Salesforce CMO Ariel Kelman

Ariel Kelman: Look, the environment CMOs are operating in now with all these digital tools and what you can do is fundamentally different. If you go back to the world where your TV, out-of-home, print advertising was the dominant way you communicate with people, yeah, your ability to really construct a brand story, your storytelling, your ability to come up with ideas that stick on a big, broad scale, that was the dominant factor in whether you're successful. But now it becomes crafting this customer experience strategy and then executing on it with a whole bunch of digital tools.

Greg Stuart: Welcome to Building Better CMOs, a podcast about how marketers can get smarter and stronger. I am Greg Stuart, the CEO of the nonprofit MMA Global, and that voice you heard at the top is Ariel Kelman, the CMO of Salesforce. He rejoined the company in 2023 after serving as CMO of Oracle and the head of worldwide marketing for Amazon Web Services. And before that, he spent more than six years on the Salesforce product marketing team. Now, today on the podcast, Ariel and I are going to talk about how AI has radically changed the CMO's job and will continue to do so, the challenges of accurate measurement, and why we can't afford to ignore qualitative data. This podcast is all about the challenges marketers face and unlocking the true power that marketing can have. Ariel Kelman is going to tell us how he did that right after this.

Ariel Kelman, welcome to Building Better CMOs.

AK: Hi Greg. Thank you for having me today.

GS: Sure, sure, sure.

GS: You guys just had Dreamforce two weeks ago, a week ago? It's been very recent.

AK: Yeah, week and a half ago in San Francisco. We had 45,000 of our closest friends here for a week, and it was pretty exciting.

GS: Yeah, it's crazy. I mean, I get a sense that's probably one of the bigger individual corporate, certainly for the marketing, CRM, and sort of customer experience kind of area. This is the biggest event going, right?

AK: Yeah, definitely. It's the largest B2B tech conference for sure, and we've been doing it for a long time. So it's just a great opportunity to get our customers together with our product people, with their peers. I mean with business technology, things just keep changing so quickly. I mean, you've seen all the change around AI, which we could talk about a little bit, but the best way we've found to get customers acclimated to how to get the most from this new technology is to just spend time with them, be able to show it, let them ask questions, and most importantly, get people evaluating our products. We say, look, go meet with people. Don't listen to our salespeople. Go talk to our other customers who've already used it before in a casual setting where you can ask them what works, what doesn't work. Now, that only works if you have great products. If your products aren't good, it doesn't work. It actually makes it worse.

GS: I feel like you've just given away the secret to the keys to the kingdom in some regard there, so I appreciate that.

GS: So listen, you actually just came on board there, too. You're the recent CMO, maybe been a year. It's been a little while, right?

AK: Yeah, a little over a year and a half, although this is my second time here.

GS: You're boomerang, right?

AK: Yes. And it's been amazing to come back.

GS: Yeah, why come back? I mean, what was behind that for you and even for the company, by the way?

AK: It was a time when I'd taken a little bit of a break from working at large companies to work at a startup, and I was sort of itching to come back and work at a large company again. And just so happened Mark reached out to me and said he was looking for someone. So it was... sometimes the stars align. I just like to say can't complain about timing. Sometimes it's good, sometimes it's bad. And I thought it was a great opportunity. I loved working here the first time, and I always thought it'd be great to come back at some point.

GS: Yeah, yeah, yeah. Listen, you've done a couple of other stints, too. I think you were CMO within, I mean, you've been a CMO twice at least elsewhere, right? I mean, one was Oracle and then Fireblocks.

AK: I left Salesforce in 2011 to run marketing at AWS and really had an opportunity to really build up a marketing team from scratch. I think then we had maybe 20, 30 people, and that company really grew a lot. So that was a great experience to join the team there and kind of build out marketing the way we wanted to make it. It's one of the great things about building from scratch is you don't have the legacy. But I did that for about eight and a half years, and then I moved to go run marketing at Oracle for about three years.

