00:00:01 — 00:02:08

We left off with a live wire. If an agent makes the wrong call, the accountability doesn't disappear. It lands on whoever wasn't watching. Today, Peter and I go further into what it actually takes to close the gap between the pitch and the production reality and who's actually making moves he says the industry needs.

If a map takes your advice and becomes a pure execution layer, it officially stops owning the data. It narrows its own surface area, and its own AI features can worsen because it's only seeing a slice of customer behavior. It's a real trade off for the vendor, but it's actually a big win for the customer in a lot of different ways.

So make the case for why a marketing automation platform CEO should voluntarily walk into this. I think at the outset, like very few will. Um, but I guess a question I would ask them is, do you want Sustainable growth? Or are you willing to take sustainable growth over large contracts up front with churn down the line?

And the reason why I say that is like if a marketing automation platform CEO is hyper focused on being the best execution layer they can be. Then the contract size might be lower to start, but I would expect it to increase over time because I expect that your messaging more people, um, and they're also not frustrated that you're charging them to store their data and then they ultimately leave.

Um, and that's in contrast to let's sell all the bells and whistles and then churn out of those bells and whistles as the organization finds. Oh, actually, like the AI agents that I have access to inside of here only live inside of these four walls and don't have access to the things that I'm not even thinking about or didn't know existed inside of the data warehouse.

And I needed to work with a data and engineering team to figure that out. Right. And so

00:02:09 — 00:02:26

when we're talking about, like, predictive models, anything AI agent, um, I expect that a lot of people will buy those to start and then eventually they'll not renew them. And now your customer success team just took a churn

00:02:27 — 00:03:05

because they don't have a leg to stand on. When the customer says, well, we do all this upstream anyway, and we're really getting a less strong model because it doesn't have access to the same amount of data, because your contracts require me to reduce the amount of data that I have inside of here to make you affordable, right?

And so I would just say, yeah, you're going to take an initial hit. The contract sizes will be lower, but you would have less churn. Quite a catch 22 for the vendors. It's either evolve and prepare or

00:03:06 — 00:04:25

stay stagnant and wait it out and see what happens. Right. Like the obvious answer there is like just connect to the data warehouse. Make your platform data warehouse native. Yeah, that takes a lot of work. So like, I understand why the traditional marketing automation platforms either are taking a long time to do that or won't do that, because for sure, the data touches every surface inside of your platform.

So when you change that model, you have to make it work with everything. Yes, I'm building on that. So the the infamous got Brinker and Franz Ramirez in their state of martech research this year. They found that 90% of marketing orgs use AI agents somewhere, but only 23% of them have it in full production, so the rest are stuck in this pilot mode or assist only mode.

And if the dumb pipe model hands more decisioning to AI agents is your thesis. This gap closes fast. There's a lot to unpack there. Um, I would also note, like Scott Brinker has broken my brain a few times in the last year with a lot of the work that he's doing around AI agents and AI in in marketing. So

00:04:26 — 00:04:47

here's what I would say is nearly everyone that I talk to has some AI initiative at their organization. And I'll quote a former manager of mine, Katie Behrens, is like the question that people need to ask is, what is the highest and best use of my time right now?

00:04:49 — 00:05:57

And to take that a step further, I think you also have to use that same question with agents like what are they actually good at? The challenge that we have here is that it, simply put, takes a lot of time to figure out for your business context and for the data that you have, what are these agents going to actually be good at?

Um, because every business is slightly different, right? Like they have a different data setup. Maybe they have a different loyalty program structure, like whatever your flavor is. Um, and I want to be clear, like I am not suggesting or never will suggest that all decisions just get handed to an agent and you say, all right, run with it.

Right. Um, but what I am suggesting is that today's decision making, in many cases, is going to be limited to the data that exists inside of whatever platform that you're using. I'm going to push and add. It's also limited by the creativity and

00:05:58 — 00:10:13

development of the individuals, because if you can't recognize the very simple tasks you're doing on a daily, weekly, monthly basis that are repeatable and are straightforward, and instead you're focusing on the advanced multi-step multi thing, you have to build it in brick by brick. And it's similar is that investment for a dividend.

