Bernard Slowey: Forget Chatbots! Smart Brands Are Moving to AI Agents!

23:50That was what we wrote. And so we taught it. We taught it to think. We were treating it like an old school chatbot. Don't do this. If that, do this. Right. Instead, it was like, no, no, you're smart. We just need you to think this way. And that solved that problem. And so that was a huge learning for me and my team. It's like this prompt design, this prompt engineer. And it's a discipline now that's in the industry. And it's like it's really, really need to think through it on what you write and how you write it. This is like such a new space we're in where people are still just trying to like figure out prompt engineering for a chat GPT.

24:22And they're like, oh, maybe instead of just like give me the answer to this, it's like I should say pretend you're this and think about it this way. like the more context. And so it's no surprise that, of course, you're like, oh, yeah, maybe we should do that here too. But it's such a fast paced world right now to keep up with. Are there any other funny moments like that that came up or just like things that happened while you were rolling this out that probably are relatable to other people who would be going through this process? I think maybe coming back to Lauren's question, which I didn't do a good job of answering on, it's like risk averse versus pushing.

24:51Like we definitely had like people internally. like I used to like we obviously Slack, Salesforce, it's a massive product for us. And so we all live in Slack. I would get Slacks every single day from people in the company giving me examples of where it didn't do a good job. A lot of the times they were people that were very knowledgeable in Salesforce products asking it a very detailed question. And a lot of the times the problem wasn't agent force. The problem is the content in the backend that it's trained on. If you don't have good content, right? Right. If you like it's only as good as the knowledge it has access to.

25:25And so like and people will go and they try it with ChatGPT and they go, ChatGPT gives me this answer. I say, yeah, but ChatGPT has access to the whole Internet. So it can pull an answer from Reddit. It can pull an answer from some blog somewhere. And the risk of that is hallucinations. Right. That's what scares companies on these agents is like. So ours is one of the one mentioned earlier on, but agent force you limited to your own content set. So with that, sometimes if it didn't have the right content, it didn't do a good answer. So we had to work on updating our content. I think it actually exposed to us areas where we didn't realize we didn't have great content.

26:00I mean, that's such an important factor, which I actually think a lot of people who are approaching or rolling out AI, building it into their systems, we all need to do it. And people are taking different approaches. But the thing I keep hearing time and time again is that your data is critical. The information that you're feeding to the AI is critical. And it's a very large and complex thing to think about because we have, it's almost like we have like all of our stuff is like strewn everywhere. And all of a sudden we have to organize it in a way where someone can come in and actually like intake it and use it.

26:38And I'm curious to know your thoughts on that piece. And like, how did you go about organizing your data so quickly to then allow AgentForce to use it? Or had you already had a totally clean house and none of that had to happen? No, no, I don't think anyone has a clean house. We had, we have an advantage at Salesforce that we have this other product called DataCloud. And with DataCloud, like people always ask me where we started with AgentForce. And what I tell people is like, it's like when you have a new employee start, what do you do? you give them access to data, right? You give them content, you train them so that they can start to be upskilled and be able to do their job.

27:16And so Data Cloud allowed us to take all of these different data sources, right? These knowledge articles, these products and actually hydrate them all into Data Cloud. And then that gave Agent Force this foundation. It's able to access all of this data in Data Cloud. That's the unstructured data. And as I mentioned earlier on, we're now going to give it access to structured data as well. So I think I had a bit of an unfair advantage where I had access to data cloud because I work in Salesforce. But when I talk to customers, I'm like, this needs to be part of the solution. Like you can use it without data cloud, but if you can have a repository that can contain all of this data and can bring it all in, it can normalize it, it can standardize it.

27:52Then you have a huge advantage as you go live in these agents. I mean, it's really the strength of your agent at the end of the day. Like as you were saying, it's not going out into the internet to find answers. It's using the information that you provided to it. And that's the beauty of it. And also it means we need to do that extra legwork to make sure that the agent has what it needs. We learned this quite in the hard way. So we're a couple of months in now and our agent force implementation still feels a little bit robotic. Like it responds to you. It gives the right answer. And so what we learned is what else do you do when a new employee starts in a service world?

28:31Right. Yes, you train them on the hard skills. Right. Like, do they have the information to answer these problems? but you also train them on the soft skills, right? The way that you deliver a difficult message, right? If you're a support engineer, like we spend a lot of times on that soft skill training. And so what we forgot to do with Agent Force is we trained it on the hard skills, but we didn't focus on the soft skills. And we talk about this as the head and the heart. We were very much focused on the head, the brain element of it and not enough on the heart. So now we're doing a lot of work where we're actually taking the training that we train our support engineers on And we're using that around some of the prompts for Agent Force.

29:08So just think about it as we're giving Agent Force access to that training. So it has much more of the soft skills on how it delivers its message. It's kind of crazy when you think about that, but that was being a big learning for us. The thing that I'm like kind of mind blown by is you're taking a training that you give to humans about soft skills and you're able to train an AI agent on the same content. You're just teaching it in certain ways. Like we always have this good example. Like if someone comes to us, let's say, with a severity one issue, right? Like let's just say Salesforce is down.

