Name: Kishan Chetan, Salesforce
Uploaded: 2026-09-25T17:45:00.000Z
Duration: 17 min 10 s

Clips
More from Dreamforce 2026

Kishan Chetan

EVP and GM, Agentforce and Service Cloud

Salesforce

play_circle_outline Outcome-Based ROI: Customer Satisfaction, Productivity, and Cost Savings from Southwest’s Casey Handling Millions of Conversations

play_circle_outline Salesforce Agentforce and Service Cloud: Contextual AI Agents Driving Personalized Outcomes via SOPs, Documents, Calls, and Workflows

play_circle_outline Auto-Building Agents from Conversation History to Fix Undocumented Processes, Fragmented Data, and Integration Issues

play_circle_outline Future focus: expand automation, broader Agentforce availability across channels and functions

14:52

17:09

15:19

Info

Kishan Chetan, Salesforce

Kishan Chetan

EVP and GM, Agentforce and Service Cloud Salesforce

George Gilbert

>> Palo Alto Studio, connecting Silicon Valley and Wall Street. I'm John Furrier, the host of theCUBE here with Dave Vellante, my co-host.

Alison Kosik

>> Welcome to theCUBE NYSE Wired program and community coverage of Dreamforce 2026. We're coming to you live from San Francisco. I'm Alison Kosik alongside George Gilbert. And I'm excited to get into this next conversation because we're seeing AI agents move beyond answering questions and actually take actions. And it's going to be interesting to talk about what that means for businesses and customers.

George Gilbert

>> Yeah. And what Salesforce has done probably more than any other vendor was put all the pieces together so that the customers don't have to. So it just works and we get beyond the request response and we can get to outcomes. And for that, you need the data, the applications, the agents, and it all becomes essentially a harness that does stuff.

Alison Kosik

>> Yeah, absolutely. All right. Let's bring in our next guest. It's Kishan Chetan. He's the executive vice president and general manager of Agentforce and Service Cloud with Salesforce. Welcome to theCUBE. So great to have you.

>> Thank you so much for having me.

Alison Kosik

>> So nice to see you as well. We're here at Dreamforce outside. It's a beautiful day here in San Francisco. at Dreamforce agents are clearly taking center stage. I'm curious about this because you lead Agentforce and Service Cloud. What for you has changed dramatically in the past year and about what AI agents can do for customers?

>> Yeah, look, first of all, thank you for having me. It's fantastic to be here. A lot of energy, a lot of excitement here. Agents are driving a lot of it. But I think a big part of what I've seen over the last year, which has really changed, is how much customers have adopted this technology. We had it in our keynote for Agentforce and it was all about ROI. We had customers stand up on stage and speak about how they're getting great ROI. So in the case of service, the ROI could be higher customer satisfaction. It could be the productivity because of lower costs, but lower cost driving more implies you can do more service. So we are seeing very clear ROI metrics and Marc actually spoke about this in his keynote, Marc Benioff. He spoke about how Casey, our agent, helps them handle 5 million conversations over the last 2 or more years and how we created a bunch of pipeline and managed pipeline with our sales agents. And that's the type of value we're adding. It's about the ROI.

George Gilbert

>> So Kishan, putting the pieces together, for the first couple of years of agents, we had a lot of technologies, the technology side of the house was often empowered to get something stood up really quickly. And it was often a standalone, you know, agent development toolkit. And there are more of those than you can shake a stick at.

>> Yeah.

George Gilbert

>> Help us understand why it's so difficult to tie in the context from all the systems of your existing systems, but also then the actions that are in a normal agent, a tool is how it accomplishes something. But in like Salesforce, the tool is how you get to an outcome. Help us understand how by supplying all the pieces, it makes it so much easier for our customer to get to those outcomes.

