Operationalizing Marketing as AI-tooling evolves

Most marketers treat AI as another app, not an operating system. Isaac Ferreira, VP of Growth Systems and AI at Shift Paradigm, explains how to build an agentic foundation instead. He breaks down recursive learning loops built on goal, context, measure, and levers, plus signal-based personalization that replaces static ICPs and segments. Ferreira also outlines a use-case-based rollout, starting with one workflow like campaign management, to prove ROI before scaling data infrastructure company-wide. See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.

• 0:00Speaker 00

  The Martech Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast? I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit ihereverything.com.

• 0:25Speaker 01

  From advertising to software as a service to data, across all of our programs and clients, we've seen a 55 to 65% open rate.

• 0:36Speaker 02

  Getting brands authentically integrated into content performs better than TV advertising.

• 0:42Speaker 03

  Typical lifespan of an article is about 24 to 36 hours.

• 0:46Speaker 01

  If we're reaching out to the right person with the right message and a clear call to action, then it's just a matter of timing.

• 0:53Speaker 00

  Welcome to the Martech Podcast, a member of the I Hear Everything Podcast Network. In this podcast, you'll hear the stories of world-class marketers that use technology to drive business results and achieve career success. Here's the host of the Martech Podcast, Benjamin Shapiro.

• 1:16Speaker 04

  97% 97% of executives deployed agents in the past year, but only 29% are seeing ROI, according to writer.com. Most organizations aren't struggling because today's models aren't capable enough. They're struggling because they're treating AI like another application instead of an operating system. The companies that are pulling ahead aren't automating individual tasks. They're redesigning their business processes to think about AI first. The shift requires a new way of organizing your data, building workflows and recursive learning systems. So how can you start building your company's AIOs specifically for marketing? I'm Benjamin Shapiro and today I'm joined by Isaac Ferreira, the VP of Growth Systems and AI at Shift Paradigm, a business and growth partner helping organizations modernize

• 2:00Speaker 04

  their marketing technology and also a sponsor of the Martech Podcast. Isaac has spent more than 25 years building systems across industries ranging from defense, healthcare and enterprise technology. And today he's going to share why marketers need to stop thinking about AI as a set of tools and start treating it as a new way to work. Isaac, welcome back to the Martech Podcast.

• 2:22Speaker 05

  Hey, thanks, Ben. It's great to be here. I'm looking forward to continuing our conversation from the previous podcast.

• 2:27Speaker 04

  Excited to to continue. And for anybody that didn't hear my first conversation with Isaac earlier this year, we talked about sort of the footprint of disruption with AI. And I feel like it's been close to six months since that happened. It's already a brand new world. We've got new models. We've got models hopping out of sandboxes. God knows what's coming next, but it seems like the pace of progress is still, you know, quickening, if that's even possible from where we were at the beginning of this year. When you look at the AI market today, everybody seems to be selling another assistant or an agent. What's the biggest things organizations are getting wrong about AI as it exists today?

• 3:12Speaker 05

  You know, I don't, I don't like to say that they're getting it wrong. I think, I think that they're in an early stage of development, early stage of maturity. And that's just part of where technology came from. And I agree with you, AI development right now is a, is a hockey ship, hockey stick shaped curve. So largely what I see organizations doing is focusing on task-based AI improvement, either task-based or efficiency-based AI improvement alone. When I say task, I mean, how do I use an agent to help streamline my emails? How do I use an agent to help develop a piece of content within an email? The challenge there is obviously the scope is small, so the scope of potential improvement is also small. When there's a massive opportunity, especially with these later models and the contextual

• 3:57Speaker 05

  capability they bring to optimize processes, not just tasks. Processes are where you get the real, not only efficiency, but you get the quality. It's where you get the margin improvements. It's where you get the customer service improvements. It's, it's where kind of AI comes together underneath the stack that you have and helps you develop a unified operating model.

