← MarTech Podcast ™ // Marketing + Technology = Business Growth
MarTech Podcast ™ // Marketing + Technology = Business Growth
The Agentic Evolution according to Salesforce’s CMO
00:44:52
Loading summary
Transcript93 lines
- [00:00]
Benjamin Shapiro
The Martech Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit iheareverything.com.
- [00:26]
From advertising to software as a service.
- [00:29]
Ariel Kelman
To data across all of our programs.
- [00:32]
Benjamin Shapiro
And clients, we've seen a 55 to 65% open rate.
- [00:36]
Ariel Kelman
Getting brands authentically integrated into content performs.
- [00:40]
Benjamin Shapiro
Better than TV advertising. Typical lifespan of an article is about 24 to 36 hours. We're reaching out to the right person with the right message and a clear call to action.
- [00:51]
Ariel Kelman
Then it's just a matter of timing.
- [00:54]
Benjamin Shapiro
Welcome to the Martech Podcast, a member of the I Hear Everything Podcast Network. In this podcast, you'll hear the stories of world class marketers that you use technology to drive business results and achieve career success. Here's the host of the Martech podcast, Benjamin Shapiro.
- [01:16]
95% 95% of generative AI pilots fail to deliver measurable business impact According to a recent report published by MIT, CMOs are rushing to deploy AI agents, but the gap between AI promise and impact isn't changing yet. Despite billions in AI investments, enterprise AI experiments are floundering left and right, while smaller, more nimble organizations are seemingly more successful in navigating the new world order in marketing. So what separates the 5% of AI implementations that actually work from the 95% that don't? I'm Benjamin Shapiro and joining me today is Ariel Kelman, the President and Chief Marketing Officer at Salesforce, the world's largest cloud based CRM and the creator of Agentforce, the Salesforce's platform for building AI powered workflows. And today Ariel and I are going to break down why most AI agents fail and how Salesforce Trust first approach is driving real results with more than 18,500 companies and what it really takes to move AI experiments to becoming agentic marketing organizations. Time for a short break to hear.
- [02:23]
Scrunch Sponsor Announcer
From our sponsor, Scrunch. When was the last time you actually visited a website to research something? For me, AI does all that work now. So if AI is doing the discovery research and deciding who or what is your brand's website really for the rise of AI bots becoming as important as new visitors is a massive change in user behavior. And that's what Scrunch is taking head on. Scrunch is the AI customer experience platform that helps marketing teams understand how AI agents experience their site. They tell you how your brand shows up in LLM results, where it doesn't and what's preventing your content from being retrieved, trusted or recommended. It's not just visibility. Scrunch shows you the content gaps, citation gaps, and how to fix the technical blockers that matter so your brand is found and chosen in AI answers. Face it, your most important site visitor isn't human anymore. It's the AI deciding what gets surfaced. And for Martech podcast listeners, Scrunch is providing a free website diagnostic that uncovers how AI sees your site, where you have content gaps and where you're showing up versus your competition. To see how LLMs evaluate your site, go to scrunch.com martech that's scrunch S C-R-U-N-C-H.com martech Ariel, welcome to the Martech podcast.
- [03:46]
Ariel Kelman
Hi Ben, thanks for having me on.
- [03:48]
Benjamin Shapiro
Excited to have you on the show. Thanks so much for giving us your time and what a privilege to have you here. Salesforce is focusing on agentic workflow, primarily through agent force. So talk to me about Salesforce's approach to agentic AI and why you're seeing this really large 95% failure rate across the industry.
- [04:10]
Ariel Kelman
Yeah, it's a good question. I think maybe I'd like to start with a little bit on the difference between making generative AI work for business and having work in the consumer space. And it's really all about context and data. You think about if I want to go search how to make a chocolate chip cookie recipe on chat GPT, you know, it's been trained on thousands and thousands of chocolate chip cookie recipes.
- [04:33]
Benjamin Shapiro
It's a great bartender too.