29:41Doesn't happen very often, but some companies can have things are down. The last thing you want is the agent going back and forth with you on potential ways to solve that problem. You want it to go, oh my goodness, you're having a Seb one. I'm so sorry to hear that. We're going to get you through to one of our support engineers and they're going, you know what I mean? Like that's an example where agent force needs to get out of the way quickly. but it needs to have a lot of empathy in that situation. Right. And so that's the example of what we mean. You don't want it to be like kind of robotic and try and answer that problem.

30:11It's like, no, no, get out of the way. Be very empathetic. And that's how you would teach a human employee to also react in that situation. One thing we haven't talked about yet is like the success that you saw once you implemented Agent Forest. I mean, one, I'd love it if you like if you had a before and after, because I'm trying to visualize like what it was like before when customers would go on there and ask a question, what it's like now to like really understand what it looks like and then what happened after how many cases are getting solved with agent fours like tell me a bit of the stats so this visual that you can see this was our health portal in i guess september last year right and so it looks like pretty much every single company in the world has some portal like this right you can come to it.

30:56You could search on it. You could access our content. It has videos, places to learn, et cetera, and you can go in and create a case from it. So that's kind of what it looked like back then. And so, as I said to you earlier, unfortunately, the problem with this is people struggle to find what they were looking for, right? Especially if you were having a problem, you need an answer quickly. You know, we hear that in all of our research with customers is like, hey, if I'm having an issue, you need to be able to answer me quickly, right? And so So let me now show you what this looks like now. So this is help.salesforce.com now.

31:30And so what you can see is it's a completely different experience. It's a full screen experience of AgentForce. So you can go in and ask it any questions I mentioned earlier. It doesn't matter whether you're logged in or not. You can engage with AgentForce. And what we've done as well is you asked me, like, how do we determine the success? Right. We believe transparency is key for trust. So what we've done is down the bottom, you can see down here, we actually have a live ticker to show how many AgentForce conversations. So since launch, you know, we've had 450 ,000 conversations with AgentForce.

32:01So since October the 11th, that's a pretty huge number. And then you can see, we also try and show our human tickets as well, because we want to try and show people. But what I want to show you is just something a little bit cool here now. And I go, you can click on this, here's how. And what that will do, it's going to show you all of our metrics and how we measure success. So this is our customer zero page. And you'll actually see there's a video there that some of me and my team are in to talk about our implementation. But what we show you is you can actually look at all of our data and you can go back to any date since we launch.

32:35What you asked earlier on is like, how do we measure success, right? And so what we look at is, think about this as a funnel, right? Someone has a conversation with AgentForce, right? Like it comes back with an answer and you go away. And we consider that resolved, right? So you ask it a question, you came back with an answer and you go away. You don't, when I say go away, you don't want to create a case with a human. And so you can see our resolve metric is like mid to high 80s, right? But we also get asked questions, but is that really resolved? Like, how can you determine that was resolved?

33:06Like maybe the question I just went away. And so what we've implemented now is we have something called customer confirmed resolution, which is in the flow of the conversation, we ask the question at the end of the conversation, hey, did this resolve your issue? Yes or no. And on that customer confirmed resolution, right now we're about 76%. So that's a really good data point for us that like all of these conversations that you're having at the highest level, it's probably about 80 something. And then customers are telling us it's about 76 when they answer that question. The other two metrics I just wanted to show you just so I can help explain it is we also look at this thing.

33:40This is the AgentForce conversation handoff. So you can see that 4%. So that is that week, 4 % of customers who engage with AgentForce got handed off to a human support engineer. So that was that kind of flow that I was telling you about. Like it's really critical to us that it's an easy flow. The other metric, which is interesting for us is we have this thing called abandon. And let me explain what abandon is, is you can kind of see the ask AgentForce in the bottom right-hand corner. So a certain percentage of customers click Ask AgentForce, they open it up and they don't ask it a question. And we're really interested on this and why is that happening?

34:15And we think some of this is to do with kind of UX UI where people are like, oh, that's one of them old school chatbots. I don't want to engage with that. Right. And so they don't ask it a question. And so that's something that we're working on is how do we kind of pull more people into the experience? But that is kind of the metrics that we measure today. And what's pretty cool on this site, by the way, is we have lots of blogs on how we did it. You can there's Jim Rote up there. you all know very well. And so we're very transparent. Here's all of our information. Here's how we launched it. Go look at it and see what you think yourself.

34:45Yeah, that's this is awesome. I love the ticker numbers on the homepage. I just feel like it's kind of gamified. I'm like, I just want to sit here and watch it go up and see how many humans are involved. And yeah, that's fun. And can we also talk about the sales, the sales strategy here? I mean, you're literally showing like we are using this and here's the results that we are getting in real time. I think it's something like this point about transparency is so critical. It's always been critical. But when we talk about AI, it is ever more important because customers, whether they're B2B or B2C, will trust you more if you are transparent with when you are using AI versus when you are not.