GS: Yeah, so you've got longstanding experiences to sort of build these big enterprises. Ariel, can I ask you a funny question? So listen, MMA is a Salesforce customer, have been I think since I've been here, but the MMA works on behalf of big enterprise CMOs. My board chair is the CMO of AT&T. My next board chair will be the CMO of General Motors, right? So you get a sense of this thing.

GS: What is the role of Salesforce to big enterprise CMOs? I understand how you do it with smaller companies. I certainly get it from a sales, but does it also relate to those big enterprise CMO customers?

AK: Yeah, absolutely. And I spend a lot of time with the CMOs of our large enterprise customers. And when I talk to most CMOs about technology, if you're a CMO at a large enterprise, your martech stack is one of the most strategic things that you have. And I'd say one of the things you're almost always fighting with because no one company has produced the end-to-end, completely packaged, easy SaaS experience marketing stack.

GS: Not yet.

AK: So usually what most large enterprise CMOs are dealing with is a combination of some SaaS technology and custom-developed software. And I mean, we're in that case. I can talk about it if you're interested. But for a lot of our customers, they want advice on how do I leverage marketing technology to be most effective with my objectives? And I think over the past couple years we have this product called Data Cloud — which is a customer data platform — which has really made a lot of our CMOs... I wouldn't say more strategic, but a lot more central to the company's overall technology strategy. Because as these technologies have evolved, having that central database of record of all of the interactions your customers have with your company, having that in Salesforce in our Data Cloud product really ends up being amazing because our customers can activate on that data. They can give all their employees access to understanding every touch point a customer's had, and then they can trigger activation.

So for example, one of the things we've done is if a customer will come to our website and look at the pricing page of one of our products and we can resolve that website visitor back to a contact in our database, we'll then go send a Slack message to our salespeople that goes, "Hey, one of your customers is looking at what the pricing is for this product. Go give him a call and see if he needs any help." And so being at the heart of that, and for the same thing for customer service people to understand what marketing events or what the customer has done on the website, having that single source of truth ends up being incredibly valuable. And that's typically a system that the marketing org and the CMO is controlling because a large portion of the value accrues to marketing.

GS: MMA has done a lot of research where a series of think tanks delve into different important areas where we think there's the biggest opportunity to create transformational change for marketers and CMOs. And we've done a lot of work around marketing org, and there's a thesis we have. We've not proven this one out. We've proven a lot of other things. We haven't proven this one out, but there's a thesis based on some research we've done that says that brand— I mean, most CMOs who got to the job of these big enterprise ones got there through brand, they got there through a Procter & Gamble style of marketing, all around brand marketing.

AK: B2C CMOs.

GS: B2C, B2C. Clearly, yes.

GS: And we actually have a thesis, though, that the winning marketing strategy of the future is customer experience. And I've actually said to some of my CMO board members that I think the job of the MMA is to convert those brand-oriented CMOs to customer experience CMOs. I have to assume you have an opinion about that.

AK: I like that a lot. I'll say with an initial caveat, I'm a B2B CMO. I work with a lot of B2C CMOs, and I think I totally agree with what you're saying. I mean, it goes along with, look, the environment CMOs are operating in now with all these digital tools and what you can do is fundamentally different. If you go back to the world where your TV, out-of-home, print advertising was the dominant way you communicate with people, your ability to really construct a brand story, your storytelling, your ability to come up with ideas that stick on a big, broad scale, that was the dominant factor in whether you're successful. But now it becomes crafting this customer experience strategy, then executing on it with a whole bunch of digital tools. The thing I find interesting about all these digital tools is there's two parts of it you have to be great at. It's building the tool set and then using it. So I've talked to some CMOs where they say, "Hey, can I get your advice on something?" I'm like, sure. They say, "Well, my CTO just finished deploying your Marketing Cloud, your Data Cloud, CDP. We have the ability to do machine-learning-driven segmentation, activate programmatically on LinkedIn and Google and Facebook and all these ad platforms. So we have this amazing digital marketing machine. What should we do with it? What's your advice?" And I'm like, I'm happy to spend some time talking with you, but I don't really know your business.