• 4:21Speaker 04

  I feel like there's this notion that you mentioned, we start with tasks and we've kind of iterated more from I'm using ChatGPT to replace Google. And then we have, you know, Claude coworker or GPT, I think they just call it work. And now I'm starting to have it this take over work for me instead of giving me answers. Some people are building agents and having it start to do work. And then there's this next iteration of whatever is beyond agents. How do we make the distinction between adopting AI and operationalizing it?

• 4:56Speaker 05

  Well, I think that most companies are in the adopting stage right now that's using AI as an assistant, maybe to help you search for information or asking it how to do things to put it into your, into your workflows. Maybe, you know, the previous example that I use, let's clean out my inbox and tell me who I should be responding to. Those are kind of like the first two levels, but the last two levels are where you use AI more as, as an employee. And then the final level is using AI kind of as an operating system or using it across all of your workflows. Those are the two levels where we have a really large opportunity to start to mature into.

• 5:34Speaker 04

  We've talked before about this notion of the agentic foundation and you speak fluent enterprise and I feel like I, you know, often make the joke of like, I'm a podcast host working out of his garage, but I'm a solo practitioner who's building his own foundation. And I went through this process at the end of last year, beginning of this year, thinking that I was going to build a chief of staff, right? And an agent that understood my business context that was primarily there to help me with demand generation.

• 6:02Speaker 05

  Yep.

• 6:04Speaker 04

  But over time, it's iterated to build out more processes. It builds my proposals. It does my research. It helps me make my thumbnails, all, all sorts of different great stuff. And now I rely on it to do the research and bring me information instead of me telling it what to do. But there was this agentic foundation component that I was building for myself. What does that look like at the enterprise scale?

• 6:28Speaker 05

  It looks a lot like the foundation that you were building, just with information security and role-based access control and ISO certifications and all the things that enterprise requires.

• 6:37Speaker 04

  SOC 2 compliance.

• 6:38Speaker 05

  Exactly, exactly. It looks like creating a contextual base where AI can understand what is it I'm trying to do? What are the limits on what I can do? What are the operations I'm allowed to perform? What was the context around decisions that were made? What was the performance and what are my improvement goals? And if you have all of those things in a contextual foundation that are readily accessible, then you can literally lay whatever agentic layer on top of it that you need to. That's what we're kind of advocating for some of our larger clients and what we're building for some of our larger clients is a way that they can build an agentic foundation where an agent can be used to solve a problem. And all we have to do is pull the data into the warehouse, trade it on other context and kind of let it go.

• 7:21Speaker 05

  Whereas compared to today, what a lot of companies are doing are saying, I have this task I need done, let me go acquire a piece of SaaS or maybe use it within SaaS I currently have, or maybe let me, you know, code up a quick little app that does it for us.

• 7:36Speaker 05

  It's ungovernable that way and enterprise needs governance.

• 7:40Speaker 04

  When I was building our infrastructure, I, you know, without realizing it, I started with context, right? I was using Claude code and said, hey, you know, here's the systems we use. We've got our production stuff in Monday and my CRM is Pipe Drive and I've got, you know, these records in Google Drives and check my Gmail, all the different systems. We use Slack.

• 8:04Speaker 04

  Everybody uses Slack.

• 8:07Speaker 04

  And so I started with this notion of context first, right? And then what I would do is I would layer on tasks. And then after tasks, there was workflows and, and now we're building other systems that are starting to get better over time. And so I'm on to the portion where I think this is what makes really AI and OS is that it is recursive, right? It, it has learning, it can improve the system based on data. Talk to me a little bit about. The idea of recursive learning systems and how you're not just automating a workflow to repeat, but how you're actually improving that workflow over time.

• 8:43Speaker 05

  Yeah, it's like a, it's an AI-enabled continuous improvement process. It's like lean six sigma, but for, for AI. I, I think top level, what you have to have is, is a goal. What am I trying to improve?

• 8:56Speaker 05

  With that, you have to give it the context. So what are the limitations? What are the targets? What are the reasons I'm doing this? Because AI otherwise is just guessing, right? And so your outputs are going to be guesses. What are the measures I'm using? That'd be the third piece of recursive. So goal contact measure. And then after the goal context measure, it's really an execution phase. And after that, it's performance. How did I actually perform against these measures against the context against the goal? But once I understand that, then I have to feed the AI the levers, which are the things that I can move around to change performance. And which AI can experiment and then go through that whole loop, goal, context, measure, performance, levers.

