How Slack Uses Marketing Cloud Next to Optimize Paid Search | Dreamforce 2025

Well, welcome to today's session. How Slack uses Agent Force Marketing to optimize Paid Search. Before we get going, I have to flash up the forward looking statement slide. Please make your purchasing decisions based on products and services that are available today. And with that I want to say thank you. If you are tuning in from the broadcast, thank you so much for joining us. And if you are in the audience tuning in live, I want to thank you as well. We hope to make a session worth your while. We have a lot of great content. So really quickly I want to spend a moment and introduce myself. My name is Sarah Pfefferman. I'm a product marketing manager here at Salesforce on the marketing cloud team and I am joined on stage by my partner in crime, Javesh.

Thanks. I'm a senior product manager working on C360 truth profile team. We are basically customer zero for data cloud at Salesforce. Thanks Sarah.

Awesome. Okay, so a quick preview of the agenda that we are going to cover. We will start with our paid search introduction. We will do a demo. Javesh will do a paid search agent demo. We will walk through the tech stack that is powering our custom built agent and then we will end with a couple takeaways for you all. So let's dive in. Paid search is a huge industry. It is a $200 billion plus industry. It accounts for almost half of every marketing team budget. Show of hands in the audience because I do have a live audience. Who here knows of your marketing team? If you are running paid search, okay, every hand is off because paid search is huge. So then you can all probably relate that optimizing page search is incredibly complex. And it's complex because it requires constant keyword identification, constant keyword testing. And this means marketers on a daily basis are balancing thousands and thousands of keywords every day. Humans can't keep up with that kind of scale alone. Right? And why is that? Limited resources, manual processes, disconnected data sources, competing priorities. These all get in the way of the marketing team's ability to to maximize that paid search impact. And here at Salesforce it's no different. We might look like this huge org if you're walking around Dreamforce. But our teams face the same challenges, the same bandwidth constraints. And our Slack marketing team is running a ton of paid search specifically because they are reaching such a broad spectrum of use cases. And this is why they set out to do it differently. They wanted to find a better way and hard to scale tasks just like this. Paid search optimization is a great fit for agentic use cases for AI agents. And that's where Javesh and team come in because they have built a custom agent to specifically help the Slack marketing team optimize paid search. So wanted to flash up Josh's picture. Josh is a colleague of Javesh and I as well. He runs growth marketing for the Slack team and he just started piloting this custom built agent and he's so excited because this means that he will have that much more opportunity to optimize and take advantage of that $200 billion industry that he's investing into. So without further ado, do you all want to see this paid search agent? Awesome. I get some head nods. Javesh, I'll turn it over to you.

Awesome. Thank you so much. Basically, this is if you're a Slack marketer or if you're a growth marketer and you want to engage with an agent to do your paid search execution, campaign planning, campaign execution, workflow. This agent really allows you to engage with this and go from keyword discovery all the way to ad copy generation within a few minutes. And I'm not even joking, we'll walk through that with all of you. So let's move to the next slide. This is the Slack interface. So as a growth marketer, if you want to interact with the agent, the interface is slack. Now in this interface, as a marketer, I want to get a list of new keywords, high performing keywords that I want to get a list of and really understand which keywords are doing well and which are not. So we give a prompt to the agent and the agent before returning the results asks us, first, do you want to do a Lexical search or a semantic search? Now even saying those words is so difficult. But for our use case, we went ahead with Lexical because we wanted to reach a wider audience. I'll explain why we chose Lexical and what's the difference between Lexical and semantic. So Lexical search actually allows you to come up with terms which are literally very identical to your actual topic. So if the topic is project management, the search terms would be around project management apps, project management tools. If you're doing semantic search, it'll return terms which are basically in terms of meaning, very identical. So for example, for project management, the terms would be workflow optimization or task planning. So the agent actually creates the results. And right here in Slack, we click on the link and it takes us to the Slack canvas. In this canvas, what we see on the left is basically a bunch of keywords that are sorted by impressions and Clicks. It also tells whether they are paid or organic. It also tells where these keywords are coming from, which channel. So we used five different channels that we brought in data from using marketing intelligence into this agentic workflow. Now, if you look at the list of keywords, what you'll notice is for example, project management app, you'll notice the impressions are really high and it is returning a very high SEO. What that means is it's already doing well. You don't need to spend any more money on that, so avoid any duplication of cost. However, if you look at project management tools, for example, it's high up there, but the SEO ranking is low. What does that mean? This is a great opportunity for us to really tap onto that and include it into our marketing campaigns. Now let's go back to the agent. Now that we have a list of high performing keywords, I want to group them into ad groups. So let's see if this agent can do that. So we ask the agent to do that and the agent comes back with a result and again adds it to the canvas. The trick here is any new prompt you give to the agent, the results are always added to the bottom of the canvas. So you need to scroll down. So when we scroll down, what we see on our left is actually the keywords that the agent returned in the past and then it groups them automatically based on the related terms into ad groups and it also tells whether they're new or existing ad groups. That's phenomenal. You don't need to go into Excel, attach formulas, do sorting. No, none of that. Now I want to show you something more. Now you will ask the agent basically to return top search terms. We really want to understand which terms are actually driving engagement and conversions. Now we again scroll to the bottom of the canvas and basically we have all our top search terms sorted again by impressions and click and again from which channel it's coming through. We also want to now take it to the next level. Now we've gone through the workflow of keyword discovery, sorting ad groups, finding which search terms are really doing well, which are not doing well. I want to go to the last kind of the finalizing step of this campaign execution planning workflow and I want to see if this agent can actually create ad copy variations and how well can it do it. So who is excited to see that? Okay, let's go. So we asked the agent to create an ad copy for search campaigns and the goal is to drive Slack Team creations. The topic is project management, so let's See what the agent comes back with. So let's go to the canvas. We again scroll to the bottom and we have 5, not 1, 5 production ready ad copy variations basically along with their headline and description. What does that mean? This means we don't have to spend time doing creative thinking and doing peer reviews for these creatives. Rather, if you spend one more second, I want all of us to just look at this slide and notice something unique. We gave four different or five different prompts to the agent. If you look at the timestamp, the first prompt was four minutes ago. So we gave five different prompts and the agent returned back with five different results within four minutes. That's fabulous. Talking to the growth marketing team, what we understood was this used to take a lot of time, a lot of manual work, working through spreadsheets, sorting data, analyzing data, reviewing creatives. All of that was done with this agent within four minutes. Now I know I blabbered a lot, I talked a lot of things, I showed a lot of things, but I want you all to kind of take a look at the slide and I want to give you a quick recap of what the growth marketer actually did here and how he engaged with our agent. So as a growth marketer, I started with what keyword should I use? And he gave a topic based on that topic. What the agent did was agent went and started looking for those keywords and those keywords were brought to the agent by data that was on the ad tech platforms was ingested into Marketing Intelligence and Data Lake. Objects in Data Cloud were then sent to agentforce where this output was created. Then the marketer actually went ahead and asked to create a bunch of ad group recommendations. The agent used the keywords that had previously shared and grouped them automatically and created those recommendations. And finally it created ad copy variations. Now I want to really talk about what was under the hood. Now you all must be thinking, this seems super easy and not at all complex. It must have been super complex. We had hundred engineers working for six months. No, that's not the case. So I want to show you a quick architecture flow of what really happens here. So as a user, when you're engaging with the agent, the interface is slack. Everything is happening within the slack. You're not jumping between different tools and different tabs. With an agent, the users are then connected to their CRM orgs and AgentForce has topics and actions that are used to pull data using rag retrievers from Data Lake objects from Marketing Intelligence, which were actually created automatically by Marketing Intelligence and On the right side you see different ad tech platforms that we brought the keyword data, search query data analytics data into this agentic workflow. So we use Google Ads to bring in paid media advertising data. We use Google search console to bring in the search queries data. We use Google Analytics to bring in user behavior on websites data. And we use Brightedge to bring in third party SEO data. Now I want to pass it on to Sarah to talk about the big four pillars or core pillars as they say.

