Building a Team of AI Marketing Agents with Ryan Gavin, CMO at Slack

0:00Welcome back to another Flashback Friday, where we dust off an episode that hits different the second time around. This week, we were visiting my conversation with Ryan Gavin, CMO of Slack. Yes, that's Slack. We spent 20 plus years inside the AI shift from Microsoft to building AWS's machine learning practice to now leading marketing at one of the companies that actually are building AI powered for future work. We get into why it is cheating to use AI, how Ryan used Slack's own AI to onboard himself into the company's entire 10-year history, and the marketing hill he would die on. If you're a marketer trying to figure out where you actually stand in the AI era, this one is a required lesson.

0:46And let's get into it. Welcome to the Marketing Millennials, the no BS marketing podcast. I'm Daniel Murray. and join me for unfiltered conversations with the brains behind marketing's coolest companies. The one request I tell our guests, stories or it didn't happen. Get ready to turn the f*** up. What is up, Brian? Welcome to the Marketing Millennials Podcast. Hey, thanks, Daniel. That's awesome to be here. I appreciate you making some time. Could you give a little background who you are? So just to set up the conversation for today. I'm middle management overhead, brother. I am the CMO for Slack.

1:37I've been at Slack now for about a year and a half. I've been, you can see all the gray hairs. I've been around the industry for a while. I've worked for about 20 years at Microsoft. Worked at AWS where I helped build out the machine learning and AI practice there. And then I went to startup land, which is incredible. I was in the supply chain industry, kind of transforming trucking. Did that for about four and a half years as the chief growth officer. And then, yeah, as I said, about a year and a half ago, joined Slack as part of Salesforce to really take Slack to the next level. It's been awesome.

2:11I don't really see the gray hairs he's talking about, but he can claim it. But I want to rename the podcast the Gen X podcast. I want to get into, you are really in the front lines of AI and work right now. So what is one thing you think marketers still get wrong about using AI in the day to day? Oh, man. How much time do we have on this podcast? It's hard to say get wrong. I think you have to start with the understanding that work's going to change more in the next five years than it has in the last 25. like almost undoubtedly some things may stay the same but it's going to fundamentally shift and so every marketer in fact every person inside of every organization really not just marketing you kind of need to come in to your job with like a beginner's mindset it's a bit of a cliche phrase but you got to think about what would this job or what would this task look like if i was to start this company today in the age of ai and what would i do differently so if you are you know working in email automation or you are working in you know like product-led growth or you know uh you know trying to optimize a conversion funnel or whatever it could be how would you do it differently if you had just one person and a whole host of ai tools and what would your process be and how you go about it and if you're not kind of starting you know all the things that you do with that kind of open-ended question and asking like all right?

3:50How do I rethink this entirely in the age of AI and the age of agents? You really are behind. And there's really no, no nice way to say it because it is going to change that quickly. And, and those leaders and those marketers are going to embrace the latest tools and capabilities are going to just far outstrip those who aren't. And it's, you know, it's changed and sometimes change is a little bit scary. You know, everyone approaches AI with a different mindset. You know, we've done research where there's people who, you know, think of AI as like, wow, this is these superpowers of tools. I'm so excited to go use them.

4:28And there's others on the other side of the spectrum that have like, is it okay to use these tools? I don't want to tell people I'm using it feels like I'm cheating. So I kind of like hide that I use it. And there's some people who are really nervous and scared to use it and don't want to use these tools. And so, you know, Everyone kind of approaches it with their own starting point, but it is going to be pretty transformative. And there's a lot of transformation that has to happen in the marketing space that we're barely scratching the surface of. I think that you make a good point of where people are, because they are different point of views out there where some people think if you're using AI for certain things, you're not doing your job as a marketer, you're not doing your job.

5:08or you're being lazy at work because you're using something else. But I would argue it's productivity instead of laziness. You're doing more with your time. If you're doing it straight out without using any of your brain power and strategy and experience, then maybe the argument would come into that. Yeah, totally. I mean, the thing I was thinking about, I was talking to somebody the other day, And at Slack, we have this, Salesforce has this incredible platform called Agent Force. And so we have all these agents that we get to deploy for employees and they get to kind of use these agents to augment their productivity.