- [04:35]
Ariel Kelman
Interesting to figure out. Yeah. How many. But. And so that data is, it's in the model. But when we talk about business AI, these models, whether you're using OpenAI, Gemini, whatever, you're using anthropic, they're obviously not trained on all the data about your business. You have to feed that in at runtime. And that context of understanding about your business. And in the world of sales, marketing and support that we live in here at Salesforce, understanding the context of the customer is absolutely required if you're going to get any useful information back out of of an AI model. And so that's a lot of what that's probably the most important thing we built underneath Agent Force is really the ability to provide AI the right context by giving you a trusted way to connect to all of your enterprise data. So you think about a customer coming to your website and trying to solve a customer support problem and having The AI understand what products have you bought, where are you in the world? Because there may be regulatory things it needs to know about. Whether you've called before, all the previous interactions you've had with customer service agents using that to give in it. And that's the context in people's brain when they're helping out. You know your a customer. And so that is really the secret sauce of what we've done. That's the difference between the failure that people are seeing when they're building agentic AI use cases on their own. And we're using a platform that's really purpose built for this, that has done a lot of the heavy lifting on the infrastructure layer for you so that you can focus on building the functionality of your AI agents, not all the plumbing required just to make it actually capable of working.
- [06:24]
Benjamin Shapiro
Yeah, I was training a friend of mine who had never used GPT before at chat GPT before and trying to teach him the understanding of how important the context is and how to train the model. He's writing children's books so I wanted the model to go search for different children's books and figure out what's popular before it started to help him as a copywriter and that understanding of context first. Actually it's exactly the same thing that HubSpot's head of AI Nicholas Holland, when he came on this podcast said too is that we, we have passed the area in the evolution of artificial intelligence where prompting is the most important part. We now need to consider context to be the most important part. First you have Agent Force built on top of Salesforce. So naturally there's Salesforce's data being fed into how people build their agents. It seems like an MCP server is also a solution to this problem for other tools. How do you think about the difference between that native integration of Salesforce data to Agent Force as opposed to marteching this together for other solutions?
- [07:33]
Ariel Kelman
Well, I mean for a lot of the use cases that we're seeing with our customers in the world of sales service marketing, we have a lot of the data already in our platform, all that context that it needs around the customer. And we have a customer Data platform called Data360 that helps pull data in from all the other systems that you might have enterprise data warehouses integrating with your operational systems. So it gives you that single place to go for Agent Force to get all the context that it needs. And so most of our customers are able to get to really use that to feed all the data. I mean a lot of it here's the thing is a lot of this comes down to a problem that most marketers and most martech organizations have been working on for a long time, which is getting their data in order and even pre AI, even non AI use cases. I mean I feel very strongly that you can't do anything sophisticated in digital marketing without having your data in order. In terms of having some type of customer data platform where you have a single source of truth about all the things that you're doing with a customer, all your internal data, your first party data, and then signals that are coming in from other systems as well, that's the context that you need. And I mean I think where MCP comes in is integrating between different AI tools and it's something that we support on our platform and you know, we've announced other systems be able to integrate, we've announced partnerships with companies like OpenAI. So you'll be able to do things like update your forecast directly within ChatGPT. So I think MCP is a pretty good useful foundational technology. That's really changing how easy it is to integrate all these different agentic systems.
- [09:27]
Benjamin Shapiro
Yeah, fundamentally there's the garbage in, garbage out problem and MCP is the translation between wherever you're holding your data to whatever large language model. I want to go back into agentic workflows and marketers struggle to actually get something of value. What's actually happening when customers try to deploy AI agents and where are they getting stuck?
- [09:53]
Ariel Kelman
Well, I mean just first of all to be clear on that 95% study, it was really just about generative AI. There is probably a portion of those that were agentic, but I would guess it's probably a low percentage. But I think just if you say broadly across AI agents and, and generative AI projects in general, I think the, the, some of the failure modes that need to be avoided that I see fairly often, you know, on the technical side, you know, it's the data and on the people side it's I think too many closed minded people that aren't willing to work differently. And I think this people aspect of it's underrated. Where to use generative AI to solve problems and do things, you have to change the way that you work. And all the people we work with, everyone kind of has a rating, you know, almost like a Madden football rating. If you, you know, in addition to like agility or you know, strength. Imagine if you had one like change comfort, change management. Some people love flexibility. You know, there's people we work with, they see something new and they're like, oh, wow, okay, I'm gonna stay up all night and go use this. This is amazing. And I want to tell everyone about it. And there's other people are like, no, no, that's not how we do things here. And you know, when we're using, and you think about using AI to make for video production in a company, something we've been doing a lot with, it's a completely different way to do things that often challenge people's, you know, internal view of themselves, of what are they good at. So if you think, you know, let's say you do something we do in technology marketing a lot, customer reference videos. You know, you go to the customer site, do a shoot. If you think your expertise is in setting the shot up, right? Like, you know, you, you don't even need a cinematographer. You tell your camera guys where to go and you're like, and now like, you can have one of your product marketing people team up with a, a video engineer adept in AI tools. They're going to be able to do things completely differently and a lot faster. And you know, some people love this, some people don't. So what, what I found is one of the biggest keys to success is having the executives and all the different, the leaders of the different marketing teams use the AI tools on their own and model that behavior for the rest of the team. And that's gone a long way because I think some people just underestimate the change management aspect of it. And it's not like change management, like you need some big consulting firm to come in. It's simply like inspiring your team that there's a reason why they should nudge themselves to get comfortable doing things a little bit differently.