And you really have to invest there. I completely agree. Yeah. Um I'm not generally a fan of like reading dusty old white dude books. Um, but me either. But I will say like that your statement reminds me of like the The Effective executive, which is like, if we don't understand where our time is going, then we have no way of changing the way that we operate for good.

Right. And so I would say, yeah, you have to really understand what am I spending my time on? Is that a good use of my time? What does an agent need to understand in order for them to do this, as opposed to me to do this? And then how do I need to wrestle with this agent in order to get the output that I wanted and not stop at good enough?

Like you stop at this? Is production ready? So like I would posit that like, this gap is gonna close fast. But it does take time and effort on the part of the people using these agents. Yeah, it's the concept of like, player coach. You need to be the coach of your agent player because they're going to make mistakes or there's going to be this unusual play that happens in a corner case and you're like, oh gosh, how did we did not prepare for this?

We need to rethink how we approach when something like this happens. I see some platforms maybe getting better at the agent side of things, but, um, it doesn't feel like it's enough. Like there's not enough context for these agents to be really good. Um, and most of them just end up looking like, uh, especially on the generative side, the crap that gets thrown out there, like, looks like any Gemini or Claude or, like, fill in the blank on your AI harness.

Like, all emails start to look the same. Um, and that comes down to it's really difficult to build in guardrails and brand guidelines that are worth their weight. Right? Um, it's just tough. It takes a lot of time. The AI harness, like there's the large language model and then there's the AI harness. Right.

Which is how do you drive the model? Like if you're talking about creating a campaign, the AI harness would be the one that breaks down. We start with an idea that also has a goal. Now we connect that to data like so. It's the order of operations, um, is what the AI harness would handle. Um, and ultimately, like you need to train that AI harness how to do what you do manually every day on a daily basis if you want to, or almost like your company objectives and goals.

It's like you have an OKR for your agent and then for the individual agents. Hey, I need to answer basically a campaign brief if this is exactly what I need you to do and not do and things like that. Exactly right. And if you don't have like that really good AI harness. With all the guardrails, you end up getting different outputs every single time you ask the same question.

Correct. And that can be really frustrating. Um, and so I don't blame marketers for, for going in and saying, oh, I tried this thing. It didn't give me the output that I want. And so therefore the product is broken. It's like it takes more than that. Like, let's not say that the product is broken yet. Let's also, again, looking inward is really important and saying, did I give it everything?

00:10:16 — 00:13:27

Hard fought token costs are rising in the post funding subsidization. And it's it's coming to an end seemingly. Maybe I don't know. I don't know how much more open free cash there is in the world because it doesn't feel like it. And you have said that the price and volume economics don't work at the current scale, so isn't let the agents make the decisioning calls a thesis that's economically unproven right now and not just architecturally proven.

I think that's fair, right? And admittedly, that's not an area where I'm spending a lot of time focusing is like, how much does it cost for me to use an agent for this versus using a human? I think, though, that again, going back to that statement of like, what's your highest and best, you know, save the really difficult things that you maybe wouldn't even be able to do, like analyzing massive sets of data.

Like you can't do that efficiently. I don't know a lot of people who can. When we're talking about at the scale of data that or enterprise organizations have. So I guess what I would say is we don't have to use agents for everything. Like if it's just a simple trigger, like you don't need an agent for that necessarily, or if it's a simple rule, you don't need an agent for that.

But where I think it's useful is when you have a problem that you've only ever dreamed of solving, and your brain starts to break. When you think about solving that problem or you don't even know where to start. That's where I think being a thought partner, at the very least with an AI agent, becomes really useful.

The economics of it. Yeah, it's it's unproven right now. I don't think anyone knows what's going to happen with token costs in the next two weeks, let alone two years. But, um, have to be really smart with where you dedicate your time and budget for these things. Um, so you you do have to be tracking it. Oh, yeah.

I mean, I know so many instances where the enterprise is like, use AI for everything, and now they're receiving monthly budget limits per cost center per department, and people are freaking out like you gave me crack, and now I cannot use it, I. You're not allowing me to do what I was doing. Enabling it. Well, yeah.

And if you're in the finance department, if you don't have a tool that allows you to easily track that and set guardrails, like maybe not the right tool for you. Exactly. Brought to you by our sponsors. If there's one theme that's followed me at every stop in my martech career, it's trying to get good data into the hands of marketers.