35:24And we see that in the data. And I just want to, I just think it's a really great, really great website. You did a great job. Dude, I got to admit, the ticker was not my idea, but I love it. And at first I was like, a ticker? Why would, because like, it's going to hit like half a million soon, you know? And you're like, wow, like that's a pretty incredible number. And then yeah, to your point on transparency, Lauren, like if we're not using our own products and we're not showing them like, why would a customer buy it? You know? So it's this, like this trust piece of, okay, Salesforce is using it.

35:53Here's all the metrics we measure. And like, we debate the metrics as well. Like people are like, well, is that really a resolution rate? And we're like, yeah, you're right. We need to do better there. And so we introduced that customer confirmed resolution, which is people actually telling us, yes, you solved my issue, you know? And so we're just like any support leader in the world. I have the same pressure as they do to kind of show ROI and all of that sort of thing. And so we're transparent. Here it is. Here you can see it. I love that you guys are like, you're like the original case study and companies can just come in and like copy paste your strategy.

36:24And by you doing all this legwork, I mean, I think about back when I worked at Google, we were always trying to figure out metrics of like, which ones mattered. And we used to get excited about ones. And then people would come in and be like, actually, you don't want that metric to be high. That's bad. That means they're having a problem. And like, there was always so much debate. And so the fact that you guys did this first and you already have like a whole entire, like basically you can copy this entire strategy and just do it with another company. But I'd love to hear who else has done this.

36:48Like who else has copied this? I love what you said there, Stephanie. One of the other things that we like had these long debates on is that handoff to human. Internally, we call that escalation rate, which is a real support mindset metric. And you say that to people externally and like, they're like, what's escalation? Is that a bad thing? And so at that whole page, you saw our marketing team built that page. So we partnered heavily with our marketing team on like, how can we tell this story easily to customers so they can go use it? And kudos to our marketing team. Hand off the customer is way better than our human is way better than escalation rate.

37:21But the thing I want to tell you about, like we had a long debate. That escalation rate was down at like 1%. And sometimes it was 2%. And now we're at 4%. And people are like, oh my God, what's going on? And it's actually a four because it was too hard to get from agent force into a support engineer with too much friction in the process. And so we want it. We've simplified that process. You can literally come and go talk to a human, talk to an agent, whatever that might be, and it will create your case and send you through. And so we think that escalation rate would probably get a little bit higher.

37:51And we're OK about that because that means it's a good experience for our customers, which is the most important thing we're doing. But it's interesting as people can get driven by metrics. You know, it's like, oh, no, it's a four percent. It's like, yeah, but that's a good thing. You know, and so so that's just like as you talk about that Google and that like debate metrics, we do that a lot internally as well. I think that this piece is something that so many companies struggle with, especially from as a CX leader. I would always say let's make the experience best for the customer. But it's not always the most cost efficient way because it's better if we don't have a human talk to them.

38:29If we just make the customer figure it out themselves, or if we just have the AI do it, it's a cheaper option, but it's not the best experience. And we all know as consumers that when you want to speak to a human and you can't speak to a human, there is probably nothing more frustrating. So I shared a story with my team all the time. Like when I was kind of earlier in my career at Microsoft in Dublin and Ireland, we had this project that was literally called automate and eliminate. And if ever there was a name on a project that tells you you're doing the wrong thing, automate and eliminate was that.

39:02And like, I remember I won like a gold star or some monthly award at Microsoft for driving up this number of automate and eliminate. And what did we do? We just made it really hard to get to the human. We put lots more clicks in the flow. And so we saw people drop off and we were high fiving each other. So it's my example of like, whatever leadership defines is like, these are the metrics. These are, I'm sure you can do things to do that, but it can have terrible, terrible impacts. I think it'd be fun to hear maybe some lessons that you've gotten from either Microsoft or any previous companies that have influenced what you're doing with AgentForce and how you're thinking about, you know, like we're talking about with metrics or building out a team or really anything you're working on.

39:40I'd love to hear a bit about that. Even if I think about my role, I'm the SVP for digital success. And someone called me recently in a webinar, a unicorn, which I didn't know how to think about that. But like, Like what it is, is that not many companies have a leader on a team that are 100 percent focused on the digital experience when you think about service. Right. A lot of the times, like I got passionate in this in my career because like I used to lead Windows support for Microsoft, massive support footprint, thousands of support engineers. But I always say like no one gets up in the morning and says, I want to contact support.

40:15Right. They're all going to go online. And so like I started to really kind of hone in on digital and spend a lot of time there. And then kudos to Jim Rolt, he created a function and I joined to come lead it. And my team wake up every single, we wake up in the morning and think about how can we improve the digital self-service experience for our customers? That's our day job. It's not an art. Like I find in many companies, they have people that run functions. And then like, it's like, oh, and also you should think about digital, you know? And so that's the big thing for me is you have to have people that are invested in it.

40:47You know, I have a team of over 100 people now that are working on like Agent Force is one thing we work on. We have other things, but they're all waking up in the morning thinking, how do we improve the digital experience for our customers? Whether that's, you know, that's not just support motions. We also think about success motions, right? You've bought our product. How do we help you onboard? How do we make sure that you're getting you're getting the value from that product? So you renew, right? Like that's critical, like support kind of that moment something goes wrong and I need help. but we also need a great digital journey to help you on.