- [12:47]
Benjamin Shapiro
Yeah. There's the question of tops down versus bottoms up. Right. Do you have your workforce start to learn it, cultivate individual contributors, and then once they have the skills, then they can rise up the ranks as opposed to, you know, and I'm curious to hear how you do this at Salesforce.
- [13:04]
Ariel Kelman
Right.
- [13:04]
Benjamin Shapiro
In theory, most of the time the executives at your company tend to be the older, more experienced people and not to age or stereotype. Generally those are the people that are most resistant to change and rely on their previous experience. So how at Salesforce are you thinking about having your leaders start to learn the technology?
- [13:24]
Ariel Kelman
I think like a lot of cultural things, it's top down. I mean, they see our CEO is using all these tools all the time, playing around with them, experimenting. And really across my leadership team, each of those leaders is getting Hands on the tools. I mean, like if just to go back to the video example, our, our chief creative officer runs our whole creative and brand team. He, I mean, he has experimented with, you know, over a hundred different video production tools, chaining them together, basically doing entire productions without shoots, with AI generated video and a mix of existing assets. He figured it out on his own first and then asked the team, okay, let's figure out how to operationalize this. Which led us to different processes, hiring some different people, and then on our product marketing team, I gave an example, this guy Patrick, who runs our product marketing team. He, when we bought, you know, we bought a company called Informatica. And when he was prepping to do press interviews for explaining to people why did we buy this company, he took the messaging document, put it into Gemini, created a gem, a prompt template to simulate the different reporters he was going to talk to, ask him questions, and then grade him on his responses as compared to the messaging document.
- [14:48]
Benjamin Shapiro
Right.
- [14:49]
Ariel Kelman
You know, you can call this prompt engineering. But he thought of it as like, yeah, I played around with Gemini for like literally like 18 minutes. He did and then he had it done. He, and then between meetings he would go and run mock interviews and get scored and you know, then he goes and tells his team, like, this is what I'm doing. I had a problem where I knew that, you know, a large language model would be able to make me more productive and make me more effective by simulating things, you know, in a text driven way that we know it's good at. So, you know, that, you know, it's like a lot of things if the leaders are getting their hands dirty. I think it's a lot more credible if you're asking people to do the same thing.
- [15:35]
Benjamin Shapiro
I, I understand the purpose of the example and I just want to share when I've done something similar. I was going through a valuation exercise for our production company and was, you know, doing mock rehearsals for here's what I hear. Everything is, how do I have this conversation in advance to try to prove that I hear everything should be acquired by this other company? And every single time I was able to convince Chat GPT that it should buy my company for a hundred million dollars. The valuation of my company is nowhere near a hundred million dollars. Like it's an absurd premise. So how do you're going through and testing these experts?
- [16:12]
Ariel Kelman
I like this example because I think, I think this will kind of reinforce maybe something I was talking before about the importance of context and data. When you did that, did you first feed in to ChatGPT, you know, all the details around your company, what the revenues are, you know, how you think about valuation. Probably didn't, right?
- [16:34]
Benjamin Shapiro
I did. Oh, you did. But. But I also positioned the acquisition talk to be quote a strategic and talking about podcast os, the methodology which we use to produce our content and how it provided incremental strategic value. And so I think that the acquisition should be on a value exchange, not on a multiple of revenue or other traditional. And of course it agreed with me and then we moved on because it wants to make me happy.
- [17:04]
Ariel Kelman
Although, you know, an AI company thinking that an AI company should be valued highly doesn't entirely shock.
- [17:08]
Benjamin Shapiro
Yeah, maybe I should sell the company for $100 million. Maybe you're right. But that said, I want to talk a little bit about what you're seeing in terms of the actual numbers. Once you have your, you know, your senior executives playing around with your AI and navigating your change management, what are you seeing in terms of the hard ROI numbers from companies that have successfully deployed agents?