Thank you Javesh. And how cool was that demo? Pretty inspiring, right? Awesome. Okay, well as Javesh alluded to, the four tools that powered this agent experience are Data360, Slack, AgentForce and Marketing Intelligence. And what I'm going to do is spend the next few minutes focused on marketing intelligence and, and how this meant that much more time to value for Javesh and his team. Okay, so the first step for Javesh and team to then build their agent was connecting those data sources that he mentioned. And how did they do this? They use Marketing intelligence data pipeline to do this. So what you are actually seeing is a three click workflow. This is actually what it looks like in the app. It is, you just have to authenticate and then you are able to that easily connect your data sources which is a huge unlock. Once your data is connected and it's into data cloud, it's into data360 I should say. The next step is to make sure that it is modeled, transformed and manipulated in a way that you can start to ask questions of your data to start to see how you're actually performing. You've got this consolidated view. So what you are looking at here is the tableau semantic model which is on data 360. So the brains of Marketing intelligence really live here. And what you're looking at is the marketing intelligence out of the box semantic data model. And what this means is that all that data that you're ingesting that's coming in, that is mapped on its own out of the box into the right fields so that you can start to analyze. This is a huge time save for Javesh's team and for marketers. This is custom built for specific marketing use cases and is a huge unlock. And what I like to say about this is this solves like 90, 95%, gets you 95% of the way there. But the beauty of a semantic model like this is that you can then customize it to get it 100% there so it's unique to your business. Which is a huge win for our marketers. So there is a lot more that I could dive into for marketing intelligence. I'm very passionate about it, but I just left it at those two specific examples because those are the use cases Powering the Agent that Javesh walked through. I do want to touch on some takeaways that Javesh and I actually talked through a lot as we built out this content. So I know you are walking around a lot here. You were seeing a lot about an agentic enterprise and building what makes sense for your business and whether you are building an out of the box template or a custom solution like Gevesh and Team built. We boiled it down to three points in three areas that you can start to think about and grasp and get started. The first two kind of go together which is one, your tech stack matters and two, AI Ready Data is the foundation to every agent. The agent is only going to be as good as the data that is powering it and it's so important to get that right when you're thinking about how agents can help your business. This final piece I want to touch on is all about refining, repeating and scaling, scaling your solution. So like I mentioned, our Slack marketing team, they're just getting started. We've already had to iterate. Chavesh and team have already had to figure out how to really make this work for them. And that's okay that we expect that. This is new for everyone. This is new for us at Salesforce. This is new as you start to think about it and that's good, that's okay. But when you're building your solution, make sure that you are building in a way that you can refine as you go so that when you do get it right, you can scale it as needed. Hopefully that helps you start to think about and boil down kind of what we talked about into three just concise points. So as a marketer I do have to leave you with a couple CTAs. This first one, this is a blog that we wrote. This breaks down everything Javesh and I just talked about on stage. We hope you share it with your teams as you go back. Use it as inspiration as you start to think about how agents can help your business. And then the second QR code. This is all about marketing intelligence. We're super excited about marketing intelligence. It lost, it launched this past year and it is doing a ton for marketers. So if you want to learn more, check out that QR code. And then my final takeaway is tomorrow is our marketing cloud keynote. It's at 10am we'll be going over tons of innovations in the marketing space. We hope to see you there. Javesh and I are going to stick around after. If you have any questions, please come see us. We would love to chat.

Thank you.

And with that. Thank you. It.