• 9:38Speaker 05

  That loop itself, if you build that out in your contextual data layer, if you have that data available, the data layer, that loop itself is fairly easy to orchestrate. The hard part's getting the data there.

• 9:49Speaker 04

  There's some, I don't know, trepidation in my mind of building these recursive learning systems. And I'm not sure how much I should give approval towards what the learnings are, or how much we should just say, okay, this is true. And I'll, I'll use the down market podcast host example of, let's say, our podcast titles. We do research on where there is opportunity in YouTube when we create our podcast titles. We try to write where there is blue ocean surrounding us. Then we publish the titles and we look at how they perform. And in theory, our agents should be able to say, okay, that episode didn't perform because that title wasn't that good.

• 10:33Speaker 04

  So here's how we're going to write our titles moving forward, which sounds great. I don't know if I trust it. So I tell it, hey, give me the suggestions first. How much should I just blindly accept? How much do you trust the AI to actually implement the recursive learning? I hope it gets better over time. How do you just have faith in what's going on?

• 10:53Speaker 05

  I think, well, I don't know if you ever just press the I believe button. I'm not to that point yet. I don't think anyone in enterprise is to that point yet. What's really important is that you have a human managed process. I know the term used to be human in the loop. But I think that the trick here is that human continuity is almost always required to make these things function properly. Especially when you're talking about something that's customer facing like titling, right? I wouldn't just trust a, a machine to make up a title that's not going to offend people, for instance. We don't know if it's going to or not, but there are ways to build models, especially for models that are non-customer facing. To make sure that their outputs are within a reasonable, reasonable area of acceptability.

• 11:35Speaker 05

  And you can monitor them with other models, but I would not advocate ever for pulling humans fully out of the loop just yet.

• 11:42Speaker 04

  I built this flow where basically once a week I get this giant document that's like, here's all the things that we think we should change. Can you thumb up or thumbs down them? And you know, my work is going from, I'm responsible for writing these titles to I am approving or rejecting what modifications we're making to our title writing rubric.

• 12:03Speaker 05

  Sure.

• 12:04Speaker 04

  And it's getting very meta. I can't imagine what that's like. Like for me, it's a little overwhelming with all the systems that we've built and all the different ways that we're tweaking and using recursive learning. I can't imagine at the enterprise level, how that is governed, how it's managed. When you're working at, you know, a multi-thousand person organization, how do you keep track of recursive learnings and improvements? And, and, and making sure that you're guiding your AI operating system to get better, but also make sure that you're the one in charge.

• 12:39Speaker 05

  Well, part of it is the, the learning isn't, doesn't have to be auto-accepted and probably in the beginning should not be auto-accepted. So, so learning can be much like, um, I don't know, I have a process where I'm trying to optimize for supply chain. Um, and I have this learning that meant that I improved the supply chain, you know, throughput by out of three percent. So I put that into my dot M D file or my memory file or wherever it's stored. And I prompt the manager, the human manager to look at it and say, is this something that I should store as, um, as a learning that we should use all the time? Or should I not? And then the human can look at that and say, well, you improved that throughput because you reduced total sales, which is obviously not something that we want. We're not going to, we're not going to put that.

• 13:26Speaker 05

  There are processes where the risk level is high that requires validation of learning. And there are processes where risk is low, where you can just let a machine iterate till it gets it right. The problem in the high-risk process is iterating till it gets it right could mean damage to the process, damage to your reputation, damage to supply chain, et cetera. In those processes, you have humans in the loop for the learning.

• 13:50Speaker 04

  I feel like we're seeing this at the bleeding edge of the frontier model development with them, you know, giving where, you know, chat GPT or, or open AI gave one of their newer models a goal of, I don't know, building or hacking something. And it topped out of the sandbox and started breaking into hugging face. We've all heard this story before, but you know, it's like, that's the example at the extreme end of what happens when you just let it run and see what happens.