- [17:30]
Ariel Kelman
Well, let me maybe start with what we've done. So you know, the first Agent Force project, I'll kind of go outside of marketing for a second. I'll come back. Was on our website for customer support. So we created an agent on help.salesforce.com so we could help customers resolve problems they're having with their Salesforce deployments. I think customer service is one of the most obvious agentic use cases. And you know, in the past year we've had over two and a half million agentic conversations for customer support at the same time, 1.6 million with people. And, and it's really freed up our human agents to deal with the more complex issues. And Overall we've resolved 77% of all the customer support cases using Agent Force. And then that's let us save over $100 million. And part of that is we were able to redeploy some of these customer support agents to be field cert, forward deployed engineers that essentially are going and working with our customers on their own deployments. So building more customer success instead of fixing problems which is a, it's a career progression, they often do anyways, but it lets escalate some of them on that faster. So that's on the customer support side marketing, we put a agent on our website to answer people's questions. You know, you can go and say, hey, why should I use Salesforce and the Data360 product as a CDP instead of, you know, segment or Adobe, Go give You product comparisons or you can ask any kind of product question or say give me, transfer me to a salesperson. Now the interesting thing is at first we saw less leads going to sales. Now is that bad?
- [19:24]
Benjamin Shapiro
Well, maybe, depends how how sales is working.
- [19:27]
Ariel Kelman
But, but you know, as you can imagine when you're letting people click connect, connect me to a salesperson on your website, some of those are useful conversations for sales and some of them are not, you know, a bunch of them are customer support halls. So our hypothesis was we could drive a, you know, basically have a smaller quantity of higher valued leads. And so we measured it in terms of total pipeline generated from the website as a, as a source of sales pipeline and we've been able to see a 20% increase in sales pipeline from using this agent, our website and then lastly lead follow up in general. So we route a portion of our sales leads to an agent first instead of sending them to the, the human salespeople. And that allowed us to work almost 200,000 more leads that we previously just like, you know, they scored too low so we wouldn't follow up on them. And that generated $27 million of incremental pipeline.
- [20:30]
Benjamin Shapiro
Yeah, I would be remiss if I didn't bring up qualified, which I guess the acquisition isn't completed. But the AI driven SDR tells us a relatively similar story which is you can implement and train an AI SDR to work nights and weekends. Right. The times when it's harder to staff and manage a chatbot. And it seems like that is essentially casting a wider net. Do you think that's sort of a similar story with customer service and your AI chatbot on the website? Is it that it's catching more leads or is there actually an improved conversion for the leads that are coming in that would be those high quality top tier leads.
- [21:15]
Ariel Kelman
Well, it's catching more. Now here's the interesting part is that there are some people, if the websites do doing a better job of answering people's questions and actually giving value to your visitors. This is like a B2B. You got to set the context a little bit. This is B2B technology. The vast majority of our products are bought through a salesperson. If you're evaluating, you know, large scale customer service Systems, email, email, SaaS, applications, you know, you're going to go look at a bunch of vendors. It's not about capturing them on that website that moment and immediately converting them like A in a more B2C concept. So a lot of this is we're taking it from first Principles and not just blindly tuning conversion rates where if we can better educate our customers, we believe good things will happen. And so that's why, like, we, you know, we are seeing lower lead volume, but the conversion rates are higher, you know, as you suspected, because there, you know, if people can get simple answers or actually get directed to the right place, if they have customer support calls and not end up as a sales lead, that's good for our business. You know, it's good for our brand too. People are less frustrated, but in some ways, you got to change the metrics. I'll give you another example. So we have our website set up such that if we recognize who you are and you click on a pricing page, we'll send a slack message to the salesperson that says, hey, you have a customer here that, you know, they seem pretty interested in this product. You may want to like, reach out to them and ask them if they need any help.
- [23:04]
Benjamin Shapiro
Right.
- [23:05]
Ariel Kelman
That doesn't generate a sales lead. The marketers aren't counting that as pipeline from sales leads. So first I had some people complaining with like, oh, no, no, this is going to impact our numbers. It's like, change the fucking numbers. Like, you know, it's, you know, it's, it's a change management thing, right? Like, you got it. You got to be fluid with your, your metrics and make sure they're always aligned with the business outcomes you're trying to drive. So in this case, we're like, okay, how about we figure out how to measure the incremental impact of these Slack messages and add those onto, you know, the benefits we're getting from our different digital systems. So we adjusted all the metrics, but I mean, salespeople absolutely love it and the customers like it too.
- [23:44]
Benjamin Shapiro
Yeah, it's funny, it actually gets into back to the change management. There's this whole underlying problem with artificial intelligence and what I called the, you know, the AI revolution or, you know, the new world order of marketing. The tech stack which we're using is changing.