>> Yeah, fantastic. Look, what you want the agents to do is, at the end of the day, it answers a set of questions and it takes actions on your behalf. I'm oversimplifying this, of course there's a lot more to that. And then of course you want it to be personalized, you want it to be contextual, because if I'm a patient going and speaking to my pharmacy, I want that agent to know who I am and ideally know what my prescriptions are, to give an example. So that all those three need to come together. So which is, it needs to know more about me, And that's where all of the technologies we have to bring the right context of who you are, but what you've done in the past are important. The second is what we call knowledge. The agent is only as good as what it knows, what it's trained on, and being able to take all of this corpus of knowledge that you have within your organization, your SOPs, your documents, your historical calls, getting that and using that to build that agent, that becomes important. And finally, the last thing is it actually needs to do something about it. If I'm calling a bank to change my address or create a new account, it actually needs to go do that. And that's where the workflows matter, and that's in Salesforce. So the combination of these three, knowing who you are, being able to have the agent trained, and then being able to do something about it, that three becomes really important.

George Gilbert

>> So I would gather that Southwest is actually having a fair amount of success with the Help Agent, Casey.

>> That's right.

George Gilbert

>> Okay, so tell us what they're accomplishing and how they measure the outcome if they do.

>> Yeah, yeah, great question. And look, Southwest was actually on our Service Cloud keynote last year in Dreamforce, and it was fantastic because what they wanted to do is they wanted to give a very specific contextual answer to their passengers who are there. So they wanted to ask about flight questions or questions about what's available. They wanted to get the right brand relevant answers and a reasonable percentage of them essentially are now answered by the agent. Then they also wanted to have the AI and agents necessary for making their own reps very productive. So which basically means that when a service call came in and a human needed to address that, that needed to be addressed properly. With the right brand tone. And both of those are things that Southwest was achieving. It was a great keynote. We actually had them also yesterday on the Agentforce keynote. So that's a continuing success over the last year and a half.

George Gilbert

>> So how do they, do they measurea cost benefit? Because, we've all been struggling for like trying to understand how do you price, how does the vendor capture value? How does the customer measure the value that's been delivered? I know there are different variables, there are different ways of doing that now that Salesforce has been evolving.

>> Absolutely. we've been evolving our pricing model over time. But just to be very concrete here, one of the key things in service, the good news about service is you have a good sense of whether a customer's question or their request or their issue has been resolved. So we've actually moved to an outcome-based pricing which basically means our customers pay us only when their customers' issues are resolved. So which means that if somebody has come in and asked a question about needing to reschedule my flight and that reschedule actually happened, that's a resolution. If that reschedule needed to go to a human because the agent couldn't solve it, that's not a resolution. So that's how you essentially have a very clear way of measuring what the technology is doing.

Alison Kosik

>> Yeah, what should be the ideal handoff between an AI agent and a human? When it comes to customer service?

>> that ideal handoff really has to be a few things. One is there are several cases where the handoff is actually necessary. Like if I'm in the luxury industry and I want to ask questions about a new handbag that I want to buy, I want to speak to somebody likely about it. So that handoff, the ideal handoff, needs to have the right context. It needs to know everything that's happened so far, what's happened in the conversation, what was the customer sentiment, What is the customer looking for? And pass all of that so that the human doesn't have to wade through all of that. And then the human needs to also have— the rep needs to have a good summary and something from the tool telling you that, hey, you know what, she's looking for this handbag. Maybe these three are the right ones. And while you're doing that, maybe you can also have this other service that she can have, which is around maintenance of the bag. Just to give an example, that is what ideal experience is like.

George Gilbert

>> So let's dive in a little more into your role-specific agents like Casey.

>> Yeah.

George Gilbert

>> Like what went into, what are the ingredients that saved the customer from having to, roll their own? Like what's in Casey that last year they would have had to build themselves?