41:18And that's what my team 100 % focuses on. And that's been a big learning in my career is like, you need people that that's their day job. That's all they worry about. So I'm naive. Is this not normal to have this kind of role? Like is this role not? Oh, okay. I'm like, this sounds like perfectly like, yes, it makes sense. We should have it. Yeah, no, it's funny. Like there's a LinkedIn community now for digital success and there's like digital success webinars and like that didn't exist. a while ago, you know, and so I get asked to attend these things and talk a little bit about my role in Salesforce.

41:52That's where the unicorn thing came up, because I think I'm the only SVP that's on digital success, which made me laugh. But no, it's not something that, unfortunately for a lot of companies, it's on top of someone's current job. Yeah, I just heard a quote in my interview the other day where someone's like, the things that look contrarian and crazy today will be commonplace in the future. And so this will probably be very common, I would assume, in the next like year or two. Yeah, I love that. The way that we can utilize AI in customer success is still yet to be seen in many use cases. And Salesforce is really, I mean, Salesforce has pioneered customer success, period.

42:29So it makes sense that you are getting to lead that function. The new wave of what customer success really means. I'd love to hear some examples of other clients that you've been working with other than Salesforce yourself, to really utilize AgentForce and how that's been taking place. Yeah, it's actually interesting. That site I showed you, if you scroll down to the bottom of that site, you'll also see other customers that are using it as well. So I feel like a majority of my job now is going to talk to the customers about what we do. So that wasn't my job a couple of years ago, but we have some great examples like OpenTable is a big company that's using AI agents now.

43:07And so we share with them what they're doing and they kind of share. I also want to learn from these other companies, by the way. Like, how are you implementing and what are you doing? What learnings can I take from you? So OpenTable is a huge one. Vivid, I actually have them in my house, their security alarm system, camera system. If you go to their website, you're going to engage with an agent. Remarkable, which probably behind me somewhere. I literally have one that I'm using right now. They're the best. Yeah, incredible. And the best, amazing. and you go to their website to engage with support and you'll engage with AgentForce.

43:42And so they're just trying to treat top of mind examples. Like there's huge companies that we're working with right now. And I can't kind of get into the names of some of them that are getting ready to do implementations. And so that's where me and my team come in a lot because we share what we did, our lessons learned and our mistakes so that they can get ready to go live. So you're going to see a lot of companies going live with AgentForce over the next couple of months, a lot. Are there any key questions that you keep getting asked, especially among these like larger companies coming in? Are there themes where people are like asking similar questions?

44:14Yeah, well, one of them you all touched on earlier, which is a kind of around the data hygiene, right? Like we actually, there's this new terminology my team are talking about recently, which was a new one for me called data collisions, which is like you have data that, you know, content is a good example where maybe you have a couple of articles on something that's similar. And maybe you're using like an acronym name or something and the agent can't pick the right piece of content. It's a collision of that content. And so that comes back to this kind of data hygiene piece that you asked about earlier.

44:47It is really important that you have the right data. We've spent a lot of time actually calling older articles, right? Stuff that's not relevant anymore so that the agent's not trained on it. So that's definitely a very, very common one. The other one I would say is the experience from agent to human. We talk about that a lot. Like, because in our experience, it's like you have a choice. You can have the human join the chat. And so, you know, they come in, it's a seamless experience. Or maybe you don't have the time at that moment so you can create a case. Say like have someone come back to you at a time that works for you.

45:20And so we spend a lot of time designing that experience. And so a lot of customers are like, how did you do that? Show me how it works. And so that's another one that we spend a lot of time on as well. Do you have to go through like expectations with them? I know when we talked previously, you were kind of talking about how a lot of people come in and they just expect AI to work right away. This better be perfect. I better fix all my problems. And you had a good analogy around like, would you treat your employee that way in week one? Be like, what did you do? Why did you mess that up? You got that thing wrong.

45:48I can't remember the name of the bias, but as humans, we have a bias for machines that machines should always be correct. And the example, the great example is like self-driving cars, right? Like we're afraid to let them onto motorways because there's like 0.000 whatever percent chance it is that we'll have an accident. Humans have accidents all the time. Right. So self-driving cars are actually a safer option than a human. Right. But we don't want that to happen until it's near perfect. Right. And so it's similar with some of these agents. It's like people expect the answer to be perfect all the time.

46:21And it might not be. And it might not be for multiple reasons. Like we talked about content. If the agent doesn't have access to the right content, if that content's wrong, then your agent's going to be wrong, you know? And so there's so many reasons, like there's so many reasons behind it. Like I gave you the example earlier on of like, if you don't write the prompts are in position the right way, then it might respond in a different way than you want it to. And so there is that expectation setting. And that's why I think that experience agent to human is critical because you need to make it easy that someone can go, oh, actually, you know what?