- [24:01]
Ariel Kelman
Right?
- [24:02]
Benjamin Shapiro
We have to learn how to use a new tech stack. The way that we evaluate everything, the buyer journey has dramatically shifted and all the marketing channels are changing as well. Right. People aren't relying on Google search as much. I'm not going to project. Google is dying. And obviously they're a big player in AI with Gemini as well. But the methodology of which we find information and then of which we engage with companies to go through the sales process is totally different. And it's changed in the last 18 months. So how do you think CMOs should restructure their teams to think about this agentic AI driven workflow?
- [24:39]
Ariel Kelman
Like yeah, what's funny is in some ways, like the fact that your customers may be getting information about your product from an agent first from AI instead of a human, drives you to make your messaging more human. And I'll sort of explain this. Like when we first put agent for us on our website, you know, we're testing it obviously and some of the, the date, you know, it's like you ask a question, you go, why doesn't this come back? And you go and tune it. And it turned out that the way you organize the information on your website, the more it looks like the way a human would ask questions, the better the LLMs understand it, the more that you make it like an org chart the worst. So for example, if you basically have your pricing information only on every product's pricing page versus having a document that clearly explains how your company approaches pricing and then gives the examples, the latter is way better for, you know, AI SEO. And also, which, which is also how our agent would work as well. And there's a lot of cases like that we're unorg charting our content and putting it more into plain English increased it improved our AI SEO results. So it's, it was, it ended up being good for the product marketing team because it kind of reinforced a customer centric way of communicating, which I thought was kind of ironic.
- [26:10]
Benjamin Shapiro
I can't help but laugh because the last meeting I had with a friend of mine who I worked with at eBay 15 years ago, I believe the company is called Everything Machines. But I'm sorry, Jeff Rain, if I got the name wrong, is exactly this, which is the text and language which you use on your website to be public facing is not necessarily the same text and language that you would use to feed to an AI agent. And so essentially you need to have two separate sets of texts and two different ways for agents to collect and analyze them. And they are essentially making the back of the canvas for your website, which is here's all the information that we're feeding to the LLM so it has all of the details. Because humans only want a little bit of information and the bots want everything and then they synthesize it down and communicate it.
- [27:01]
Ariel Kelman
We've done it by putting like really long FAQs for everything. But I, I would argue, like, let's say you didn't have all those FAQs and you could only do one web page. I think that it would be better for the customer if you did the same content that was more readable by the, by the LLMs. Because in many cases we have this tension between, between you know, clear communication and like conversion rates and brand and how you want it to look. And again, I'm just saying the B2B context where immediate conversion is less important than actually getting your ideas into the brain of the website visitor. I personally believe we could veer a little bit more into clearly explaining our messages.
- [27:47]
Benjamin Shapiro
I could look at 10 B2B companies taglines and not know what the company is because the wording is so fancy. Right. And nobody just wants to say we're another CRM, right? We're a little different. We're a cdp everybody.
- [28:01]
Ariel Kelman
Yeah, you got to have a functional description. I mean this is another AI use case for us where we, we're using a, a product writer for some of the, the generative AI. And you know, we set it up so that I have like a 15 page product messaging kind of bible. This is like here's how we think about product messaging and one of those rules you just reminded me of is, you know, always start with a functional description. People you, and it sounds obvious, explain what your product does first before you talk about the benefits. But you know, we often fail to do that a lot in marketing. And so when people are writing, are writing and editing messaging for our web pages, copy for emails, we have writer go and check it against those product messaging guidelines and reject, say sorry, like that's not going to work. You introduced a product but didn't say what it does. So I think this can really help us scale.
- [29:01]
Benjamin Shapiro
So talk to me about your advice to the marketers. There's thousands and thousands of marketers that are using Salesforce and they're going through this transition. And honestly, Salesforce is, you know, one of the things that keeps them set in their ways. I'm sure lots of people have been using the platform for a decade, but now they have to think about marketing in a completely different way. What advice do you have to them to make the transition to going from sort of static AI pre AI marketing and outreach for established companies to this new way of working.
- [29:39]
Ariel Kelman
Well, we're doing a lot of work to make that transition easy by building the AI functionality directly into the product. From something as simple as like summarize this opportunity for me for our salespeople to in our marketing cloud product we've added something called two way email. So you think about sending email campaigns with just a static offer as opposed to Sending an email campaign with an offer that says, I'm an AI agent, you can ask me any question as well, and then letting the customers reply to the campaign and engage with it so they can be more interactive as opposed to just spraying out all these static offers. You know, we think that it should be an evolution, not a revolution. Now people are thinking, and so we're trying to architect it to be simple and I think on the, you know, the SDR agent side as well, letting people set those flows up in a very seamless fashion so it's integrated with the rest of our platform.