>> Yeah, a few key things. The first one is because we've done a lot of things to make it easier to set up, you can automatically connect to what I said about knowledge and connect to the context of what the agent is supposed to do. Second, it's easy to connect it to the channels where the agent is working on. Is it on the phone? Is it a chat? Is it on WhatsApp? That part becomes easy. But what the real secret sauce is this whole idea of an engine to kind of build and optimize the agent. And what it does is it looks at my historical conversations Let's say somebody's come in and had a bunch of phone calls about needing to change my address or updating something. You look at all that historical corpus of information, your phone calls, you take that and make that agent with that so that the agent is automatically starting with that context. So that's a big part of it. Then second, you have this out-of-the-box analytics, which means that, in a service context, you'd care about resolution rates, you care about handle time, You care about how many times it's escalated. These are metrics which are very specific to that context. So it's building the agent, it's optimizing the agent, it's measuring the effectiveness of the agent. And finally, we make that engagement between the agent and the human seamless and making that seamless. So those are the 4 things that we've done out of the box.

George Gilbert

>> Okay. So now, how might that— how much time might that save a customer, from rolling their own? If they just get a generic agent development kit and let's say they have Salesforce, but then they would have to pull data from multiple sources. They would have to attach, to Salesforce objects or the APIs. How much time does that save?

>> And we've done a bunch of work broadly. We also have what we call Enterprise Knowledge, which makes it easy with our Data 360 platform to get data from multiple places. So we've solved some of those problems to make it easier. To answer your specific question, we have customers who have been able to set up and get up and running in a matter of weeks, right? And if you build everything from scratch, depending on where you are, it could take several weeks to months. Again, I think the saving is pretty significant in terms of setup. But what I wanted to point out too is the setup is important. But what's really important is once the agent is up and running, can we make sure that the agent is actually doing the right thing, that it continues to do the right thing as your business processes change, you introduce new products. That's where the real power is in the optimization, not just setting it up.

George Gilbert

>> Oh, so there's a loop. That loop. So that's why you emphasize the analytics. Yes, because you're going to always be tuning it. And so what's— what— oh, this is a learning for us, which is it's not just getting it up and running, it's that it's tied into the outcomes that you can measure in Salesforce.

>> Exactly.

George Gilbert

>> Tells you how to continually tune the agent. And what would that tuning look like?

>> as an example, if you have an agent which is in a retail industry and you realize that whenever there's an order or you're adding new products, it always needs to go to a human. And that's what you're seeing from your analytics, you perhaps need to go and understand why is that the case? Do I have something which is missing in my corpus of information? Do I need to actually create a workflow so that it can automate that? That is what the tuning is doing. It's identifying gaps in your agents' build, in your knowledge, in your workflows that you can then further automate.

Alison Kosik

>> And what are the challenges in this that you're finding?

>> The challenges that customers are finding quite often is a few-fold. One is the knowledge and all of the data that's there. The processes that are important are not always documented somewhere. It's in somebody's head. It might be in some historical phone calls. So taking that information and making it available for the agent to use, that's a pretty big deal. That's the reason why our Agent Optimizer looks at historical conversations and automatically learns from it. And the second one, to your question on the challenge, is to make sure that you can orchestrate the work in different systems. So for example if I'm running an airline or a train service, one of the things you want the agent to do is to book a ticket or reschedule a ticket or maybe book a lunch that needs integration to other systems. Being able to build that integration is going to take time.

George Gilbert

>> So also, how does Salesforce and the customer have to engage differently now where you have to involve the technology side of the house. There's a business owner for the outcome, but that Salesforce also maybe has to have more professional services upfront or bring in a service partner in a way that they didn't before where the customer has to The technology side of the house is the enabler. The business side of the house owns the outcome. And those two sides have to work together in a way they didn't have to before. And then Salesforce has to maybe orchestrate that. Is that—

>> Yeah, a lot of that used to happen before, but there are a couple of things I think which have changed. One is the cycle of how quickly this has to happen has been really shortened. So you want to get the benefits in weeks, not months, days, not weeks, right? That's one thing. The second thing is in the previous world, you do a— you demo your capability, people would look at that capability, then they'll get a partner, who'll try to understand your processes, document it, and have an implementation. You don't— you can't do that in the Agentic world because what you really want is you don't want a demo, you want something working so that you can understand how is it working and continue to tune it And in order to do that, we bring our own, what we call FDE, a forward deployed engineer. Sometimes it's a partner as well. And our product teams are quite often engaged with both the business leader and the IT leader. So shorter, very focused, start, show success, and then grow.