46:52I just want to go true to the human. Mm hmm. How have you gone about creating the guardrails beyond just the here's the content to pull from? But especially as we think about the more like empathetic skill set of the agents, how do you like set values in place with the agent? And how have you made those decisions around what to train it on? It's interesting, actually, we spent a lot of time as a team just thinking about like, what do we not want it to answer questions on? And so one example is we have a simple guardrail that you should only answer questions on Salesforce products, which I think every company would do.

47:33Right. Like, it's funny. I look at I look at our conversations every day. I go in and read some of the ones and I can see like we have a sediment analysis that runs on our conversations that tells me whether they're positive, neutral or negative. And I can go look at the negative ones and see what it was. And so like some of the negative ones, people ask, like, how do I cook pasta? And so like Agent Force didn't respond. And I'm like, wow. Like, and I think it's because like people think chat GPT. And so they want to go ask it anything. And like, no, we've confined it into the world of please don't answer anything.

48:05Just answer questions related to Salesforce. You know, so that was one. It seems obvious, but like that's a that's a design choice. If you want, you could give like these things access to the Internet and say, hey, go go answer all of these things. Other things, it was funny, like we were sitting down going, God, we don't want it to try answer questions on Mark Benioff. Like we don't want people to be asking questions about our CEO. And it comes back and says, you know, because because that's not ours. It's a support persona. It's a support site. That's not what it should be for. So we set guardrails.

48:33If someone asks about one of our leadership team, it will point them to our investor relations page. You know, so it's really interesting. Like it's an exercise you need to do with your team. mom, where do I want it to answer? And then probably more importantly, where do you not want it to answer? And I think that that's, you know, it's one of those steps that has to be talked about before implementing. So many companies want to just dive right in. But yeah, this is one of those things. It's like, make sure your data is clean and then also be aligned on where you want this agent to play and where the boundary is.

49:09Where's the fence to the playground? Yeah, totally. And the other thing I'd say, Lauren, is like companies kind of treat it like a traditional software release, right? Like, hey, we're going to go do all this work. I'm going to bang, release this thing. And like, it's not a one and done. It's a constant refinement. It's a constant looking at the performance. Why was it doing things this way? Like, did we not have content for that? Do we need to adjust our prompts? Because like we said earlier, it's an employee, right? And it's grown. It's getting better all the time. Like ours is far from perfect right now.

49:41It'll get better and better and better over time. So you've got to be thinking like that. It's not an old school, traditional software release. It's a digital employee and you've got to be constantly refining it. I feel like this is a perfect segue, knowing the timing that we have left into more of that, like the closing the loop between sales, marketing, CX. Because right now it feels like we're entering into a different world when it comes to businesses. And it used to just be, you know, you've got your awareness, top of funnel, your customer comes in, you sell to them, hopefully retain them.

50:12Customer support is sometimes, you know, either in the middle of that or at the end, depending. Now it feels like everything's kind of merging together and like more conversations have to be happening between these teams. Like you just mentioned, Bernard, like when people have an issue, you might want to come back and like change the model. OK, this happened over here when someone was coming into the flow like this now needs to be adjusted. How do you see the world going forward with this like cross-functional collaboration space that we're in? Every company in the world has this problem of they show their silos to their customers.

50:42You know, you have marketing that's all about lead gen and like, how are we driving the pipe, etc. And OK, now you have sales. We're going to create that. And we have a customer. Now you kind of have your post-sales function support success. And they tend to be siloed. And suddenly you now have this agentic layer that goes across all of them, right? And so like an example of that is like, I see what companies that they struggle to pass the relevant information between the different departments. Like I'll give you a very simple example. Sometimes a success team doesn't have the relevant information of why the customer bought the product.

51:18What do I mean by that? Sales sold it for a certain reason, right? They sat in a room and said, hey, you buy this product and it's going to give you this feature is going to give you this ROI. They sell and they're gone. I mean, I'm not, this is not Salesforce, by the way. I'm just saying this is what a lot of companies tend to think about it, right? I've seen it many times. Yeah, yeah, right. And then suddenly it's like customers are okay. And so your success team needs that information, right? So that they can now figure out how do I do a tailor on board for Lauren to make sure that she gets that value of why we sold it, right?

51:52Or why we marketed it originally, right? Right. And so now you have this agentic layer across the mall that's able to see all of this data. It's able to see what did marketing do? Why did we sell it? What was the business outcomes the customer said they wanted? And now I can tailor like we get really excited with agents is that today our agents really about the support persona. But we're going to light up these other personas like a success manager. Right. It would help you with your onboarding. Right. It would even today our agents text. We're getting ready to launch voice. right, which is pretty incredible.

52:23Our agent would ring you and say, hey, you know, I see you, you just signed up for Salesforce. I see you bought it for these reasons. I'm here to help you onboard. Let's schedule an onboarding session. You know, that's something today that not every customer might get because maybe they're a smaller customer or something like that. So suddenly these agents give you this ability, this digital workforce to do all of that at a huge scale. It seems like, too, this is a an amazing time for like CX leaders and marketing leaders all to also be at the product leader table to be able to influence products that are being developed because they have a lot more intel and everyone's going to know more about the customer and more about what sold them and why they're excited about that than ever.