• 14:16Speaker 05

  Absolutely.

• 14:17Speaker 04

  I want to bring this back to marketing a little bit more, you know, for years, we started talking about the ideal customer profile. And, and now we're getting more into personalization and the idea of actually one-to-one personalization and building campaigns around people. How does AI fundamentally change the process when you're thinking about building this recursive learning and personalization into marketing?

• 14:39Speaker 05

  You know, that, that sounds like a martech question, a technology question, but it's really a, a marketing operations question and a, and the way we're going to market in the future. AI is going to fundamentally change the way that we market. Today, we essentially push marketing out. Here's our campaign. We send a bunch of emails. We'll target you with ads.

• 14:59Speaker 05

  We'll get you with, with media as we're doing right now and, and, and develop interest. The challenge in the future though, is that most digital media is going to be intermediated. We talked about this a little bit in the past, in the, the last, last podcast. Meaning AI will essentially decide what a human sees and doesn't see based on contextual relevance. So that moves us into a world where as marketers, we're starting to market based on signals, not based on predefined journeys. So what we have to learn to do is to, to identify those signals and even create opportunities for the signals to be developed. That means that our original ICP, we might start a little broader, push several versions of marketing out to that ICP and see who's responding

• 15:40Speaker 05

  demographically or with any other attributes and create subsegments within there. But within those subsegments, further develop it. AI allows us to iterate until we're talking to an individual all the way from your TAM down to an individual, uh, based on those, uh, those preferences.

• 15:59Speaker 04

  That's interesting. I feel like we've gone from, we need a really tight TAM. We need to identify the exact process, uh, that we're going to hit with, you know, this segment with these definitions. And we're going to basically send the same message to everybody and see if it responds. And if not, we're going to iterate on the message. And now you're basically saying, you know what? Spread the peanut butter thin and wide. And then see, you know, what sticks to the bread. I'm using a PB and J analogy for some reason.

• 16:28Speaker 05

  I don't know why, but yes, develop the signals.

• 16:30Speaker 04

  I am hungry.

• 16:34Speaker 04

  So, you know, you're going to basically try to see who might be interested in, and whoever sticks their head out of the hole. Then you're going to sort of focus on that individual prospect, which is a very different dynamic than what we've been at for years. Like this notion of personal targeting, you, you still have to have a signal to figure out who is in the market. Give me some examples of ways that companies are effectively doing this now.

• 16:59Speaker 05

  Well, if a company are using tools like six cents to look for signals for someone looking for, you know, a platform on the marketplace or a service in the marketplace. But the other way you could use email. So let's send out four different versions of an email with four different value messages and see who responds to each. And then you're continuing, you're continuing marketing as long as that same value chain instead of just a standardized campaign. Even further, you can get, you know, even greater resolution by saying, well, I have a, I have a dog in this image and this is the value. And this was a text and this person responded and that person has these attributes that now look like this segments. Now let's go push this marketing out to this segment, see what kind of response we get.

• 17:42Speaker 05

  So you have this recursive loop that can be developed that we really didn't have the capabilities, you know, to do previously at scale.

• 17:48Speaker 04

  I have this theory that segments are actually going away.

• 17:51Speaker 05

  Yeah.

• 17:51Speaker 04

  Right? And, and what you're saying is, look, you've got your total addressable market, do something that hits everybody. Anybody that gives any sort of indication of interest, now you can create a customized campaign that is specific to them. And so this notion of identifying multiple target markets essentially goes away. It's like, not even a target persona, it's a target person. And you're creating.

• 18:14Speaker 05

  And at the lowest level, absolutely. What's interesting though, is you can use the people that, that respond, the signals that respond to develop segments to get more individuals. Instead of starting with a segment, assuming that's going to work and trying to figure out what fits, you can invert the, invert the process.

• 18:32Speaker 04

  So help me connect the dots here. We talked about how do you build that foundation? You've got this data infrastructure. You are building your operating system to help bring the information to you. Are we to the point where AI is starting to help us build our campaigns and doing this like micro-personalization? Who's doing it? Well, how does that actually work?