5:49And one of the things I was kind of, you know, thinking about is like, we've had all this explosion of tools over the last, you know, 20 years. Like, just think about the number of new, like, applications and SaaS tools that have launched the space. It's not like anyone's going home any earlier. Like, we're still filling up the day with work. The question is just you're getting more done. Ideally, you're doing higher value work. Ideally, that work is having greater return for the business. But it's not like anyone's like, hey, cool, it's a three day work week now. You know, that doesn't exist.

6:21So we're all being asked to do more. And unless we're kind of someone embracing some of these tools, it's going to be really, really hard to keep up because like the compression algorithm is doing what it always has done, which is expect more for less. um what is a few ways your marketing team at slack uses ai to get done more with less yeah i mean we we feel pretty we're pretty blessed honestly i mean i'll tell you a couple examples um you know one my personal example so when i joined uh slack you know just like i'm sure many of your listeners a lot of folks you know you join a new company and you're learning the culture, you're learning all the new tools and processes, you're learning the people and all this kind of stuff.

7:04And that learning curve is kind of daunting anytime you've taken a new job. And, you know, whatever. I'm a middle management executive, whatever. So I have all these people who are like, hey, we're going to help you and we're going to create documents for you and all this kind of wonderful stuff. My number one onboarding tool when I joined Slack and Salesforce, number one, without a doubt, was Slack AI, which was, so Slack AI basically allows, So if your listeners aren't familiar with Slack, Slack is this incredible kind of work operating system collaboration tool that allows organizations to really kind of orchestrate all of their work.

7:37But one of the most amazing things about it is it also stores your company's long-term memory. So you think about, you know, Slack's been around for greater than 10 years. All the conversations that have happened while the company was building, all the conversations that happened when it was having its first customer meetings, all the conversations that have happened since, you know, strategy reviews, et cetera. All those live in Slack. And those conversations, because the way Slack is architected, I have permissions to all the public ones. So if you create a channel in Slack, there's permissions you can set.

8:08And some of them are public, means everyone can have access to. And some of them are private, meaning a set of people. And I have access to those public ones as well as some of the private ones. well previously those conversations were kind of like just trapped like they happened they went away and then you never really see them again and you never hear from again well what happened with slack ai is we're able to query those conversations so myself as a new employee coming in someone would be talking about something i have no idea what they're talking about i could just quietly rock up to slack ai and ask it a natural language question like an example is one is salesforce uses this thing called the v2 mom which is a way of our setting our our vision and our values and the measures and methods of, of our, our goals for the year.

8:48And someone said it like on my first day, I was like, I have no clue what you're talking about. So I just go up like, what's a V2 mom. Um, and there I get this full rich answer. Um, I also could go do the things like, Hey, you know, the, the founder of Slack was kind of Stuart Butterfield. I could ask Slack of like, Hey, show me the original presentation Stuart did to the first set of customers. And what was the original visions like he had for Slack? because all those conversations are there. That long-term memory exists. And I could query it. I was rocking up to Slack AI like 10, 12 times a day just to get insights and understand.

9:21And it was this amazing treasure trove of insights. And those are the types of things where it's one example of AI able to tap into this long-term memory of a company, which has previously really just sat there and been idle. So that's one example. I could give you like 50 others. We do a ton with agents in Slack, which is basically how every single employee gets their own workforce. And that's pretty darn cool. I'm happy to talk more about that as well. Yeah. I mean, first you want to describe, like, for the people who don't know, like, if someone says I'm using, like, AI versus, like, using an AI agent, like, what the two differences?

9:59And then, like, what are some, like, cool use cases that people should be thinking about using for agentic AI? Yeah. I mean, it's a great clarification. So, you know, AI obviously gets thrown around. It's like the, it's the drinking game du jour. So, uh, every time someone says you have to take a shot kind of thing. Um, uh, I don't know, it's 8am here, so I don't expect you to be doing that, Daniel, but if you want to, if you want to, I'm just saying it might be fun. It might spice up the podcast. I, you know, so you think about like AI capabilities broadly, where it can be doing anything from, you know, what you see inside of, you know, ChetGPT or Gemini or, you know, hear suggestions for your writing, or I'm going to summarize this podcast, or I'm going to create a podcast for you.