And it's not just that you know your business, it's understanding the intersection of how your business is going to use marketing and then what's possible with these tools. In some ways, for some CMOs that are just pure brand CMOs, it's like a radio on TV problem. The TV first came out and they're reading radio broadcasts. You have to adapt how you work to the new media form. So I fully agree. And I think just maybe one other thing on customer experience: I think it is a good way to kind of explain this because it talks about how marketing really is transcending the marketing department. Because when you think about your job in marketing is to do marketing, it's like, okay, well, I'm going to do promotions to people through the channels I control. But if you think about customer experience, you're thinking about how do we promote our products to the right people through all the channels? So let's say for example, you're a company that's going to go do, let's say you're going to go build customer service agents with our Agentforce, which is technology for AI agents that can autonomously help your employees and customers and take action in some pretty amazing ways.

If you're a CMO and you're like, our customer service team is doing some sort of AI agent deployment, that doesn't affect me. Well, you're not thinking in a customer experience mindset. You have all these new marketing opportunities with this person, with the customers. Because if someone, let's say you're a retailer, someone calls up your call center or talks to an agent and says, "My sweater's too small, I need the larger size." Well sure, we should ship them out the larger size sweater, but we also want to say, "Hey, did you realize that we have a hat that matches that sweater? Do you want to buy that?" And the reason why they recommended that hat is based on some segmentation logic that we've created in our marketing tools. You want to be able to inject that into your customer service experience. CMOs need to think more broadly about getting into every channel.

GS: Well, and they need to manage the whole of that journey. I mean, the way that MMA has often looked at customer experience is how do I both ingest the product itself, whatever that might be. And by the way, I think that even applied to Coca-Cola, and I'll explain, but also then how do I manage my ongoing satisfaction or experience with that product once I have it? So listen, my point to Coca-Cola: Coca-Cola said, well, we're a brand company — and they're a member of the MMA — and they go, we're a brand company. I go, but my problem is not disrespecting your brand, I've already decided to buy you. The problem is I can't get the shit in the house fast enough. It's too big and bulky. So how do you deliver? That becomes a part of that customer experience. If you can make it easy for me, I would double my volume with you all.

AK: Yeah, it's all interrelated.

GS: And then they start to know who I am, which is a really big deal.

AK: Yeah, yeah.

GS: Changes everything.

GS: Before I get to my key topic here around Building Better CMOs, I saw a quote that you made, I wanted to check this with you. You made a point as you were coming into Salesforce that you're there to reset KPIs. That's what I thought I saw. And it says here, "overhaul metrics, re-engineer attribution, get away from last click, and move to a deep learning model" that you also thought other B2Bs should adopt. We're big measurement people. So I'm curious what that means.

AK: So this is where I'm going to complain as a B2B marketer that it's so much harder to measure everything. I've spent a lot of time, as a lot of CMOs, debating and working on attribution models because ultimately you're trying to measure what's the impact on sales pipeline or revenue of your marketing programs. And I feel like how to do it right has sort of, in the most cases, been decided on or solved. We've gotten consensus on the right way to do it, which is to do a deep learning model to go back over a period of time... to get your data science team to go there, many different algorithms, but whatever. Some type of deep learning model to take in, as signals, all of the marketing touch points over the, let's say, some five-year period of time, four-year sales opportunities in Salesforce.

And to take in all the marketing touches and all the sales touches, and then to basically do a deep learning correlation on which types of programs at which point in the cycle have the most causation. And then to end up with a model that you can plug in a campaign ID and get back out, in our case, what we call marketing-driven pipeline. And so that would be the opportunity value dollars that we are attributing as being driven or caused incrementally by doing that campaign. But if you look at it from a campaign perspective, we did this executive event, it outputted $5 million of marketing-driven pipeline. But if you look at it from an opportunity perspective of we have, let's say, the Bank of America million-dollar opportunity that was created in January, closed in November, you can see that it was touched by, let's say, 75 marketing activities and 200 sales activities, and our model will literally go attribute an amount that million dollars of opportunity value to all of those 200 touchpoints.

GS: Yeah, no, I mean it's in essence a variation of multi-touch attribution, which by the way, I happen to have been the co-founder of multi-touch attribution 20 years ago, funny enough. But you're really moving away from, I mean, I guess your big advocacy there is move away from last-touch attribution.