• 18:53Speaker 05

  Yeah, I think we're getting there. Adobe is, is probably the farthest along from what I can see. They've actually taken the AI as OS idea and layered Agentic AI under their entire system. And their ultimate goal is obviously to be able to go to a command line and say, I want to build this campaign for these people doing X. Um, and it uses all of its systems, all of its creative capabilities, the data and its real-time CVP. It's, uh, it's marketing and activation capability and develops the campaign, runs a campaign with human, uh, uh, input tests that iterates on it. That's, that's probably the, the company that's farthest along from a platform standpoint. There are also individual companies who are starting to do this for themselves.

• 19:36Speaker 05

  One of them is one of our clients where we're building a system where we can actually automate, um, all of those steps from a centralized console.

• 19:46Speaker 05

  And that's kind of, once you have the platform, you can build the agents a piece at a time. So it's not this giant bite that costs millions of dollars and tons of time. It could be, you know, at first I want to just, let's develop the campaign brief and push the campaign brief into our project management system, step one.

• 20:02Speaker 04

  There's this idea that my AI will help me identify my prospects. It will create the campaigns for me. It will execute them and I can go to a beach, have a Mai Tai and then check an AI-generated dashboard and see how everything did.

• 20:22Speaker 04

  And the reality is, we're not there yet. So how do we meet the enterprise marketers where they are so they can take the first step in going from, I've got my static dashboards, I've got my legacy processes and I am chat GPTing to try to get some information done, to building these workflows? What's the first step they could take today? What's the order of operations to get to the beach with a drink in their hand?

• 20:49Speaker 05

  Well, they need to do what you did because you're about to go to the beach with a drink in your hand.

• 20:53Speaker 04

  Ah, not yet, but for everybody listening, this is my last meeting before I go to Hawaii. I will come back in the next podcast and be much more tan.

• 21:00Speaker 05

  So the trick is building that agentic foundation and that, I know I keep saying this over and over again. You'll see a lot of people on LinkedIn and then a lot of other folks who talk about this saying, you've got to get your data right, you've got to get your data right. But what does getting your data right mean in an agentic context? Getting your data right means I need to have access to the data, which we generally have. If you have static dashboards, generally speaking, you have an observable enterprise and you have a lot of data. What I have to do now is put it into a form that allows my AI to understand what it is. So I'm looking at this data, what is it? Where did it come from? What are the limits on it? Why is it important to me? Then I have to tell the AI with this data, you can do these functions. And then when I tell the AI what functions it can do, and by the way, you have to follow these roles.

• 21:43Speaker 05

  And once I get those things together and I can start pulling in performance, then I have a process where any operational capability I need, I have to pull the data into that structure, build the agent, connect it, test it and let itself improve or improve with the human in the loop. Whereas before the process was, I'm going to go buy a piece of technology. It might not do what I want, but it might do some of it and I'm going to go buy it and I'm going to train it and I'm going to get some limited adoption on it and it might work, it might not. But now I'm going to have to support it. And now I've got 300 tools in my arsenal that are not being utilized, but they all have AI. It's just, it's a much more sustainable way to build, even though it seems like a bigger bite up front.

• 22:24Speaker 04

  I had this conversation with my boss. His name's Ben and I went to Ben and said, look, we're not going to get shit done for like three weeks because I'm going to be unifying our very little set of data across a very few set of systems with only one person building this out. And Ben said, God, you're stubborn, but do it because there's no stopping you.

• 22:51Speaker 04

  I had a nice conversation with myself.

• 22:54Speaker 04

  At an enterprise level, you actually have to go to somebody with a different name who signs the paychecks or digitally approves what you're getting paid. That's a tough sell, apparently of saying, hey, look, we're going to be less efficient, probably not for three weeks, for three to six months to go unify all of our data and build this contextual layer. So then we could start building out our agents. So then three months later, they're really doing work to getting back to probably where we are right now. But fundamentally, we all know it's the right thing. Like we all know that we should be building out these operating systems. So if, when you're working with people that are in the enterprise, how are you helping them make the case

• 23:40Speaker 04

  that they should make an investment in this infrastructure that helps you build the agentic layer, that helps you build the recursive system that helps you get to the beach with your Mai Tai?