53:00And so it seems like it's just there's going to be a lot of collaboration opportunities that bring marketers and CX and other leaders to a table that maybe they were not really at previously. Totally. And one of the conversations we've been having, Stephanie, is like, I think the danger with some companies is you could go create agents for all of them divisions. You can have a sales agent, a marketing agent, a support agent. And for us, we want to make sure in our customers experience, it's agent force, right? So it's one agent, but it might be behind it. You have an army of agents that are skilled in certain things.

53:33You could have a renewal agent, a success agent, a sales agent, a support agent. But depending on what you ask it, it brings the relevant agent into the conversation. So for the customer, I'm just talking to agent force in my world. The companies will call it different things. I'm just talking to that company's agent. I don't realize there's all these different agents behind it. So you suddenly, because what companies do today is like they show their organizational themes, right? I always like my example with this is portals, right? Companies will have a support portal. They'll have a, you know, sales portal.

54:03And of all in Salesforce, we did it. And we're working to try and bring all of them together. The danger is you do that with agents as well, is you build all of these separate agents versus leveraging the capability that it can work across at all. So then we have to train our agents to be cross-functional. So I'm wondering how, like how Lauren, do you train these agents and people to think differently about building this structure when you're like, okay, don't learn, don't think about anything I've ever learned before. Now think different, do different. Like how do you do that? You can still have, so you still have like your, your subject matter expertise and your company should still build your sales agent, you know, and your support team should still but what you have with agent forces, we have this, this agent to agent transfer.

54:43And so what it should do is based on the questions Stephanie asks it. So let's say like you ask it, like maybe you have a licensing question or a pre-sales type question. In that scenario, Agent Force is able to recognize that question is related to this domain. I'm going to bring in this agent to answer that question. It's more about the technology recognizing what the customer's intent is and then bringing the relevant agent into that conversation. OK, that yeah, that feels a little easier. So instead of me being like, oh, I have a, this feels like it's a customer support question. I don't need to know that as the customer.

55:16I just need to be like, help SOS. This is what I need, blah, blah, blah. Like I'll just voice it into the system. And then behind the scenes, they'll know like, oh, actually that's a finance thing. She's having problems. Coming back to if we're using structured data and let's say you are a customer, right? And you're logged on. We're then able to start to use some of this information as well to know what we should come back with, right? Like maybe we can see, okay, Stephanie's day two with our product, right? She's, she's onboarding, right? We need to use our onboarding aids. Like, so there's, there's ways we can use our structured data as well, like to help us and understand where you are in your journey and what you're trying to do.

55:52I just wanted to bring it back to a real life example here that this is, this is a real life problem that this is solving. When we call up a number and we wait on hold and then we get through and we end up being on the wrong number. And then they're like, oh, let's transfer you to this other place. And you can wait on hold over there again, which is the most frustrating thing in the world. And this is the beauty of agentic AI is that the consumer or the customer no longer needs to deal with that because the AI knows what it is that we're asking for and can directly put us in touch with the right agent where we don't have to wait on hold because it's AI.

56:35Totally. And it knows who you are. That's the most important. It knows who you are. And it's tailored that experience based on who you are. I'm very excited by this. I was just on a chat line with Adobe trying to cancel my like crazy enterprise account that I accidentally signed up for two years ago. And I still don't even know if it's canceled. I was on with like four different chatbots and they kept being like, but actually, do you want to buy this thing? I'm like, no. And then I'd come back to it and be like, oh, that never got canceled four times. And I think it's still a problem. And they didn't know enough information about me.

57:06And they kept asking me stuff. I'm like, I don't know. I think you should know this. I think I'm your customer. You should know this about me. They probably do, but they just don't want you to cancel. They just don't have agent force yet. It's a problem. So I wanted to talk a little bit about the impact on the humans in the loop here. And how this is, I mean, obviously AI is having a huge impact on business. and with new employees, these AI employees, it's going to impact the existing employees. And the thing I keep coming back to is what are the skills we need to be learning to really work with AI now that we have these new types of friends that we're working with day in and day out?

57:49It's interesting. So one of the things, I'll come to your question in a minute, but just I think one of the other things is this like fear of what to kind of do for my job, et cetera. And so one of the things we've been really trying to do is like with our support engineers is making sure they're part of our AgentForce launch. So we use our support engineers to look at the performance of AgentForce to help go, actually, what would you have done in that scenario? Because they have more expertise in our products than anyone else in the company, maybe outside our engineering team. So since day one, we've included our support engineers so that they feel really part of it.

58:22And, you know, hopefully it takes away some of the volume that's probably easier for them. And they're left with the more complex scenarios. is we're always going to need like incredible support engineers. So that's the first thing I'd say. On your question, it's been interesting. I was having this conversation with my team today. We just hired a prompt designer slash engineer. And, you know, that role, I don't know how far back you go, but it didn't exist. There was no such thing, right? Like, you know, and so I was even laughing because I was super curious on what type of applicants we would get.