AK: And move away from last-touch attribution and then also just moving away from humans deciding how to weight different parts of it.

GS: Agreed. Yes. Yeah. You've got to build up models to become predictive about that.

AK: And so when I first got here, you go around and talk to people, what's working, what's not working, and one of the areas people were unhappy with was with this attribution model. They're like, "I don't trust the model," especially our salespeople. And I think for CMOs, if you're starting a new role, this is one of the biggest things I recommend is you've got to get the measurement right first. Otherwise, you're never going to get consensus in the company on whether anything you're doing is working or not. So, luckily when I started here, they were already about halfway through that project, but I focused a lot of energy on that initially to make sure... do workshops with all the sales leaders and to say like, here's some choices we have on the model, and really bring them in the tent and say, we're only deploying this if you guys say you believe it. And showing them here's the 50 ways we're being conservative to not take too much credit.

GS: Yeah. I don't know about B2B as much, but I'll tell you, I mean, I've done all the research around click attribution over the years, and I mean most marketers don't understand that there is zero relationship, zero relationship between last click and a sale. Zero. Yeah, I've done all the research, I've studied this for 20 years. I can tell you it doesn't matter. And yet the industry sort of relies on kind of easy, I think sometimes, unfortunately. They think it's directional, but it's really not. It really just annoys me. So good for you. Yeah, it caught my attention when I saw you said that.

GS: It's time for our favorite question: what do CMOs get wrong about marketing? Let's find out what Ariel Kelman thinks in a minute. But first, let's take a quick break.

This is Building Better CMOs. Let's get back to my conversation with Ariel Kelman, the CMO of Salesforce.

Here's the big question I always ask people. So MMA, as I mentioned, is a nonprofit trade association. We're here to better marketing and CMOs. We're here to help raise the stature and gravitas of marketers and make them more important to the companies that they lead. And so in that journey, we're always looking for places where we think that marketers have maybe a misunderstanding. Maybe they don't have the knowledge, maybe they've just not been able to do the learning to really help advance the role of the CMO and its contribution to the business. So the question I have is what do you think either marketing maybe doesn't get, doesn't fully understand, maybe doesn't appreciate, maybe doesn't prioritize enough in your opinion that we should be more focused on? We'll take off the mix here, get measurement figured out first. I think that should be the headline of the show here at some level. That's a very basic one, but what do you think we're kind of missing, Ariel?

AK: I think part of it is what we talked about earlier around thinking about marketing as the customer experience versus just traditional marketing and the example of thinking about customer service as a marketing channel, and how you can use that as a channel to get your messages out and to have impact is pretty key.

I think other things where marketers are missing the mark — I'll go back to measurement a little bit — is only thinking about hard science measurement. The way I think of a hard science measurement — I'm just making up that word now, it's probably not the right one — is traditional metrics of adding things up with all of your data and ignoring the qualitative, anecdotal data points. Because, look, we deal with a lot of uncertainty and things where causation is difficult to prove a hundred percent. So let's just take an example of a large conference like a Dreamforce or, when I was at AWS, re:Invent conference. You'd say, well, what's the ROI of it or how important is the executive summit program that you're doing inside of it?

You can track the people that went there, correlate them with opportunities that were created there. Look, definitely go and do all that, but what people don't do is they forget the easy, non-hard science, math-oriented stuff: why don't you go take 20 customers that have executives that were there that you actually trust as they're real customers, they're smart people, and go ask them, "Hey, you attended our executive summit, and I saw you bought some stuff. You bought this big project afterwards. Did what you learned at this event make a big difference? You brought five other people to this event. Did it make a difference?" Or think about content marketing attribution. You have this white paper, go tell your team to go interview 20 people that read it and ask them, did this meaningfully help your ability to understand whether our product is different? And people kind of skip over that. They get enamored with like, I have this great Tableau report that has this data in it. Great, but you get an opinion from someone you trust that you know has the same needs, desires, issues as other people.

