The Warehouse Won, Now Someone Else Gets to Monetize It | Luke Ambrosetti

00:00:01 — 00:02:26

Welcome to the breakdown, where we delve further down the martech rabbit hole and try to make sense of it. Today's guest has spent a decade telling Enterprise Brands the data warehouse should be the center of their martech universe. So the question is, isn't whether he believes in warehouse native architecture, it's whether the industry is quietly becoming this all in one thing.

He told everyone to stop buying in the past. So first and foremost, a little bit about my guest. Luke Ambrosetti is a principal industry architect for marketing and advertising at snowflake and one of, in my opinion, the most connected practitioners in our entire martech data ecosystem. Prior to snowflake, he ran Solutions Engineering at Message Gears, a platform built on that same premise that brands should stop copying data out of their warehouse into a separate ESP.

The operationalized it for household name brands. You ever meet him in person? There's some crazy brand names. Well, the rest of the industry still treated the warehouse at the back end. Data only concern for someone whose career is built on making data pipelines invisible, he's unusually candid in public.

He's got his LinkedIn posts that always go viral. He's got medium posts, and they also read a lot less like thought leadership only, and more like someone arguing in real time about the complexity of our industry. Because that's the point of this podcast. Nothing makes sense. It's all contradictory.

It's all over the place. So on that note, Luke, welcome. Thank you for being here. Absolutely. Thank you for having me, Jacqueline. Appreciate it. These are my thoughts. My opinions. Definitely not my employers. On behalf of my own personal self today as Luke, not here as an employee of a company. Understood.

Now that we got that out of the way, let's dive in. And we're going to start with some rapid fire to warm us up. What was your first martech tool? My first martech tool is I've dirty secret about me. I've always been on the platform side. I've always been working at vendors. My first job was not necessarily martech, but it was comms tech or communications.

That was technically my first tool because I helped our customers, you know, use it and deploy it. And it was just a communications platform called Cedar. I always call it boring documents, you know, as, um, bank statements, you know, insurance documents. That's what I did before I got to the actual marketing communication.

Understood. Okay. We're going to play word association. I considered ad mad Libs, but we're going to go with this instead. So one word for each company about tonight. Salesforce SaaS.

00:02:28 — 00:02:34

That I got. Yeah. Databricks. So hard to do. Just one word I know.

00:02:37 — 00:02:43

Competitor phrase. Leader. Mm. High touch. Composable. Fair.

00:02:45 — 00:03:04

Iterable. Interesting. Helium pipes. Customer. Io. Curious. Yeah. Growth loop. Compound. They'd be very happy about that. Message gears. Fond. Hmm. Treasure. Data. wonder. Interesting. And last but not least. Snowflake blue.

00:03:06 — 00:09:28

Very neutral. Yeah. Because you have a background of working with pretty much every major brand. What do you see as the most unloved stack that enterprises are still running? And what is the short answer of why they won't kill it? The most unloved stack I would say something that that uses some sort of what I would call like the the legacy reverse ETL.

It's something that if you've read anything that I've written in the past, I don't know, just a few years, even worse details out of new concepts by any means. And so I think about like the legacy reverse ETL, like the if you see brands use unica or like the older versions of Adobe Campaign campaign or Adobe Campaign Classic as, as it's called, there's sometimes those are just very embedded into the workflow that some teams have, and it might be because they haven't really fully migrated to the cloud either.

I see that still. Even that snowflake. I thought I would not see that anymore past 2019, 2020. And yeah, I still still see it a lot. I can only imagine and how painful that is to witness. Yeah, if you are starting from scratch tomorrow with no legacy, no budget constraints, no political baggage, what is the first martech tool you would buy?

I would resist buy anything for as long as possible. It would have to be conversation with the business and with the business priorities. Ah, I mean, you can go so many different directions because there isn't a right answer. I think sometimes people put too much emphasis or too much priority on making sure that they have the right tools when it really is.

It's got to be a business conversation first, I agree. Okay, what is the most overrated buzzword right now? Composable agent tech or real time personalization? Those three. It's a genetic, but my favorite buzzword right now that's not in those three is context. Mhm. Yeah that's fair I think adding on that, it's like also semantic layer.

Semantic layer. Yeah. I've also said I think people last week at Cannes got, you know, tired of me saying serving layer. That's also, you know, semantic layer, serving layer. Interesting activation layer. Yeah. I have contacts, I think they're all all buzzy right now and I think for some good and meaningful reasons.

But otherwise, you know, very much an AI buzzword as well. Yeah. Okay. So for a little bit of context for listeners, I think Luke Ambrosetti I actually met in email geeks a long time ago, and also specifically like duking it out about the value of reverse ETL and KPIs, and that B2B businesses could also benefit from it.

When we finally met, not a slack component, I think we both finally were like, we agreed to disagree, but also we're trending in the right direction. It was funny back then. I know we disagreed, but things have changed so much now that I do agree. Right, so you could say you were ahead of the curve there. Jacqueline.

Whereas I, you know, honestly don't think I don't have experience to. To see it at the time we weren't working with B2B minded brands typically. Not at all, not at all. And if anything, I think B2B can benefit more. Only because it's always been neglected. Interesting, no. Or easier for. B2B marketing team to actually go in and use.