58:51I was like, God, does really have a, you know when people put like 20 years experience or something and there's no possibility of 20 years experience. So I was like, I wonder like Like, who's going to apply for this? We actually got a rock star. I can't wait for her to start. But like to your point, there's these new roles starting now. We have a director of our LLM operations. You'd have told me two years ago I'd have a director of LLM operations. I'd be like, what are you talking about? Is this English you're speaking? And so I do think it's up to people to start training on some of these, like depending on what your interest is, like, you know, UX designers, there's now this prompt design, which is a new flavor of UX design, you know, And then there's understanding these models and more about them.

59:30So I think it's really exciting. And I said it to my team. It's like, wow, look at us. We're blazing a trail. We're creating these new opportunities. There's opportunities for existing people in my team are getting more deeper into it. It might create new roles for them. So it's kind of cool, to be honest with you, on some of these roles that are opening up. Yeah. I mean, with any technological innovation, new roles have been created. And of course, right now, a lot of people are afraid, oh, is AI going to take my job? And as you know your job to be today, probably in some cases, but there's going to be whole new jobs, whole new careers, whole new departments maybe that are even created.

1:00:07I know I've been talking to a lot of people about the new role of the AI manager, the agent manager. Like as a human needs to now manage this team of agents and train them and guide them and coach them. Yeah. And I think I think your point is I feel in a lot of situations they augment the human. You know, I was just on a call earlier today. I have a team, a video production team. So we have a Salesforce support YouTube channel. And it's an amazing use, like the second fastest growing YouTube channel in Salesforce. And we have a lot of things happening in Salesforce, which shows you like how much it's grown.

1:00:42But they were talking to me now about how they're using AI for their video scripts, you know, And something that would have normally taken hours and days is now taking minutes. So it's not that it's replacing their role. It's allowing us to go faster and to implement more videos at a more quicker time frame. And so that's great because it means we're getting more out there for our customers to self-serve on. So I think you'll see more examples like that where it's like it's augmenting my job. It's allowing me to be more efficient at my job. And then it frees up time for more valuable activities as well.

1:01:16I mean, I think about, especially in the realm of customer success, if we can spend more time listening to and thinking about our customer instead of doing operational, you know, even scheduling meetings and things like that, that just like eats up our time. We're much better off thinking about our customers, spending time with our customer, listening to our customer, and then bringing that insight back into the organization. Because those human to human interactions, that is what AI will definitely never be able to replace. Right. So let's lean into that. So one of my favorite questions that I ask of all my marketing trends guests who are all normally CMOs of the Fortune 500, Fortune 1000 is what's something that you believe that very few agree with you on?

1:02:00I love this question, by the way. I'm going to steal this question and I'm going to use it with people when I'm interviewing them because then I think you'll see how their brain thinks. I had to think about this a lot, by the way. So kudos to you. You had me kind of like, oh my God, what will I say to this one? The thing I landed on is like, I don't think data is everything. And I think a lot of leaders in marketing, support, whatever it is, they just live in data. Now, I live in data. I look at data all the time. But sometimes I think the best decision goes against what data tells you. And so, like, I always push my team is just because our data is telling us something like, is that really what's happening?

1:02:40doesn't often tell you the why and the why someone chose to do something, the why they made that decision, the why they went that way. And so I challenge people. Of course, you should have data. Data is foundational to everything you do. I live in scorecards and metrics, but I really challenge people. Sometimes it's about going with your good instinct, you know, and saying, hey, no, I think this is the right idea. I've been doing this for a long time. I'm going to go against the data and I'm going to try to do something myself. Have you done this recently? I'd love to hear a story if there is something that comes up around this?

1:03:10I don't know if I have a perfect example, but even coming back to AgentForce, like, you know, data probably should have said, don't release AgentForce to the whole world at once, right? Like it's doing an okay performance or resolution rate is X or whatever. You know, you look at the data and you're like, let's just continue to go slow. Let's just iterate. And over time, we'll get it better and better and better. And then we'll go large. And if I'd have made that decision, I don't think we would be where we are now because because we went big, even when the data told us it wasn't doing an amazing job, it was doing a good job.

1:03:45It meant that suddenly it was exposed to more people and there was more pressure on me and my team to make it better, faster. And so that's like that's like where it was like it forced us to go faster by going against the data and releasing it to everybody. I think about this often. I'm so glad that you share this take or this opinion, Bernard, because going back again to what AI will not be able to replace, it's our intuition. And it is so important that we hold on to that intuition and if anything, strengthen it now that we have AI doing a lot of the logical thinking for us, where we can say, is this right?

1:04:25just because the data says it doesn't mean that we need to do it that way. It is insight. It is information. We can take that, but we still need to use our human critical thinking skills to decide, is this the right way to go? Totally. And your intuition is built up over years of decisions, bad decisions, things that went wrong. And so AI doesn't have that. And so sometimes your human instinct is the best way to go. And it's the best way to go. Yeah. AI probably a lot of times has all the risks that have ever happened in many years. It's been tracking data. And I think what's amazing about humans is being like, I'm going to do it anyways.