George Gilbert

>> And that— okay. And then that loop, does that include when you're learning? Because this is really important. There's this— there's that looping just to get into the technical terms, the observability traces, is that what you're

>> learning?I spoke about analytics and I spoke about optimization. You can't analyze and optimize unless you have great observability, and observability is going to be different across different channels. Like, what you need to do for voice observability is very different from text observability. So observability and understanding how well is it doing, have you tested it before, can you continue to optimize, it becomes crucial.

George Gilbert

>> But that's part of what you're learning from. That's it.

>> That's right.

George Gilbert

>> Okay, okay. So this is, so just to sum up really quickly, that's why you needed this integrated approach, because you can get started really quickly with a standalone tool, but you're not going to be able to run that sort of learning loop and that optimization unless you have all the pieces.

>> Precisely. And there's one part of it, just to reiterate that, and it might be obvious, in our mind, but just to reiterate it, that optimization that you need to happen does not need to happen only over agents. It needs to happen over agents and humans. So the real thing is, let's say I'm trying to— somebody calls in working with an agent, it went to a human and the human did something. You really want to optimize the agent based on what the human also did. And that's the power. That's why the integrated platform really matters. And you also want to do it not just for service, you want to do it based on what happened in a service transaction and a commerce transaction and marketing, bringing it together is what really, really drives success.

Alison Kosik

>> Okay, take out your crystal ball for the final answer. Where are we going over the next year? How do you see things dramatically changing?

>> I think, for me, the one which I'm really focused on is helping customers drive success and really drive metrics which are great business metrics that leads to success. And the big technology thing that I expect to happen. We've got this optimization that we've launched. I want this automation to be much bigger and much broader. And you also heard about Agentforce, where a lot of our capabilities are available where the customers are. And that's really important. We spoke a lot about customer service, but my team also owns our employee service, IT service, HR, and we want that service to be available in Slack, available in Cloud, available everywhere where people are. So optimizing the agents and being available everywhere for all of the functions. That's, I think, what we are looking to do over the next 12 months.

Alison Kosik

>> All right, Kishan, great conversation. Thanks for stopping by theCUBE.

>> Oh, thanks for having me. Very nice to meet you.

Alison Kosik

>> Nice to meet you as well. And you're watching theCUBE, the leader in live tech coverage and in-depth expert analysis. We'll be right back.

Kishan Chetan, Salesforce

George Gilbert

>> Palo Alto Studio, connecting Silicon Valley and Wall Street. I'm John Furrier, the host of theCUBE here with Dave Vellante, my co-host.

Alison Kosik

George Gilbert

Alison Kosik

>> Thank you so much for having me.

Alison Kosik

George Gilbert

>> Yeah.

George Gilbert

>> So I would gather that Southwest is actually having a fair amount of success with the Help Agent, Casey.

>> That's right.

George Gilbert

>> Okay, so tell us what they're accomplishing and how they measure the outcome if they do.

George Gilbert

Alison Kosik

>> Yeah, what should be the ideal handoff between an AI agent and a human? When it comes to customer service?

George Gilbert

>> So let's dive in a little more into your role-specific agents like Casey.

>> Yeah.

George Gilbert

>> Like what went into, what are the ingredients that saved the customer from having to, roll their own? Like what's in Casey that last year they would have had to build themselves?

George Gilbert

>> Exactly.

George Gilbert

>> Tells you how to continually tune the agent. And what would that tuning look like?

Alison Kosik

>> And what are the challenges in this that you're finding?

George Gilbert

>> But that's part of what you're learning from. That's it.

>> That's right.

George Gilbert

Alison Kosik

>> Okay, take out your crystal ball for the final answer. Where are we going over the next year? How do you see things dramatically changing?

Alison Kosik

>> All right, Kishan, great conversation. Thanks for stopping by theCUBE.

>> Oh, thanks for having me. Very nice to meet you.

Alison Kosik

>> Nice to meet you as well. And you're watching theCUBE, the leader in live tech coverage and in-depth expert analysis. We'll be right back.