GS: You didn't say this, but I wonder if you're getting at kind of the falseness of quantitative research sometimes, which is that it tends to go for an average. And as my research guy says, "If you're in a bathtub and your hair's on fire, on average, you're fine."

AK: Yeah, no, no, it's the same ballpark, but the other area of this is people operating with false precision. If you're trying to pull out very specific insights from something where the certainty over the data isn't a hundred percent anyways... For example, if on your website, you can only identify 40 percent of those people on the website and of the 40 percent of the people that you can identify and track, there's another 20 percent of data you're missing because the ad blocking and privacy filters are filtering those people out, and then you're trying to go draw these very sophisticated conclusions off of some small subset of the data... you got to hold on. I think you're making conclusions at a level of granularity that outstrips the granularity of trusted data.

GS: And is that, in your opinion... I think what I hear you saying is that it misses the sort of nuance of what a customer might relay back to you that captures new essence. In some regards, it's kind of like, did we ask the question the right way in a survey to begin with or do we hear sort of a different opinion when we ask them how the feedback was? I get that all the time.

AK: Yeah. Well that also makes me think of my other advice of you can never spend too much time reviewing the survey questionnaire...

GS: Oh my god.

AK: ... before you do the survey. Something I always tell my team is do not delegate responsibility on reviewing that survey, especially these large brand surveys and for any programs that you're doing, and you've got to have people that are good at this. It is a science.

GS: I agree. You know what I do all the time is I make — my team's going to laugh when they hear this — I make them put an open-ended question at the bottom of every page because give the customer opportunity to tell me what I didn't figure out how to ask them properly.

AK: For sure.

GS: A hundred percent. Hey, listen, just funny enough, given this point about really focus on or be aware or don't ignore the qualitative, that's really what you said, just to be clear. Do you have an example of where maybe that's come up for you, something you've learned? I mean, it feels like it's a lesson you had.

AK: Yeah, I mean when I was at AWS, it came with our exec programs. I mean, this is one of the Amazon cultural things is they're very data driven, but also they place a lot of value on customer anecdotes in terms of go talk to the customers. And so with executive programs, they're hard to measure for many reasons, but you have a small volume of it. Let's say you go and do some executive event series where you have 30 people come to each event. It's not like if we want to go draw conclusions from, let's say, Dreamforce, we had 45,000 people. You've got a lot of data to work with there. So we were struggling to sort of think about the ROI of a program and Andy Jassy, who was running AWS at the time, he's like, have you talked to any of the customers? I'm like, I haven't. He's like, call a few. And so we're like, okay. Of the 300 people in this program over the year, we went and called up 20 of them and we got pretty consistent information: this part of what you're doing is super impactful to getting us to do more. This part irrelevant, waste of time. This part, I don't care.

And so it's like when we looked at a combination of the qualitative information and the data together, it gave us a good picture. This is kind of why I recommend when there's areas of your business that you can't measure precisely, if you come up with three different imprecise, imperfect measurement techniques that are very different, then you can see it can actually end up being useful. If they're all aligning, your chances of them being statistically flawed in exactly the same way are pretty low.

GS: Right, right.

GS: So listen, Ariel, let's shift here a little bit. Although this is very interesting, and I do actually have other questions about B2B I might come back to, but can we talk a little bit about Agentforce? Because I mean, listen, I think any of the big tech companies that are launching AI products we should pay attention to. It feels like you guys, though, took a different approach there. It feels like Salesforce is doing something different than I'm hearing from anybody else. So can you just walk through for the listener, what is Agentforce and what is it meant to accomplish? And just give some orientation, especially for CMOs.

AK: Sure. Quick functional description is Agentforce is a platform for building and deploying AI agents that are designed to both work directly with customers and to augment the capabilities of your employees to make them more productive and effective. The way I think about it being really different is we kind of evolve with generative AI... You think about basic LLM or the co-pilot, which is something that's passive and static. It's sitting there on the side where you can ask it questions, and it will give you an answer.

So for things like how do I do something or do we have a product that does this? When you have a big catalog, doing content RAG, it's useful. It's there to ask questions. But there's two things that were missing that we felt required a new paradigm or a new approach to AI.