That's why I'm so excited to tell you about our sponsor, High Touch, the leading composable CDP and AI decisioning platform companies like Domino's, Chime and Ritzy and PetSmart trust high touch to power their data. And here's the kicker 90% of customers have a real use case live in production within their first week.

That means you can implement a world class CDP in months rather than the usual years long headache. That's why top brands choose high touch to personalize every customer interaction at scale. See what high touch can do for you at high Touch.

00:13:28 — 00:14:42

Awesome. And now back to the hot seat. We've been getting super duper nerdy in the best possible way. AKA my favorite. And what are everyday conversations are like? So taking a step back to our industry at large, or maybe like the main line if we're talking plumbing. So the map market is no longer a theoretical debate.

It's a live restructuring with winners and real losers or casualties. Depends on how you look at it. And let's name what's actually happening, not what everyone's decks are saying. So your dumb pipe thesis has a dependency problem. You still need a map for email sending workflows, forms, and campaign management.

So the composable stack doesn't eliminate maps, it just denotes them a bit and changes their role. Does that mean that the business model requires maps to survive? And isn't that true of every composable or reverse ETL vendor right now. And this is not inclusive and solely just about high touch. I think broadly, what we're going to see in the next couple of years is

00:14:43 — 00:17:08

CDPs or composable platforms start to do some of what marketing automation platforms do and vice versa. Right. Like we're already seeing that in the space. It's been a race to the middle for years. Exactly. The convergence, even just the term CEP, all of these things are converging and also extra confusing amongst alphabet soup.

Totally. So everything regresses to the mean. But I would also say that it's not really true today that the composable stack requires a true marketing automation platform. Like I have, I've spoken with many people and know a lot of organizations that go direct to an MTA, and they manage templates themselves like that is possible for you to do.

You absolutely need to have the right butts and seats on your operations and your engineering team in order to solve that. You could probably meet a lot of those people if you go to like Twilio signal, like those are all people that are just building it. Like the people that go direct to infrastructure are a real special breed.

They are fascinating people, super smart people. Um, but that's a specific type of organization. So I think it's not true today that a map is required in a composable stack. Um, nor will it be tomorrow, because I expect that, you know, the data activation side, um, converges with the map side. And we're art like we're seeing this on both ends.

Like Treasure Data has engaged studio Bres has CDI and zero copy personalization like they're already working on this stuff. It's just like, how much are the lines going to blur moving forward, and who's going to be the best of both worlds? Um, forms are maybe a separate issue. Like a lot of marketing operation, marketing automation platforms themselves don't actually solve this problem at the level needed.

Um, maybe another feather in the cap for Salesforce marketing Cloud. Cloud pages was like pretty awesome. Absolutely not. Know that you're not getting away with that one. It's a bridge too far, I know I crossed that's crossing the line.

00:17:10 — 00:17:49

People need it. People use it. I'm just saying, oh, I'm not saying they don't need forms, but Cloud Pages is not the answer. I don't disagree there. I don't disagree, especially when you start breaking down what super messages actually are. Oh right. Yeah, that uses a super message. But it's not a message.

But it's still a super message. Yep. Yep. Okay. In your essay that has, I think, rocked the martech world in the best way, between all of the debates that are happening, the private DMs, the messages, and really, I think it's been an illuminating moment for those who just haven't quite gotten there yet or really

00:17:51 — 00:19:29

put the words in such a succinct manner. I think everyone had thoughts of this or or worries, but didn't really understand that it's kind of happening. That's it. You did also give for reasons and laid out the reasons for maps to survive, and you were very clear. And so just for those who haven't read the essay, we've got build bigger API's, own more of the funnel without owning the data Support both head and headless interfaces and stop forcing customers to store data inside of the platform.

Every single one of these cannibalize an existing revenue stream or completely kills a roadmap item. Which one is the hardest sell internally? The one a map's own sales team fights hardest against, and why? I kind of see this specifically on the sales side. Like I kind of see it as being the opposite. Every sales team wants to have all four of these things.

Um, I think though, if we're talking about which is the hardest to sell internally, it has to be storing the data inside of the platform. Um, because and this is not a sales point, I think, again, we're seeing more and more RFPs where one of the top line asks is, can you operate directly off of my data platform data.