• 23:52Speaker 05

  So I think the trick is don't fight that battle. We've known for over a decade that we needed to unify our data in one place that was accessible. But getting the corporate will to spend the money and time on it has been challenging. At almost every company I've ever worked with, in or with. So the way that we proposed before with this is as use case based. All right, I see that you-

• 24:14Speaker 04

  Give me some examples.

• 24:15Speaker 05

  Yeah, yeah, okay. I see that you need to operationalize your project management system for campaign development. Right? I don't have to get all the company's data into one place. I have to get the data I need for campaign management into one place. So build out the base layers, get the campaign management data in, build the agents to assist the process, run it, show the value. Once you've run it and showed the value, what's the next use case? I don't know. Let's say I want to automate my content supply chain. Well, I can build on top of the project management system I just had and manage it through that. Let's go get all the data I need for content supply chain into this system and let's automate that, get the value from it.

• 24:58Speaker 05

  And with that, you start to be able to build this, this snowball that's growing of scalability that you wouldn't really have in the past without having this massive disruption to your operations in six months of no efficiency and all the things that come with centralizing all your data. I would, I do not now and I would never say, go to a company and say, centralize your data. I want you to give me $10 million to fix your data problems because one, it will never complete. Two, I will never, never show an ROI and three, you're going to throw me out.

• 25:27Speaker 04

  I almost threw myself out at the end of our integration and it was a couple of weeks where I really just, without realizing it, I just kind of went down a wormhole and was like, you know what? I just feel like I'm going to need, I want to build this chief of staff. I wanted to understand, you know, revenue and marketing and content supply and all of the things.

• 25:49Speaker 05

  You have the scale, you can do it all.

• 25:52Speaker 04

  Yeah, the scale where I can spend three weeks to get all these systems. What I learned, and I like your snowball methodology, your snowball, and I like your snowball metaphor. In the sense of when I did the initial implementation, it was a pain in the butt. Setting up databases and stuff that I don't even understand, how does it work? I don't know, Claude did it. I hope you guys understand, I'm sure you understand better than I do how the infrastructure is actually built, or at least I would hope so. But now every time I have to add a different data component, the answer is always, okay, cool. We've done this 10 times, we have the learning, we, Claude. Right? I understand how to wire all this data in, it's going into the existing system.

• 26:34Speaker 04

  Let me just add new columns essentially into the big table that you have. So it does get easier over time, I think is my point.

• 26:42Speaker 05

  It gets much easier over time. One, you've developed a process to start with and every time you iterate that process becomes easier to execute. And it becomes a matter of what are the attributes for this data that I need to store? How are we getting into the system and who's doing the work? And once you have that part down, the data side is relatively easy. It's the developing the process side that takes the thought because it's something new that we haven't really done before. We're, we're trusting machine to do large parts of activity for us.

• 27:09Speaker 04

  One of the things that I have learned is, okay, once you have this infrastructure, you replace your existing workflows first and then you evaluate them and then you start improving them. Is that the same thing that happens at enterprise?

• 27:26Speaker 05

  It can. It depends how big of a bite a company wants to take and how risky the process is. If you have a high-risk customer-facing process or a revenue impacting process, it's generally speaking, lower risk just to copy it and get it into an agentic format with humans. If you have a lower risk process, my recommendation is to go straight to an agentic enablement because you have the flexibility to learn as you go.

• 27:52Speaker 04

  I find, it's funny, we're rebuilding Podcast OS, our production infrastructure. We're always rebuilding our production infrastructure because it keeps getting better. And it started off with all these manual workflows and we took what we were doing manually. We built Zapier workflows out of them. And then we're taking those Zapier workflows, which are fine, the models got better, the content got better. But we're replacing them with these agentic learning systems that are doing research to get more data into the workflow so we can make sure that what we're saying is right, that content we're talking about is interesting to our audience. But it was always built off of this understanding of what we're trying to replace. So even though I might be replacing a workflow and, and just, you know, making it,

• 28:37Speaker 04

  not just replacing it and making it do the same thing, but replacing the ethos of what it's supposed to be for, I still have to educate the models and my agents on what's the problem I'm trying to solve? Why does this workflow exist and why was it built the way it was?