One is the desire for AI that is proactive and autonomous. So doing things without you asking it a question. And then secondly, that instead of it just being an answerer of questions, that it will take action and do things for you. So I'll give you sort of the difference in a customer service example.

You think about the co-pilot approach: I'm a customer service rep. I'm talking to someone and they say, "My machine isn't working. It's doing these three things." Well, they can have their co-pilot type in, their machine is doing this, it's spinning, it's shaking, there's smoke coming out of it, whatever.

And it'll tell them, ask the customer this, maybe this is what's wrong with it. As opposed to an agent approach of saying, "I'm listening to your conversation with this customer, and I'm 90 percent certain that there's a part that needs to be replaced. Please let the customer know there's a good solution to this." Lets the customer know the solution and then the agent will go and say, "We have this part in stock. We can ship it out. Go tell the customer that it will arrive at their house in two days. Would you like me to send them a confirmation email?" And so this is autonomously doing things and taking action of checking inventory and placing an order and generating a follow-up email. Think about how much easier that makes it for that customer service rep that they can focus on actually having empathy, helping the customer with the problem, having the human-to-human conversation. All that sort of operational, menial stuff, that's outsourced to the agent. And then the really low-end calls of someone saying like, "Does widget A work with plug B?" That can just be an AI agent on their website or that someone talks to on the phone, and that can handle the case without the human.

GS: Just so we make sure, because as I think I mentioned, I have another podcast which is Decoding AI for Marketing. So are you able to comment, how big a deal are agents? For the uninitiated listener here who maybe hasn't really spent a lot of time on AI, what's going on with agents? Can you give perspective on that?

AK: Yeah. I'll call the marketing tagline because we're on a marketing podcast, we're inside the boat here.

GS: Yeah, yeah.

AK: This is what AI was meant to be. And what we mean by that is what do people want from AI? They want to automate, they want the AI to do things that previously required humans.

And there's a lot of ways we can think of how that would make it easier. And as I said, the difference between an agent and a co-pilot or just like a ChatGPT type thing is that it will work autonomously and proactively. And then secondly, that it can take action and actually do things.

And the thing that we found that week at Dreamforce and having over 10,000 customers create their own agents is that by combining Salesforce and an LLM — we'll let them use OpenAI or whichever ones they've used — that they can build these agents really easily and quickly. They don't need any new skills because all the really sophisticated stuff of taking action, it's already built into our platform. So let's take my example of ordering the new part and checking inventory. Well, our platform that billions of customers are using, if you want to let a sales rep or a customer service agent check parts inventory, you would let's say build an integration to an SAP inventory system.

We have this MuleSoft integration technology that a lot of our customers have used. Now, how do you give that functionality you've already built to check inventory to the agent? All you do is you say add an action, and then one of the types of actions is integrations. You pick this integration you've already built, which is an API, that's it. Done. So you have to do nothing different to let the AI agent use all of the workflows, all of the automated actions that people have created on our platform. And so they're like, "Whoa, that's amazing." As opposed to if you're going to go build your own agent from scratch, all of that taking action, you're all custom coding, you're writing your own stuff. And by the way, once you've built all that, then you go back and say, how do I make sure the right people see the right data? If you're a bank and you put all your customer information into, say, a custom-tuned OpenAI running on Azure, it's not going to know that I'm not supposed to show this data about Sally to someone that's not her financial advisor. Then you've got to go rebuild the whole sharing model and all the security rules. We have that all built in our platform already.

GS: But how is an agent different than a rules-based approach? You could set up the rules, say, do this, then do that, do this. Why is AI...

AK: Oh, yeah, yeah, no, that's important because in customer service for years we've had chatbots and we create these rules that go, if someone says this, do that. It's the fact that it's inflexible, it can't handle lots of scenarios.

GS: Can we agree on something? Again, we're in the cone of marketers, right? Just here for this one, right? Because I asked this question to a group of marketers the other day, they started talking about chatbots. There's never been a chatbot that anybody's ever liked.

AK: Yeah, correct.

GS: Yeah. Okay, good.