So like they want to be able to check that yes box. Um, so

00:19:30 — 00:20:58

I would say it's probably where the data lives if you have a solid MCP server. Like, I don't think your sales team particularly cares whether or not someone logs into the platform or not. Um, I could see customer support and, uh, customer success and product caring about that, right? Because they have metrics that they have to work against.

Um, especially on the product side. Like is your product piece that you've if your product feature that you've developed actually making money, is it being used? Um, they are held to a higher standard. Their, um, bigger API pipes that allows you to sell into organizations that you couldn't before. Right.

So like You're selling into the organizations that, in my experience, have to, because of their needs, build their own setup and go direct to an MTA. Nothing bothers me more than really small API limits. Oh yeah. It's so frustrating and limiting. It's it's incredibly reasonable. I mean, I have questions about why are you sending this much, but it's incredibly reasonable for an organization to say, I only want to trigger messages, and I want to trigger 150 billion of them a year.

00:21:01 — 00:22:24

Like R.I.P your inbox. Yeah. I have questions with the overall strategy. Right. But there are certainly organizations out there where that's a reality. Right. If you're talking at the scale of of like an Amazon and granted like they have Amazon SES so like, they own their own pipes and, you know, they drink their own children use other maps.

Just kind of saying that multiple actually, they definitely do. And organizations of that size tend to. Right. Um, but it's reasonable to ask that question, um, even owning the data, like I think so long as the salespeople and the engineers and the products are getting the product, people are getting the feedback that, hey, this is awesome, and I can activate my data in the way that we need to inside of this map.

Like, I don't think it's a tough sell internally. I do think it's a tough thing to actually build. I feel like it's a tough sell. Top down, bottom up. Makes sense. Sure. Yeah. The frontline people. That's worth noting. Like the frontline folks who are in my DMs or are texting me, they're like, yes, okay, great.

Like the marketing ops people, they get me. Um, and I get them.

00:22:26 — 00:23:35

It's the it's the higher ups where sometimes it's like, no, I don't see that in a world, because then how do we be strategic? Like, we don't want to be the dumb pipe. We want to be the brain for your marketing. And I'm sorry, everyone can't be the brain. You can't have five brains. That doesn't work. There's a reason why we don't have five brains.

Not five brains. Exactly. Yeah, you got a lot to digest. Five stomachs? Totally fine. Um, I have brains. I feel like that would be difficult. Agreed. So if maps really do shrink in the future to execution only does that kill the martech job market as we know it? Like the strategist, the ops person, the platform admin, all of it.

I mean, I'm not super doom and gloom here. Like, do I think in general from AI, there's going to be a general reduction in terms of the total headcount that we need across every industry. Yes, I think that's probably a realistic expectation. That's also been super well documented. Um, I do think, though,

00:23:36 — 00:25:22

what this opens up is those same functions may be operating in a different environment. So like instead of being masters of tools like Salesforce Marketing Cloud, Bray's iterable fill in the blank like they become masters of their organizational data. I also think like this opens up avenues that didn't exist before.

I'm a marketing ops guy. I am not a creative like you asked me to do. Creative. Maybe I could do some copywriting. Like that's basically where it ends. But like, I'm not that full, creative, sweet person. But if I have a thought partner in an AI agent that's actually good at creating on brand imagery and things like that.

Maybe I can dip my toe into the creative side. So, like, to answer your question, yes, there's probably a reduction in force across every industry known to humanity right now. Um, but I think there is still time spent in the other things that we've talked about before, like making sure that these agents and guardrails are being reined in and having their work checked, and that we're building the guardrails to make it have a good output.

Um, they're just not solving the same problems that they are today. Yeah. New problems require new solutions and up leveling and upskilling in a lot of ways. Mhm. All right I'm going to ask the spiciest question of the hour. So which ISPs are actually adapting shrinking towards execution, opening their APIs and getting out of the data business in which a refusing name, who's adapting and who's stalling.

And if you won't answer, tell me why.

00:25:23 — 00:25:25

I mean, blanket statement.

00:25:26 — 00:27:24

Pretty much no one's adapting in the way that they need to yet, so that's why I won't call out a specific, uh, organization. Um, what I do think, like, let's talk about the shifts that we're already seeing. Zero copy, composable warehouse native. Like, fill in the blank. That's the new alphabet soup, by the way, is like, what do you call just building?