10:42So these broad applications of AIs. What happens with agents and where agents get really powerful, and there's really kind of a couple elements of it, is agents can then go beyond the AI insights, and then they can really start to take action on your behalf. And what makes agents really powerful is I can then deploy an agent. Let's say what we do inside of Slack with AgentForce is we deploy agents that have specific tasks and skills just like teammates. So an example being we have an engineering agent that we built here at Salesforce. One of the challenges that happens at a big company at scale is we roll out all these products.

11:22And all these products have lots of technical depth and details to them. And as our sellers go out and sell them, they're working with customers or our customer success teams are working. They'll have questions about implementation details, et cetera. And typically those questions would go back to a Slack channel and we'd have a bunch of, you know, expensive in reality engineers who would help our customer success and our sales teams answer those questions and then, you know, help them take action on it. Well, we built this engineering agent and this engineering agent was then wired up with all the knowledge of all of our knowledge bases, all of the details of these products, all the past history of questions that we've answered on these topics.

12:02and this engineering agent, and we deployed it inside of a Slack channel. So now when a sales leader or a customer success team is going and they have a question on a product, they go into, and this engineering agent is the one that's actually answering the question. And I honestly, you sit there, you watch it in Slack, it's an unbelievable thing to watch because you're like, holy crap, this is working. And what happened is like in like six months, this thing had to answer like over 18 ,000 questions. The success rate in terms of being able to answer them was like a 76 % accuracy. Like it was able to ask, you know, the question was asked and answered and like closed out.

12:36It didn't need to be handed off to a human. And then you can kind of even go beyond. These agents can then take actions and say, hey, great, I see you've answered that question. Now, can you actually create me a Canvas doc that can summarize this for my customer and summarize the next step the customer needs to take? So the agent now has the autonomy to go and take that action on your behalf and is now doing that work for you. And you can go to bed, go get a cup of coffee or whatever, the agent does the work, then you're ready to rock and roll with the customer. So these agents become incredibly powerful.

13:05I mean, this one example, I just gave you the engineering agent. You know, we saw it save over like a quarter of a million of engineering hours, like a quarter of a million. And then, you know, we projected if we built that out across all the different topics that we have, it would be like 20 million or$20 million hours saved. Did I lose you there? Looks like. No, that's insane. I'm just listening to say. Okay. Yeah, no. I heard an echo for a second. We deployed that agent across a dozen different channels. We're going to deploy it across 700 million channels. The projection is it's$20 million of engineering hours saved.

13:42There's not a lot of$20 million ideas just walking around inside of companies. These are the types of things that we're seeing inside of Slack, that we're doing as part of Salesforce that every customer out there is thinking about. And we're working with some really cool ones that are doing the exact same type of stuff. I want to also ask you is like, I know you said you use like Slack AI to query like past conversations so you can like stay up conversations and meetings and have the history and be able to create more like alignment in the team. But what are some things that like if you were talking to other marketing leaders, like what would you suggest they should use in their day to day to make them more productive as a marketing lead?

14:28And or or the other question is like, how would you suggest them empower their team to do use AI in more of an efficient way? Well, yeah, I mean, that's another one where I mean, there's so many incredible tools out there that you can be looking at. You know, our team recently redid, like, you know, we have an email system. Our team recently kind of redid all of our email stack, essentially kind of with an AI first mindset, like I was talking about a little bit earlier, to go through and think about not only just who we send emails to, the copy that we use, how do we optimize the timing, all that kind of stuff.

15:08and you can go back and take something as simple and as antiquated, I kind of hate email, honestly. But we still have to send them. You can take something as antiquated as sending emails and you can go through and really go and build out a full optimization. And despite all the work that's been done in something like email optimization over the last 10 years, there's still way more that can happen and AI actually takes it to a whole other level. So we're seeing incredible results with that. The other one, maybe on the other end of the spectrum, that's a little bit more out there is how do you start to reimagine your customer experience in an agent first world?

15:48And what I mean by that is what happens in the world where no one's ever going to go to your website anymore? Like there's going to be an agent first internet. And like, there's estimates out there. There's a Mary Meeker report I read recently. It talked about this idea of like, you know, there's a billion people who are about to go over the next, you know, five, 10 years, whatever it is, and experience products and experience without ever going to your website or whatever going to any content that you've created, like it's all going to be behind agents. And so what does a website look like that's agent first or where the interactions and experiences someone's having with your brand are all through an LLM.