AK: People hate working with the chatbots because if the customer says, I want to return, and then the customer says, I want to give back this product to you, and it's like, "Oh, I can't help you with that. What would you like next?" Whereas that's the magic of you add an LLM in. Since all of our friends that have built these massive large language models, they've trained it on the internet, it knows that a give back, a return, however they call it in Australian, it knows all that stuff. And so you don't have to program it on that. You can basically, so you create these actions on our platform on Agentforce to say, I'm going to create a return.

And you'll say, well, what can returns do? You'll assign it that look up inventory, ship new parts, give a refund on their credit card. But the way that the Agentforce knows to initiate the return business process is based on how you describe it. So when I create this action, I just said, here's all the actual mechanical things it needs to do, but then there's a description. You'll say returns, and you'll write a paragraph or two on what returns are. And if you go like, "Hey, guess what? Everything I've described works for everywhere except for New Jersey. If they're from New Jersey, tell them we don't do returns in New Jersey." You're writing it in plain English in there, so you don't have the flexibility and the rigidness of these linearly, programmatically built bots.

GS: And nothing against New Jersey, by the way. The agents don't know to go against New Jersey.

AK: I love New Jersey. It's great.

GS: Okay, just being safe there. Okay, good. So I guess the other question that I think the listener will have around Agentforce, it seems to be turning a lot over to the machines to kind of figure out what to do next, how to communicate. So how do you mitigate some of the— I'm sure this is another two-hour conversation, but how do you mitigate some of the risk from the AI making the wrong decision on your behalf in Agentforce?

AK: Sure, and this is obviously super important, and it gets to implementing this in the right way, but we've put a lot of guardrails in there. I think first of all, we have this whole trust layer that we've built underneath all these interactions that make sure that we don't have any offensive language, toxicity, but then also that inaccuracy is the thing people are most worried about.

GS: Or hallucinations.

AK: Hallucinations, yeah, exactly.

GS: People don't understand, they treat hallucination like it's a mistake, but hallucination is the point of AI.

AK: Hallucination is a feature of an LLM, but that doesn't work in a customer CRM context. I'll give you a perfect example. I don't know, six months ago I was talking to one of our customers, Heathrow Airport, about this and they're like, someone asks our customer service agent through Agentforce, when they ask it what kind of food should I get? It should be creative in what a recommends. But when it says, when's the flight to Milan leaving? It should not be creative.

GS: Not be creative.

AK: This is the most important thing that we make really easy is we're grounding the LLM in data. And so we make it very easy when you set these up to go tell Agentforce, these are the areas that are actual structured data. And this is where our Data Cloud product comes in. So like in this Heathrow case, they have an integration that lets our Data Cloud reach out to their operational system with all the flight times. So in what we call a prompt template, we'll go and say, when people are asking about flight information, don't worry, LLM, you don't need to make something up. What I want you to do is to reach out and make an API call on this other system and do a query and give them the result. And so you apply that forward to someone says, what's the balance on my bank account? Or am I eligible for an upgrade on my phone? There's integrations to systems that people have built that the agent is told in our platform, for those type of questions, make an API call and get that. But if someone says, which clothing items should I wear for a barbecue? Then it can start to be creative.

GS: Okay, got it. Got it. Listen, I appreciate your Heathrow. I'm actually in London right now, so I'm flying out of Heathrow tomorrow, so it's good to know that the AI's not running wild there, so thank you for that, I guess at some level. What else do you think marketers should understand about Agentforce? I'm going to ask you one other question here to close up in a minute, but what else do you think we've not talked on that would be important for them to get about this new development, this new application that I'm sure none of them have thought was even possible?