What do you call building on top of the data warehouse? Right. I call it modularity because I like being able to plug and play different bricks of so I can create the best in breed, no matter what it is for that set of business. Yep. Lego blocks and like the promise of the Lego block is that it connects with other Lego blocks, right?

What we don't need are is a composable approach that is connecting a Lego block to a Duplo to a magnet tile. Right. Like zero copy and modularity is getting talked about more and more. This is a good thing for everyone involved. Where it falls short is often that it's only partially modular or composable.

Like I can create an audience without moving the data. But if I want to use personalization tokens, merge parameters, whatever you want to call them, dynamic content inside of a message. Like inherently there has to be some kind of data transfer, right? Um, so that part is really challenging. So like it ends up being more warehouse connected than warehouse native, right?

Um, what I would leave the audience with is like to the organizations that are evaluating these platforms. Some important questions to be asking, like, can you build and activate an audience using, you know, 50 million rows from a warehouse table without copying the data into the platform?

00:27:25 — 00:27:34

When building the audiences using an agent, can you show me the generated SQL that it's using so I can validate it?

00:27:36 — 00:30:33

What does latency look like? Like what is the latency? Um, and by the way, saying it all needs to be real time is a cop out. That is not real. Stop saying that. There's so few use cases where real time is actually needed. Password reset. Yes. Right. One time password. Right. Yeah. It is so rare. And yet I think because of, uh, you know, inventing, becoming a lot faster in the industry.

We're all chasing that next thing. Like we can't get enough sexy to say this is in real time when it's like, you know what? A schedule of, like, once a day or twice a day is more than enough. Yeah. Think about the use case. Think about the customer experience. Think about what is actually needed. But to say that it's all needs to be real time.

That's just categorically false. Like just stop it. And also a waste of resources. Every stretch you're going to pay for it. Like you can get real time on everything, but you're gonna pay through the nose for it. Um, and then I think the, the final thing to be asking, if you're really invested in a composable or a modular approach is what are the artifacts that persist when I send the message?

Because that's always the place where the most artifacts are going to exist. What exists? How long does it persist? Where does it persist? Um, and what controls do I have over that? And then finally, like, is this just an audience builder or can I actually do personalization with this? Can I go beyond who and making a decision in a branch split to, oh, now I can get inside of the content of the message and personalize it without storing data inside of the platform.

I loved your answer, even though it was a cop out as well. It's a total cop out. No one's getting free advice or kudos here. Uh, it's just not going to happen. I think there are two platforms that do deserve at least a shout out, and they've already been mentioned via Luke Ambrosini in some capacity. Just because Luke gets it and he's always gotten it.

Message gears did start this years and years and years ago. This is not new, but to your point, they never really capitalized on making it friendly for the marketer. It's primarily friendly for the engineer, and it's a you need to satisfy both personas. Yeah, and I will say it's not the most proven platform yet, but Salzman is a is data warehouse native and they are doing what they're doing.

And so there are two platforms, one far more mature and one far less. But

00:30:34 — 00:32:41

it all matters on what your requirements and needs are and what already exists before you have to uplevel, because this is up leveling your entire infrastructure and totally puts you in. It's like we we prefer to like legacy players, like Marketing Cloud and Pardot and Marketo and then like next gen as iterable Bray's boom reach, you name it.

And it's like, okay, what is this next frontier? Because next gen Is next gen. There's no question about it. But this is a completely different business model and also operating model. And we've got to coin that term at some point. Yeah I don't know what it is. Like um C copy model CMP composable marketing platform I don't know.

We're quoting it here. Zero copy marketing I don't know zc, CCF. So if an agent is deciding channel timing, message content, what's actually left for a human marketer to do in five years? So many things. You and I talk about this a lot, but, um, agents don't have taste. Um, and agents and AI in general is inherently looking in a rearview mirror.

Right. So the things that humans are doing there. number one. Obviously, they're reviewing agents calls. Um, so the decisions that agents make. You know, everyone calls that human in the loop, right? Um, which I kind of roll my eyes at. But it's true. Like you need to have a human in this process. Yeah. Um, but they're inserting their taste into the equation.