16:28and then they're making a decision to purchase based off of all that information. We're seeing that right now inside of Slack. It's one of the interesting trends. You know, marketers talk about SEO and all the funnel optimization and bringing people in, blah, blah, blah. The reality is people are doing more of their research in ChatGPT and Gemini and Anthropic and all these other tools, and then they're coming straight in and they're ready to sign up for Slack. And so like the traditional optimization of how do you get someone to your website and then you click on this button and move through that, like all that's getting, you know, passed through.

17:03And so what would your website look like if it was just agent-based and, you know, started with a search-based experience that was conversational? That's a lot of what we do inside of Slack is, you know, you start with like a conversational first experience. So, you know, there's two ends of the spectrum. Like one is like the traditional, the T's and C's of marketing and all the wonderful stuff that people do. And how do you reimagine that with AI, all the AI tools and capabilities? And then the other one is like, hey, how do you get ahead of where the puck's going and really think about a totally different experience on some of the things that have been pretty fundamental for the last two decades?

17:37Yeah, I want to. Does that make any sense, Daniel? No, it makes total sense because, for example, I bet you a lot of marketers out there are searching, like, what is the best productivity tool for a marketing team to chat with there, like, internally? And probably Slack was probably coming up there. And then they go deeper, like, what is Slack? And they're asking. They don't even like going on Google to search that. they're going they're just having conversations I mean I do it all the time just to see what what pops up to stay ahead just to say and then what I do is reverse engineer and saying what sources are they tracking back to to see what websites are they querying that's a question I had for you so if you work backwards from all those people who are searching Slack in these tools what are some ways you are thinking are helping you show up more in traditional, like all these AI chats?

18:48Yeah, I mean, there's, you know, everything from some of the traditional elements of like, you know, what is the, are you adding their esteem to your pages? Are you doing the right indexing your pages? Some of like the, you know, what I'll call like really, really good traditional SEO work. But there's also there's a lot of companies out there now. And, you know, there's a lot of good research you can go read of like, what does an LLM look for that's different than, you know, traditional, you know, Google search engines or, you know, Microsoft search engines. And, you know, there you're really getting into longer form descriptive content.

19:25You're getting into places where like you've got better built out answers and you've got that index really, really well. and you can go and build out that content set that is going to be more appealing and for an LLM that's looking for the next best word and the next best answer. And so you really have to kind of get into this like next wave of like, what is the LLMs looking for? How do you optimize more? And the reality, anyone's lying if it says it's not a bit of a black box. It's a total black box. Like there's no formula thing. Like these LLMs are evolving every single day. no one can tell you exactly why it picked what it picked but you can look at the patterns and practices and you can go and try to optimize for those patterns but that's kind of the the the game that you have to kind of be in nowadays um uh particularly as as web traffic is going to you know most likely continue to drop in favor of kind of where where things are moving with lms the the other one just just uh just to kind of just jog my memory a little bit on um The big one that I'm seeing where a lot of customers are leaning in for marketers are leaning in is kind of the age old challenge of, hey, I generate all of these leads and I generate all of this, you know, marketing opportunities and my sellers never follow up on them.

20:40Like probably probably the one of the biggest areas I'm seeing customers deploying things like Agent Force is on this like lead routing, lead optimization where you can go through like Box, I think has like a, you know, did a bunch of work with Box and Agent Force, and they did like a 500 % improvement in like their lead optimization. And you just get far better signals on what are quality leads. And then you can do a lot of the qualification work with agents before it actually gets to the sellers. And that becomes pretty powerful as well. Like basically, the SDR role gets replaced with agents in many cases, and that becomes far more effective.

21:21Yeah, I remember back, I mean, even just like 10 years ago where our lead routing formulas were all in Google Sheets and they took weeks and weeks to optimize based on new data that you had to pull into a Google Sheet or Excel and then have to feed it back in to keep changing. But now you could just feed these things into AI and it does it in like two, like five minutes for you, which is like, which would took you three, four weeks, 10 years ago, which is crazy to think about. I was just talking to Anthropic. They have like a deal desk where they deployed agents working with humans and they increased like their deal desk velocity by like 60 % with like agent force agents.