- [30:47]
Benjamin Shapiro
I could not agree with you more in the sense of evolution over revolution. And I personally, and I think that there's probably a lot of other marketers feel this sense of anxiety of needing to rip and replace everything that we've done all at once and start a new system, as opposed to evolve the existing systems we have. Not every platform is salesforce and is taking the agentic first, you know, help you do its workflows. And so do you see the, the sort of SaaS business changing? Do you think that there's going to be turmoil around SaaS businesses that aren't evolving to be agentic first?
- [31:27]
Ariel Kelman
Yeah, the. For SaaS, the, the successful SaaS companies are evolving their product functionality to include gentic workflows in every portion of their products. And this has been a key part, you know, really the core part of our product strategy. And I think we're in a good place. We have the, you know, what I consider the hard part of the infrastructure, the data and the context and integrating that with, with the large language models in a trusted and secure way. This is this kind of the challenge with AI in the CRM world is the data that it needs access to is both, you know, off is for sure the most sensitive data in most cases for companies, but also in many cases the most complicated data. So finding easy way and a secure way to feed that in, to get that context in every single call to the LM is absolutely critical. But, you know, we've seen some companies try to say, oh, I'm going to throw away my enterprise app, train an LLM and all the data, and that didn't work. And what people are finding is that this data, the security, the deterministic workflows they built up in their company are not an alternative to AI, but a foundation to doing AI in a way that actually works.
- [32:48]
Benjamin Shapiro
Yeah, that's interesting. I feel like there's not only evolving your existing stack, but also a different mindset. That goes into taking the stack and then enhancing it and sort of supercharging what you've been doing for years. It is a very confusing time for marketers. I want to move on to our lightning round where I'm going to ask you a couple of questions around the rise of agentic marketer and a little bit about your career as well. Are you ready?
- [33:14]
Ariel Kelman
Yeah, I am ready.
- [33:15]
Benjamin Shapiro
All right, here we go. You've worked directly with CEO heavyweights Mark Benioff, Andy Jassy, Larry Ellison. What are the traits that they all share?
- [33:25]
Ariel Kelman
I think very high levels of intellectual curiosity and eagerness and ability when they see new things to, like, jump on them and ask question after question and question until they fully understand it and can adapt to all these new things that come at them all the time.
- [33:44]
Benjamin Shapiro
It seems like there's a personality trend that is similar between the three of them. And, you know, the. The term that comes to mind is alpha male, which may have a negative connotation, but, like, people tend to see those three leaders and want to follow and believe what they say. Am I interpreting it right? And why do you think that is?
- [34:06]
Ariel Kelman
Well, I definitely agree. You gotta be. You gotta be inspirational as a leader when you're selling change and when, you know, if you're running a tech company, what's important now is very different than what was important a year ago. And what will be important, the first future. So you have to take all these new things and help your company understand, you know, with the right level of. Of inspiration and leadership and, you know, leading by example. I think, you know, all three of these, these. These guys are really great at it.
- [34:37]
Benjamin Shapiro
Let me ask you the opposite question now. What's the difference between the three of them?
- [34:42]
Ariel Kelman
Oh, I don't know. We've talked for hours. They all have, you know, they all have their own personality. I get asked a lot about, oh, you know, wow, you know, the team struggling on this thing. Oh, that CEO is really demanding. And I think the one, you know, the one constant in, you know, B2B technology is CEOs love to torture the marketing team. They just have very different torture tactics. In the end, it's all from a good place, you know, because you can't communicate effectively, you're never going to succeed in these technical pursuits because at the end of the day, you're building technology. That's hard to explain, and that's what the marketing team is there to really explain at scale.
- [35:24]
Benjamin Shapiro
All right, I want to move on to our next question. What's the biggest misconception CMOs have about what AI agents can actually replace.
- [35:33]
Ariel Kelman
Today, I think the biggest misconception is an over focus on using it to save money and using it to replace people. You know, maybe I'll just put on that last one of that. What we're trying to automate are humans as opposed to thinking at a more granular level, that what we're trying to automate with agents are tasks and those tasks often span multiple people. And that if you can automate these tasks, you make your employees so much more productive and they can get more done more quickly.