Um, and they're looking forward and determining how they are going to steer the business. Right. Because of that rear facing nature of context, it's. What's the next thing? Um, what are the things that haven't happened yet in our business context that I need to now prepare the agent for? If you're using an agent.

Um,

00:32:42 — 00:33:20

and also, like, if you're getting into a new channel, like, how can you evaluate emerging channels like, say, RCS? Right. Without that foundational understanding of where do I want to take this? Because an agent can be your thought partner, but they're probably not going to be that good at coming up with the first thing to test.

Um, you need to do that without a doubt. I, um, I just think it means a combination of you get more time to be more strategic, but

00:33:21 — 00:34:17

you have to get the foundation right. And that's honestly the hardest part for any and every aspect of marketing. Because so few I have that. Yeah. It's also like a conversation that a lot of marketers have been edged out of, uh, correct for the wrong reasons. Like, they are absolutely a valuable person to have at the table.

I would love to see need to have a seat at the table. Otherwise you're misinformed. And I think, you know, that's incumbent on the data and engineering side, making sure that they have a seat. But it's also incumbent on the marketers to say, I like have some sharp elbows and say like, hey, I need to have a seat at this table.

Yeah. Agreed. Well, Peter, this conversation, in my personal opinion, is what everyone needed to hear. What I've been wanting to hear in a less Babli way that we normally talk

00:34:18 — 00:35:10

in a more streamlined approach. And so it's really, I think, the conversation that particularly those who've held the keys to a map SP ep and it's it's interesting because you're both an insider and have done things on the outside, both as a marketer and on the agency side. And so we'll definitely be watching whether this timeline you're hedging proves to be conservative, too generous.

And I think this level of critique is so valuable because if we cannot self-reflect on where we are and where we can be going And where the market is already choosing and going. That is the downfall of any business model, product, platform, system, way of thinking. Yeah. And before I let you go, who is someone we should have on the podcast?

00:35:11 — 00:36:32

I namechecked her already in this episode, but you need to have Katie Barron's on this show. Um, Katie Love Katie was a manager of mine at iterable. I would love to see her on the Making Sense of MarTech podcast. I would too and talking about taste, she has completely transformed my wardrobe and it's all her fault.

And she already knows this. She is a tastemaker that is undeniable, without a doubt. Well, Peter, thank you so much for coming on. Where can folks find you and also listen to more of your musings. Um, mostly on LinkedIn is probably the best place to see me. Yes, we miss Professor Pete. I miss Professor Pete.

It's gonna come back. I don't know when in what format, but, like, low production video is my love language to this industry. Um, I just I love talking about things, um, in a really unstructured format and just rambling. I never edit anything, so I'm there. DM me like anyone who's DM'd me on LinkedIn or any other channel knows that I respond.

Um. And I would love to chat.

00:36:33 — 00:38:39

We covered a lot of ground today and in part one, and the uncomfortable truth here is most of it is actually really fixable. So here's where to start. The data tax argument cuts both ways. Maps charge to store data you already own, but composable stacks carried their own headcount cost. The difference truly is ownership and not price.

And that means before you sign any contract, ask whether the investment you're making stays with you when you leave or disappears when you leave the platform. In my opinion, that's the real audit. And it's not just an invoice. Total. The next takeaway. Vendors must support composable architecture.

It has a near-zero survival rate past the six month implementation because most platforms market as warehouse native are actually warehouse connected. The data still moves, which means stop asking vendors if they're composable. Ask them to show you the generated SQL when their system builds an audience, and ask what happens to your data after it is sent?

If they can't answer either. Particularly specifically, you have an answer and takeaway number three. Only 23% of organizations running AI agents have them in full production. The other 77% are stuck in pilot or assist only mode. And Peter's own framing is that the gap closes slowly, because it depends on understanding what agents are actually good at for your specific business, not a universal timeline across all businesses.

Which also means if your team is waiting for agents to just get better on their own, you're waiting for the wrong thing. You're in the wrong line altogether. The unlock is building the harness, the guardrails, and the order of operations, not the model. Thanks for tuning in to the making sense of martech.

Try not to think about your map as a leaky faucet all week, or a kitchen sink for that matter. See you next time. A special thank you to Christine Murtaugh, who edited this episode. In an extra special thank you to Jenna Carter for believing in this passion project meets business. Stay curious.