1:05:05And that's actually where most of the breakthroughs have happened. The biggest companies, the most life-changing things is like, yes, it's showing me this might be a terrible idea to do based off these things. And I'm going to send a rocket to space anyways. Yeah, exactly. Right. No, it's great. No, no, you're crazy, right? Why would you do that? People are going to die. Yeah. You know? Yeah. Have you changed how you lead your team at all now that AI is such a big part of your day to day? I think it's a lot of my focus right now. And I actually kind of worry a little bit that I spend so much time here.

1:05:38Like my team has lots of other elements. Like I talked about like our YouTube channel, I manage all of our email campaigns for onboarding and there's just like a ton of things to my team. And I worry actually a little bit that I'm so much of my time as an agent force that people that aren't working on it in my team feel like they're not working on the big shiny thing. And so it's about prioritizing my time and making sure I'm still spending time in other areas so that they can see that it's still, because it's not all agent force. You know, we've lots of other things we do that's super important.

1:06:08I think that the AI hype cycle definitely has a lot of people's attention at the moment, but it's really important to remember that it's not everything. thing. It's the shiny, it's the shiny new thing, but there's still a lot else. Although I do tell my team, I think everything that we do today will change. Like even like, if we go in that example of like, you know, we send emails to customers, you know, and like, Hey, you know, you bought Salesforce, you should do these things. Like, and it feels so 1980s now, you know, and like, you can imagine a world, like I said earlier, a proactive agent, right?

1:06:42That's looking at signals and saying, Oh, Stephanie hasn't turned on this feature yet, you know, and it either reaches out in text or voice or like, you know what I mean? And so like it all becomes automated. So like, I do think everything we do today will change. I think some things will change completely. Other things will augment. Other things will be a little bit different, you know, and so, but I think everything will change. It's kind of similar to that quote, whereas like if you have a hammer, everything you see is a nail. And that's kind of what I see with everything right now. I'm like, I do think AI will be in many of these things and they will be completely different.

1:07:15And so I'm kind of like seeing everything as a nail, like everything can be disrupted and it's amazing. And what a great future we get to live in with like all the things we don't have to do anymore when this comes to be. Yeah. Like how much do we use chat GPT now in your daily life, you know? I mean, on that point, I actually have a question for both of you, which is what is your favorite personal use case of AI? Yeah, no, I have a good one because someone actually shared it with me was we're going on spring break vacation in a few weeks. And someone shared to me about like actually putting into chat GPT, where you're going on vacation, who you're going with.

1:07:53I have an eight year old son. These are the things I like, et cetera, et cetera, et cetera. Here's how I'm going to be here. Here's the food we like to eat. And it gave me a tailored inventory of what I should do on my holiday. And I was like, wow, like some of this is really good. Not saying we're going to do everything on it. I want to relax at some point as well. But like, what a cool use case, you know, of like being able to get that inventory of what to do. Yeah. Oh, that one's way more fun. Cause I'm like, I think about the stuff that I'm in now and I'm still, it's still so work related, but I love having it just like poke holes in anything I'm doing.

1:08:26So if I'm, for example, coming up with, you know, some questions for an interview with Bernard and uploading the whole document and being like, pretend you're like a top podcast host and tell me which questions are going to result in a lame answer and which ones are going to go nowhere and which ones are more thought-provoking and contrarian and like and i keep making it poke and then i get a new one i'm like now look at this one and tear it apart and pretend you're a cmo like what won't resonate with you what parts don't you care about like basically be really aggressive with everything i'm giving you and it just i feel like gives me very cool such good output by saying pretend to be this like cutthroat killer in the podcast industry i'm like ah okay this is great i'll give you a good one stephanie because i know you have kids that someone shared with me recently is they were at home and it was a rainy day in seattle and they asked chat gpt and it gave them games to play in the house with their children and that they had never played before oh i love that especially if i could say like here's the things that i have i've got this random tree you can climb i have you know yeah oh that's a good one i love that i'm stealing that for all four of my kids yeah there you go parenting 101 from burbond here I'm taking it.

1:09:33It's hot tips. Yeah. Well, Bernard, we have one question that we ask all of our guests on this show to close it out. And that question is, what is one piece of advice that every customer experience leader should hear? Oh, that's an easy one for me because I talk about this a lot. I don't think the word customer obsession. I find people say them words and they don't really mean it. Like it's like it sounds good. It's a nice thing to write on a wall, but like you got to be spending time even in the digital space, right? Like Agent Force, a good example. Like I read every day some transcripts from Agent Force, the ones that were bad, right?

1:10:11I'm meeting with customers. I'm meeting with our MVPs who give very critical feedback on like, what should we be doing different? How do we make it better? And so like, I think to be truly customer obsessed, you have to be looking at your calendar and going, how many hours a week am I actually talking to customers or reading something that was interaction with them. That's what it means. And I just think if you're in the world of CX, that is critical to anything you do. And it goes back to the data point you made. It's not just the data. We have to actually have those conversations, really be there with our customer, listening to them, understanding what's the need behind their need and digging deeper with our questions.

1:10:50So I could not agree more. Thank you so much. This was awesome. Thank you all for letting me jump in and be in this conversation. Super fun and excited to run it on Marketing Trends as well. Thanks, Bernard. Yeah, great to meet you both.