to talk about the data side?

I think you've sort of apprised it very, very well. The things that I think you saw in one of the earlier Madhav's slide is, for us, extensibility and working with your enterprise is a very, very critical part at every level of our stack, right. From the model layer to the data layer to the agency layer, work layer and so on. And we are extending that same thing to multi agents as well. And the question specifically was on context. And this we all know, right, even humans, when we get transferred from one operator to the other, we have to go repeat all the context back again to saying what my problem is. And so we want to avoid that. And to that extent, the Data 360 layer that is creating the context with our Agentforce has open APIs through REST APIs, MCP, JDBC, et cetera, that all the other agents can also leverage. So when there is an agent-to-agent handout, there is no loss in context or translation. So that's one. Second, to Madhav's point earlier, as these agents are running, if you're using our agent Fabric, all the logs across those agents are also going to come back into data 360 and surface through our Agentforce Studio. So that means you can actually understand how the lineage is working, how these agents are calling and actually start optimizing as well. And that helps into that circular loop that we can actually do to go optimize your agentic enterprise. So the short answer is, our data layer can help you go across these agents and to also grab all the logs. Our MuleSoft layer can help you govern and monitor these agents and our Agentforce Studio layer can actually help you go optimize them through our analytics.

Awesome. The next 1 here comes from an investor who is wondering about the maximum margin impact Agentforce can have on gross margins. Let me start and then Madhav, you can jump in on some of the things we're doing to optimize. First, we, of course, have a product portfolio that's pretty broad and diverse. And as we launch new products, we always are looking to long term optimize what the gross margin structure looks like. Our expectations that we've been clear with, with investors about is we expect to maintain our gross margin structure. We have some things working through this year as we invest in Hyperforce. We invest more in third-party that longer term, we actually expect some efficiency on the gross margin side. But importantly, when we think about the construct from monetization through to optimization, the monetization side captures more than just sort of pass-through costs for LLM. As you've seen today, there's so much more in that overall agentic enterprise architecture that needs to be right, that needs to be working in order for these to scale, to be reliable and to be really successful for our customers, and that includes much more than just kind of a pass-through cost. So I'd say when companies are out there saying, "Oh, we're worried about LLM pressure and that impact." There's a lot more that we're doing than just sort of a pass-through. The second side of how we're thinking about optimizing is we can view the kind of margin by customer view of when early adopters launch, how that scales and ramps through time, how we adopt things like hybrid reasoning, determinism. The testing center that you saw us show in demo, there are a lot of ways that we can optimize fine-tune. And then, of course, importantly, we're able to leverage a lot of different models in that architecture to be able to use the right model for the right task to be able to leverage some of that efficiency. So maybe you can touch on some of that.

Yes, absolutely. Look, efficiency at the infrastructure layer is something that we do every single day. And that applies to compute. It applies to storage. It certainly applies to the use of tokens in the model here. This is the newest kind of infrastructure and the newest kind of swappable infrastructure, especially as we have model choice with customers that we will continue to leverage how we are efficient at that infrastructure layer. But there's a really important point here that I think ties back to the agentic work units. Now what the agentic work units allow us to do now is they allow us to compare the work being done with the infrastructure supply. In this case, the inference capability and the intelligence capability measured by tokens that are going into those work units. So over time, we can understand which kinds of work are leveraging how many tokens over time. There are probably some kinds of work, whereas we build richer and more complex experiences the amount of tokens go up for a little while, and then we plateau as we start to think about it. There's other kinds of work, and we talked about this earlier, as reasoning becomes more deterministic. As we're not relying on LLMs for complex task execution, the number of times we have to use an LLM in that reasoning will drop. So we now have the ability to say for those kinds of work units are we driving efficiency from a system design perspective, from an agent construction perspective. And this really allows us to kind of dial in the amount of infrastructure costs we are spending for the work that's delivered. Now this is not different from what we used to do with regular compute technology and regular storage technology, but now we're going to be able to do the same thing with this new model technology. We also fully expect just has happened on the infrastructure layer in the model layer as well, there's going to be continued improvements, efficiency, lowering of cost in these models as they continue to get more commoditized. We're already seeing some of that, and we expect that trend to continue in the future as well.

Great. Our next question is about MCP. And now that MCP allows external AI agents to interface directly with enterprise data, how do you prevent other companies, foundational model companies from eventually bypassing Agentforce entirely and utilizing their own agent orchestration capabilities? As those are getting better and are maturing, how do you prevent that from happening?

I think even before agents were in play, we were an open company. We shared all our data. And yet people are still coming to us for doing all of their data processing and for all their work needs. And same thing here. We still had all our APIs open. Anybody could have built an agent outside us. And truth be told, a lot of customers did try it 2, 3 years ago when LLMs came out. They said, "Oh, all I need is just an LLM and I just give all the APIs and things will just work." And as we sort of -- and Gabe showed the demo and others and we showcased it, what we have is really an understanding of the business over the last 20, 30 years. What our customers have done is really build those business workflows, business data context and metadata and all of those. So to really get your business successful, it's not just that data context or just the raw data. It's really everything that goes on top of it, including your system of work, your agency all of the hybrid reasoning, everything that we talked about. And the context that gets build, it's not just raw data, it's really the context that you need to go build as well. So that's kind of why we are very, very confident that people need that stack that we talked about to get real business success.