• 28:51Speaker 05

  Yep. And what are, and why were these decisions made that were made? And what was the result of the decisions made that were made? And how should I approve it? What are my goals?

• 29:00Speaker 04

  Going back to making this actionable for marketers, there's building the data layer, making sure that you're carving off one use case to try to implement an existing workflow. And then the last piece is, how do you start to show ROI? You're carving off a use case to, to go from where you are to some sort of agentic workflow? Where does the revenue actually show up so you can prove that you should do this with the rest of your models?

• 29:26Speaker 05

  You know, it's interesting. I think in the beginning, not I think, in the beginning, we all saw companies running for dollars per FTE. So they're going for labor efficiency, which is essentially a savings maneuver to justify the cost of AI. I don't really see that as the true value of AI today. In fact, you know, a lot of the clients we work with, it's how can I do a heck of a lot more with the people that I have? Because then what I'm doing is I'm developing margin, I'm developing throughput, I'm developing speed to market, I'm developing customer, customer success, customer happiness in a way that I couldn't do before. So the ROI isn't how many FTE can I cut in my mind? The ROI is how do I improve revenue? How do I increase efficiency?

• 30:11Speaker 05

  And how do I drive margin? And AI is a massive margin enhancer. I mean, some of these solutions that we've built for our clients have driven, you know, 10, 15-point margin improvements without getting rid of people. It's just that you don't have to go that direction.

• 30:30Speaker 04

  It's interesting. I feel like we are at the portion now and I agree with you and you speak fluent enterprise, like I said, and, and you know, I am down market from that, but I see the same thing where I can replace all of the SaaS tools that I was using for our infrastructure, so my margins are getting better. And I try to stay away from the stuff where my team is working on it because I want my team to still be doing their jobs and I'm trying to make their jobs easier. But at some point, we get to the level of maturity and sophistication with artificial intelligence where the AI can do what the people are doing. How far do you think we get, you know, and how far off are we from saying, look, there's just one person at the head that is replacing all of the human labor and maybe there's a couple of human in the loops checks?

• 31:18Speaker 05

  Probably a few years from that. I think right now the challenge is context and management and quality, right? I don't see it happening next year that we're just replacing all of our marketers with AI. But what our challenge just by any company was is if you get to a point where you have one marketer running your entire marketing organization and you have these great people who understand your company, what's the next product you want to sell? Go develop it. You know, what's, what's, what's the next innovation you want to develop? Go develop it. Like AI should be a tool for growth, not, not, not a tool where we fire everyone and stop growing, which doesn't make a lot of sense in my mind. You need humans for that, that stimulative growth.

• 31:57Speaker 04

  I think that there is the combination of humanity and artificial intelligence that allows us to seek growth and that AI can be used essentially for good, not to replace the labor force, but to use the labor force to grow the business. That's the fundamental goal that we're all focusing on.

• 32:14Speaker 05

  My goal in every single engagement at the enterprise level is how to show a true ROI, a real ROI projection and actually an execution without having to use labor to justify it because there's so much there other than labor.

• 32:31Speaker 04

  And that wraps up this episode of the Martech Podcast. Thanks for listening to my conversation with Isaac Ferreira, VP of Growth Systems and AI at Shift Paradigm. If you'd like to get in touch with Isaac, you could find a link to his LinkedIn profile or check out our show notes on martechpod.com or you could visit his company's website, which is shiftparadigm.com. If you haven't subscribed yet and you want a daily stream of marketing and technology knowledge in your podcast feed, hit the subscribe button in your podcast app or on YouTube and we'll be back in your feed next week. All right, that's it for today. But until next time, my advice is to just focus on keeping your customers happy.

• 33:32Speaker 00

  Thanks for listening to the Martech Podcast and I Hear Everything production. Looking to launch or scale a podcast like this one for your brand? Then visit ihereverything.com.