22:06Like it's, it really is nuts how fast you can move on some of this stuff now. One thing I also think is, which I think is an advantage for like companies that have been storing data is like, how are you using these like past internal, basically like first party data or like first party content that you have created for the last 10 years and then deploying that now as like content for your customers? Because back in the day, you would have to like sift through all these conversations you've had with customers, these Salesforce records, Slack conversations, whatever you've having with like group conversations you've had with customers and it took hours to sit through just to make a testimonial or case study or anything now like what are some things you're using it for like internal slack things that you you built up knowledge bases to deploy for marketing material now it's it's not exactly what you're you're poking on but it kind of it triggers for me a little bit.

23:11There's one of the, you know, again, really cool uses of AI again, is there's so much knowledge that's trapped in these conversations, like conversations are context and context matters. So, you know, we have Slack channels for everything, obviously it's Slack and we've got all these account channels for all the teams that are, you know, working with IBM or easy cater or slalom or whatever it is. And you think about all those conversations are happening in those channels and there's tons and tons and tons of context in those channels. Well, one of the things we've done is we've also taken all that conversational context and we've connected it to what's considered structured data.

23:45So look at the data that actually sits in this CRM. And we brought that into Slack with a new capability called Salesforce Channels, where it's basically bringing all those opportunity records, all those contacts. Where's the deal at? Where's the process? With all the conversations that are happening. Hey, we're getting ready for this meeting. How do we prep for it, et cetera. And now with AI, you can go through and you can look at all that and say, cool, agent, sales agent, which we have, generate for me a customer brief that is ready to go for this meeting. So when I'm getting asked to go meet with a customer, et cetera, the past process oftentimes would be, hey, some account team member or many account team members are going to spend four or five hours creating a briefing doc for me.

24:33here's everything you need to know here's what the account's up to so now with like with ai with these sales agents we can just point it at the structured data and this unstructured data that all lives inside of slack and say create an account brief for ryan make sure it emphasizes the latest opportunities make sure it gives an overview of what's going on with the executives uh and please uh give them you know a summarization of the account history from the last three years boom like i get this beautiful canvas in slack it's ready to go no one had to spend any time I'm working on an agent did all the work.

25:04And then honestly, you know, and no offense to all my lovely sales teams, it's way better than the big docs that I used to get from the sellers. So that is an example. You talked about more like customer evidence and stuff like that, but that same thing, that same applies to customer evidence. It's all this context that you've missed for a long time. Like when you ask like sellers, sellers are not going to be able to like remember every conversation for the last three years or, but that also makes me a point. What are some things that inside... Because some of these things are as good as how people enter it into their CRM or the questions they ask in the conversations with the customers.

25:49Because if you ask bad questions or you don't get enough information or you put bad data into Salesforce, it could be spitting out things that are... And it's not like anybody's fault, but it could be spitting out things that are going to lead you down the wrong path. So what are some ways now that are you encouraging better questions or better data entry? I mean, before, I remember we used to have rules like you have to enter these fields. And we would go, as marketing ops, I would have to go audit. Did they put enough data into the field and stuff like that, which agents probably can do now.

26:26But like back in the day, you had to audit like all these records at weekly of like, did you put in the right data? Is this the right information? Yeah, I feel like, Dana, you have some scars from that. I'm sensing the pain from your past. Totally. Like garbage in, garbage out is always going to be a challenge. And, you know, at least in my experience, telling people they have to do something or, you know, putting rules in place only gets you so far. and typically just makes people resent it because they're like, I'm actually trying to do my job and like, why do I need to like, I'm not a data entry person.

27:02So the real question is, is how do you make it just brain dead easy for whether it's a marketer or marketing ops person or a seller to share the data that's important and have that data get updated to a system of record. So, you know, in the future, like I've got an agent in Slack and let's say you and I just got out of a customer meeting well shouldn't I just be able to tell the agent like hey Daniel and I just met with Plative. Plative's got an opportunity and they need help on these five things. I'd like to do a follow-up call with them in three weeks. We think the opportunities may be a million dollars for you know these three products.

27:45Please make sure that these five people are aware and schedule that follow-up call for me. Done. All right the CRM gets updated. the meeting gets scheduled the right people get an email or a note in slack from the agent and like you're on to your next thing versus before i'd sit down and say all right now i'm gonna launch my crm i gotta apply the opportunity then i've got to go into my calendar and set up that meeting now i need to send the follow-up notes now i need to send the slack message to the other people need to be aware like that is the the level of kind of reimagining with things like Slack and AgentForce that companies are able to start to go do.