- [36:09]
Benjamin Shapiro
I understand the idea of automating tasks and one of the beauties of LLMs is they can shift with no sort of problems between, okay, you're a product marketer, now you're a designer, okay, now you're a copywriter. But the, the concept of automating humans is a little fuzzy and confusing to me. Explain what you mean.
- [36:32]
Ariel Kelman
Well, I mean people go, oh well, you know, which jobs can we replace with AI? Okay, oh this job. This, oh this. No, no, no, this job we could never do because we can't have, you know, an AI do all this stuff. It's just, it's the wrong level of abstraction. Because if there's a certain job that has, let's say 50 things that it does, any one of those that you can automate with AI, you are increasing the productivity of that person. And when you increase the productivity of that person, you can do many things as, as a business, you can give that person better work life balance by letting them not have to stay up all night for a crazy deadline, which we have in marketing a lot. You can allow them to say yes instead of no more often. Right? Marketing, we're always asked to do like almost always three to four times as much stuff as we have, you know, resources to do. Amen. Okay, so you can do more things in the same amount of time or you can do the same amount of things with, with less people, of course. But like what I've seen mostly is in, in the vast majority of cases across marketing, the way that people are, are cashing in these productivity benefits is doing more with the same amount of people and doing it more quickly and then trying getting a little bit of burnout protection with all these date driven deadlines that we get.
- [38:08]
Benjamin Shapiro
Yeah, I feel like the biggest misconception for me. People fear, fear around the new technology as if it is going to replace them as opposed to thinking what an opportunity it is that they now have the ability to have essentially super Intelligence, which I know is a broad term that everybody's throwing around right now, but we all can get access to so much more information and have it synthesized to be right size for what we can interpret so much faster than ever before in human history. And people are sitting here saying, but it's going to take my job as opposed to this can make me better at my job. And so there's a mind shift around what the possibilities are that I think.
- [38:50]
Ariel Kelman
Is, yeah, that saying that's become kind of famous now is like, you're not going to be replaced by AI. You're going to be replaced by someone who's great at using it. AI.
- [38:57]
Benjamin Shapiro
I could not agree with you more. Okay, let's move on. What's the most dangerous thing a marketer can automate without human oversight?
- [39:04]
Ariel Kelman
I, I mean, there's a lot of things you can think of, but maybe I'll sort of reframe that to say the things where it's most important to have a human in a loop. I, I, I think I, I think messaging, like, you know, we do a lot of work on creating derivative assets with AI, and I think the mistake is to think about it as a message generation, you know, factory, as opposed to thinking about it as it's going to do 80% of the work for you to create new messaging. So it's like you got to read it.
- [39:35]
Benjamin Shapiro
Yeah. It's hard because it spits out so much information so quickly, and you actually have to read it.
- [39:40]
Ariel Kelman
People just, Yeah, I mean, look, I tell my team, like, don't ship any messaging is done, whether email, website, whatever, until you read it out loud to another human. And does it sound like something you want to say that doesn't go away just because we can use AI to help us generate five times as much stuff in the same amount of time?
- [40:00]
Benjamin Shapiro
Yeah. To me, creativity is, that's my answer, is that you can't automate the uniqueness. Right. That the human thought just fundamentally, large language models are great at predicting what humans would say in this scenario. It doesn't mean they're creating something that new. And so there is that initial spark.
- [40:19]
Ariel Kelman
That actually don't use it to create the spark. Use it to bring the spark to life.
- [40:23]
Benjamin Shapiro
Yeah. To, to fan the flames. Yeah. All right, next question. In five years, what will marketers be embarrassed that they used to do manually?
- [40:33]
Ariel Kelman
I mean, I'm gonna say spend large amounts of money on traditional filmed TV ads.
- [40:40]
Benjamin Shapiro
Really? Tell me more. Why is that the thing that's going away?
- [40:44]
Ariel Kelman
I, I think that what you, what you can do with these video editing tools is amazing, I think. And it's just, it has gotten exponentially better like every six weeks.
- [41:00]
Benjamin Shapiro
Give, give me an example of the, the current state and, and what makes you think that video editing is the thing that you think it's.
- [41:08]
Ariel Kelman
And it's not say video editing, it's, it's production. It's the ability to take stills or short video clips and, and make large videos out of it. But we've just been able to generate things that are incredibly high quality that previously would have required, you know, a massive amount of people and time and money. Now we're still, we're still, you know, to be clear, like we're still hiring, you know, we're traditional film crews and we're, we're doing it the old way but it's getting to the point when if, you know, if you say in five years, I think very few companies are going to be manually shooting 30 second spots.