And we touched on this when we showed a little bit of what the model companies have talked about earlier as well. There is no question the context layer is incredibly critical. There is no question that the business process and the workflow layer is incredibly critical. There is no question that the interface where a customer engages with the company, where employees work is really critical. And these experiences are not just unlocked by connecting up a bunch of MCPs together. It sounds easy in theory, but the experience really matters. The efficiency matters, the accuracy matters, the latency matters. These are implementation decisions that are really, really important that can be kind of talked about in a theoretical way of well, you can hook up any 2 things together and get them to work. But that's not actually how CIOs and companies implement things and how efficient and how well these systems work really does matter. By the way, our partners at the model companies really agree with us on this point. And so the question that you should be asking yourself is, are customers going to recreate all that from scratch. An entirely new stack to maintain, an entirely new stack to observe. And entirely new stack that, yes, you Vibe-coded, but now you have to Vibe operate over the next several years. Are you going to recreate where your employees are working when they are actually getting incredibly productive work done today on surfaces like Slack? Are you going to recreate every single business process that you have spent years and years really honing in sales and commerce and marketing and service? And that is where we believe the real work gets done. Capabilities like MCP are going to be incredibly impactful. We're going to be tying those to agents outside of sale force. Agents outside of Salesforce are going to be tying into us using all of that. And those are important building blocks, but a building block is not an architecture, and it's not an implementation, and that's really what we're focused on.

Great. Our next question is on Tableau. And specifically, is there any competitive threats or challenges with LLMs and agentic analytics that you're seeing today. Obviously, Tableau had a weaker performance in Q4. So what is happening there, if that's not the case.

MK, do you want to start?

Yes, I can start there. I think one of the big things, Tableau as well, with Tableau, I think it's certainly different. The agentic analytics is where sort of analytics itself is going. And you want to get the answers right. This cannot be a hallucinated answer. Like if I -- if Mark wants to say, "Hey, what's going to be my quarter next year? And what's your thing?" It can't just say it could be 80%, it would be 50%. It could be 30%, right? You need precise answers. And that's really what Tableau always has strived for. Making your data available to the -- to all of us that is guaranteed. Now one of the key things that we are working on very closely is to make sure those agentic analytics can get you the right answers. So that starts with making sure you get the right what we call a semantic data model, right? And out of the box, we now ship for all the customer domains and also making sure that the conversion into sequel and the semantic models is correct. And that has kind of helped. We had a bunch of acquisitions we did as well to make that better. And this is what we believe is the future of analytics. Like how do we make analytics really give you in a conversational pattern and embedded in your line of work. And in fact, some of the demos that you might have seen already where we now have Tableau through Agentforce, Tableau through Slackbot, all of those experiences we are trying to bring in Tableau relevant in your line of work. And all the applications that we now ship, like marketing investment everything, you saw the Agentforce Analytics and so on is also now all built on Tableau. So Tableau is much more deeply integrated into our stack. So we have expanded the TAM for Tableau into all of our existing Salesforce stack as well. Plus we are making sure Tableau will be the premier solution that you use when we have this agentic analytics taking over.

Madhav, do you want to add anything more?

Yes. I think -- no, I think you answered the Tableau question, but I think this is kind of the broader point we made earlier, which is when you think about Agentforce, it's not just the Agentforce part of the technology. It's the impact and the uplift that it has on every part of the portfolio. We're making our apps better with these agentic capabilities. As customers use the platform, they now will leverage Tableau to do the analytics that MK said. They're going to be leveraging Data Cloud and Informatica to really harden that context layer. They're going to be using MuleSoft to connect across. So we really think about the Agentforce momentum, you should really think of it in terms of all of these pieces coming together to create these experiences, leverage many, many parts of our stack to create that overall experience, and I completely agree with MK. The utility of a tool like Tableau in a world where the optimization and the management of agents is going to be a critical part of how every business works incredibly, incredibly valuable capability in that world.

Great. We're going to take our last question here, which is on Data Cloud or Data 360. Clearly, Data Cloud is growing really well. It appears to be a prerequisite for preventing AI hallucinations and Agentforce deployments. So are customers who adopt both Data 360 and Agentforce together, showing materially better retention and expansion rates than those who are using Agentforce alone?

Yes. I'm happy to start and toss at MK. We see a huge overlap in the customers that are using Data Cloud to connect across their systems and the customers that are using Agentforce. Data Cloud is also a key part of the Agentforce architecture. We use Data Cloud to make sure that we get all the analytics. We have -- Gabe showed you a slide that showed as you're testing the agent, every single moment that the agent is making a decision, executing something. We call that a session trace. All of that session trace really sits in data cloud so that we can leverage the power of all of these analytics in this data. So the 2 products are very closely tied together. And then, of course, customers also use data cloud to extend across the enterprise, bring in all of this context. So a huge overlap in that customer base and I do think that the products really are complementary and drive each other's momentum, and we expect that to continue. MK, anything you want to add on data cloud?

Yes. See, one of the things that we've also seen, in fact, even recently with a very big customer, is that as we lay the foundations of this, we can easily upsell so many different technologies, because all our platform works together. And so that means somebody who is using Agentforce and Data Cloud, we can now upsell our Tableau stack on to them. If they're using service, now marketing works better because the same data and the same agentic experience now can work with our marketing deeply. And so this thing causes us to be able to go upsell and cross-sell all our technologies together because foundationally, we have architected our stack so that the data and the agent stack underlies all our application experiences and all our systems of engagement. So that is a huge boost to us. So it's not just about customers -- retaining and growing our customers with Data Cloud and Agentforce, it's really growing and upselling all of these other clouds as well.

Awesome. Well, thank you all for joining. Thank you, Madhav, MK, Gabe for hosting today. We look forward to seeing you on the road over the next few weeks. And we have a bunch of questions here that we'll get back to you on, but we really appreciate the time. Have a good rest of your weekend.