AK: I think what's exciting is the way that we've built it is we've built it as more of a platform than sort of a spectrum for SaaS applications of how much is it pre-wired where you can figure the last 10 percent. This is pretty open ended, so I would encourage marketers to be very creative in thinking about what are all the manual, repetitive tasks that they're doing that they would like to be automated by an agent that's sitting next to them and either waiting for them to ask to do something or, more importantly, that is doing things proactively based on instructions and a strategy that you've given them before. We haven't talked about this. One of our other agents is an SDR agent — a software development rep, the people that will qualify leads. So instead of scoring all your leads and then giving those to your software development reps to qualify, what Agentforce SDR does — and it's going to be available in about three weeks — is it will go email those customers and engage in a conversation with them to get more information, and based on engaging and engaging, it will decide when they're actually at the point when they need to talk to a human and then route them over. So you can give your salespeople, your human salespeople, a smaller quantity of much more highly qualified leads. It's almost like if your SDR team that before is just getting a list of leads, it's like they have their own SDR team working for them that's preventing them from talking to all the people that are never interested.

GS: Right. They're getting the Glengarry leads, I think is how that...

AK: Yeah, so we're very excited about that. My team's going to implement it, too. Pretty fired up about it.

GS: Hey, I'm just curious, where are you in Agentforce? I didn't even really think to ask, is it available now? Can people go into Salesforce right now and start to deploy?

AK: So it's in preview right now. We are releasing it out in the wild, generally available in the end of October. So later this month it'll be available for everyone.

GS: Okay. Okay. So listen, I'm going to take a real future state here in the last question for you, Ariel. Obviously if you have agents, the business, the corporation, Salesforce is providing agent technology to companies to use. I think it's great. I totally get how it makes sense.

GS: Okay. So listen, I'm going to take a real future state here in the last question for you, Ariel. Obviously if you have agents, the business, the corporation, Salesforce is providing agent technology to companies to use. I think it's great. I totally get how it makes sense.

What happens when consumers have agents, which is part of what's being predicted out there, that I will have an agent that will make the flight arrangements for me that would, as we talked earlier, will buy the car for me and take on all these actions. What do you think happens then to companies interacting with consumers? Do you have any view on that yet?

AK: Things are going to change. I mean, a cynical view would be like it's going to turn into the people that make tank armor and the people that make the projectiles that try and get in the tank armor, but I think you're going to have to adapt. If the consumers have their own agents, you have to adapt some of your policies to be a little more sophisticated. Right now, it's very reasonable, a lot of people do this to say, well, if people call to complain about something, refund them every time because you'll do the math and say, not that many people will complain because it's kind of hard to figure it out. But classic example with this is a price adjustment. So there was, "I'm mad because you put my product on sale two days later and I want a price adjustment." Give them a price adjustment every time.

Well, if every person has built in as a feature on their iPhone check for price adjustments, then companies are going to have to adapt what they put on sale and also the modeling they're doing on what their ROI from that sale will be. They'll say, well, okay, I'm going to assume that of the people we've sold to before, 20 percent of that revenue will decay or go away if we go and put this product on sale. So it'll be an adjustment. I think, look, it's all good because it's transparency. Anything that forces a more transparent relationship between companies and consumers, I think is ultimately a good thing.

GS: I can't thank you enough for doing this. I mean, listen, you've got a big job there at Salesforce. It's a big marketing department, isn't it? Do you guys disclose how many people are in marketing?

AK: We don't, but some days it seems too big and some days it seems too small.

GS: I think that would be every CMO I've ever spoken to. So there you go. Listen, I can't thank you enough for doing this. I really appreciate you taking the time here. I did think Agentforce was pretty interesting, and I thought the community should start to hear about that. There's a whole new world order coming isn't there?

AK: There is. It's a brave new world in a good way.

GS: Thanks again to Ariel Kelman from Salesforce for coming on Building Better CMOs. Check the description of this episode for links to connect with Ariel. And if you want to know more about MMA's work to make marketing matter more, visit mmaglobal.com or you can attend any one of our 44 conferences in the 16 countries where MMA operates, or really write me, greg@mmaglobal.com. Now, thank you so much for listening. Tap the link in the description to leave us a review. If you're new to the show, please follow or subscribe on Apple, Spotify, Amazon Music, iHeart, or wherever you get your podcasts. You can find links to all those places and more at bettercmos.com. Our producer and podcast consultant is Eric Johnson from LightningPod.fm. Artwork is by Jason Chase. And a special thanks to Angela Gray and Dan Whiting for making this happen. This is Greg Stuart. I'll see you in two weeks.