- [41:56]
Benjamin Shapiro
Yeah. Tell me a little bit about the tech stack that you're, you're using and what's caught your eye.
- [42:01]
Ariel Kelman
I haven't personally been using them. I've, I have in my head I don't have the brand names but essentially this chain together set of tools that will do all these things that used to be manually people doing them all before. But I think I recommend for future podcast I can give you. Our chief Creative Officer John Zismos has really been playing around with this a lot or you should have on some of the CEOs of some of the companies in this space because it's you know, and, and like other things, it's letting you do high quality things in many cases where it wasn't possible. So for example like next week we're doing a pretty big activation at the World Economic Forum Davos because they're using Agent Force to power their mobile app to help attendees schedule meetings, do briefings, all this kind of things. So all of our creative for different spaces and all the demos and functionality we, our team went and built a complete animated video fly through of all our spaces and the demos and all the signs. And we would have never done that before but it provided a really easy way to see one place how what we're doing in four different locations is all going to look and it was built by a video AI engineer we hired on his second day in six hours and fascinating. Never, I mean we never would have even thought of creating an actual video fly through for that. So it's like new things that are possible that weren't before. That allowed all. All the different leaders in the company be like, are we aligned on what we want all this to look like? Yes, we're alone.
- [43:53]
Benjamin Shapiro
Last question for you. What's the one signal that tells you a company is truly ready to start implementing AI agents?
- [44:02]
Ariel Kelman
The company is alive. They have a heartbeat, and they have people. I think everyone needs AI agents, but, I mean, it's hard for me to think of a company that doesn't need AI agents. I mean, I. I have to think that. Well, let's go back to what we were talking about before of data and the context of being really important for AI to operate in a successful way. And I think when you have a reasonably good data foundation in that area that you're planning to automate, that would provide the data that's sufficient to provide a human with the context to operate effectively, then you're ready for the AI agents.
- [44:43]
Benjamin Shapiro
Yeah. To me, you hit the nail on the head, essentially, which is you have a good data source, you have context, you have something you could feed to an agent. It's also an understanding of a process.
- [44:54]
Ariel Kelman
Right.
- [44:54]
Benjamin Shapiro
Agents are great about taking repetitive processes off your plate. So I think about what we do in. In, you know, production, we have plenty of agents. They're doing our guest research. Right. They're helping us write our copy. Like, same process, different information every time. Agents are wonderful for that. When you're creating new processes, sometimes they can struggle. Right. They're not creative. And to me, that is when you've got these repetitive workflows and you've got the right data to feed into it, you don't need a human to do the repetitive tasks all the time.
- [45:26]
Ariel Kelman
Yep. Yep.
- [45:28]
Benjamin Shapiro
All right. Well, Ariel, I appreciate you coming on the podcast. I appreciate you sharing your time and knowledge about the agentic workflows. Thanks again for being.
- [45:36]
Absolutely.
- [45:36]
Ariel Kelman
Thank you.
- [45:37]
Benjamin Shapiro
All right, that wraps up this episode of the Martech podcast. Thanks to Ariel Kelman, the president and CMO of Salesforce, for joining us, and.
- [45:45]
Scrunch Sponsor Announcer
A special thanks to Scrunch for sponsoring this interview. Scrunch is the AI customer experience platform that helps marketing teams understand how AI agents experience their site. They tell you how your brand shows up in LLM results, where it doesn't, and what's preventing your content from being retrieved, trusted, or recommended. And for Martech podcast listeners, Scrunch is providing a free website diagnostic that uncovers how AI sees your site, where you have content gaps, and how you're showing up versus your competition. To see how LLMs evaluate your site. Go to scrunch.
- [46:18]
Benjamin Shapiro
Com.
- [46:19]
Scrunch Sponsor Announcer
That's scrunch s C-R-U-N c h.com if.
- [46:23]
Benjamin Shapiro
You'D like to contact Ariel, you can find a link to his LinkedIn profile in our show notes or on martechpod.com of course, you can always visit his company's website. It's salesforce.com and if you haven't subscribed and you want a daily stream of marketing and technology knowledge in your podcast Podcast Feed, hit the subscribe button in your podcast app or on YouTube and we'll be back in your feed next week. All right, that's it for today, but until next time, my advice is to just focus on keeping your customers happy.
- [47:03]
Ariel Kelman
Foreign.
- [47:15]
Benjamin Shapiro
Thanks for listening to the Martech podcast, and I hear everything. Production Looking to launch or scale a podcast like this one for your brand? Then visit iheareverything.com.