I mean it's one of the things. I actually I get to see up front and personal every day is, you know. The company that I work for, our marketing team to go and use it, and they do a wonderful job of doing that. But yeah, it's definitely the bar has been lowered and a very significant way, I would say in the past 18 months or so specifically, I would say barrier to entry, not the bar.

Yes, but that's what I mean. That's what I mean. Sorry. You're exactly right. I'm probably using the wrong term here. But yeah, the barrier to entry has been lowered. You think about any of these more technical platforms, it's a different persona, right? It's your you it's a different sales pitch. It is a truly different thing.

And there are certain perceptions along the way, you know, uncertain perceptions there, but I think everyone has to ask themselves. As they move up the stack here, or even down the stack in the other direction as well, right? Is is this someone actually really prepared to sell to. That's a very fair question.

I think no matter who you are, that's the question I've been posing. I mean, as Databricks announces their own CDP, the next frontier is actually not them building it. It will be great because it's built off of their own warehouse, but it's actually how do you get in front of marketers as opposed to just the data engineers and the data team?

Historically, particularly, I just came in front of it, but make it a really good experience for them as well. Exactly. Make that barrier to entry less difficult and less scary. Yeah, that is, I think, the singular hardest part of all of these companies kind of running to the middle. And we will talk about it more as we delve in.

And so I want to talk through we've talked about one tension, but I want to talk about another one that the our industry has spent over the past five years really consolidating about this concept of warehouse centricity. And now literally every vendor, whether it's Databricks, Salesforce, the CDPs, they're all racing to the warehouse and or trying to make it irrelevant in the confusing way.

And you sit at the center of all of this, having been at messengers previously and now at snowflake, and the composable CDP movement was supposed to free brands from vendor lock in. However, if the warehouse has become the activation layer, haven't we just moved this lock in up a level from the SP to snowflake?

What is your perspective on this? So I'll actually take this as two different questions right as one. First, why are they all racing here. And that's something I was just talking about a couple of weeks ago as Databricks made this announcement, is everybody wants to be the serving layer, right? AI needs a really good governance serving layer foundation to to operate.

It's not a sales pitch. It's the reality. Otherwise you get crap, crap and crap out. Correct. And so that's why everyone's racing towards that is because they know that if you go in and build this intelligence Layer for an agent to to operate from. That's that's what wins the ball game. I think at the end of the day, does it get you lock in?

I want to answer that. I want to have to probably pass on it. I hope that we can maybe come back and talk about another day, whereas others I don't think will, but that's still TBD on on where that ends up. Yeah, it's always interesting how every cycle ultimately ends up into this sweet category, and it's just a matter of how it plays out.

And yes, we will have this conversation when the time comes.

00:09:29 — 00:14:42

There is a running joke that warehouse native mostly just benefits the warehouse vendors. What's your rebuttal? Not the sales pitch talking point. I would say what you're looking for here is you're looking for benefits to two main players looking for a benefit to the brand. Right. And what the brand needs to do and go and operate their marketing business.

And you're lucky looking for a benefit for the the marketing platforms or the martech SaaS out there. Right. For the brands. There is a level of like governance. It's been kind of always positioned as security. I don't think security is the right way to think about it. It's more about governance is making sure you know who has access to my data, when, how, etc., right.

So that the governance I think is extremely important, especially in the regulated spaces. That's one major benefit to to brands is getting the ability to govern your own data, getting to own your own schema right, getting to own and how what a customer looks like to you and not having to kind of force fit that into whatever kind of, you know, platform, you know, you decide to use.

So if you look at how even like the difference between what ad platforms do versus what martech platforms and how a customer, you know, what you need to find as a customer or even a conversion event, looks like it could be pretty different. And so being able to have your own of what that is, I think is very powerful because it makes makes it easy to actually swap and actually to build out different integrations or different pathways to different tools, because if you own that, you're not relying on some other party to tell you or to kind of force that for you.

So that's mean is probably the bigger thing for brands is being able to own that, that and you'll probably hear like someone say, oh, well, your brand can do that. Any, any platform even like the bigger marketing clouds. And I would say kind of, but I would push back because there are you look and again, I'm not saying these are wrong things.

You look at what some of the big marketing clouds do with kind of defining their own schema or defining their own way for marketers to think about the world, I think can be very powerful. But I'm also wondering, is that the world that marketers really want to live in at the end of the day, too? Those are two for brands, I would say the the big one for the martech platforms.

This is definitely a sales pitch. I'm just going to qualify that right now is reducing Cogs, right? Cost of goods sold. If you think about what martech vendors are doing is or have done, they have basically taken compute and storage from either the cloud providers themselves or like a Snowflake or Databricks, and they have kind of abstracted those and mark those up and kind of resold that to brands.

And it could be in the form of a SaaS fee. It could it could be in the form of, you know, like a seat based fee, or it could be in the form of compute and storage. But that's what they've done in the past. And it's like, hey, is that really is that what your highest margins are as a SaaS company? If it is. Is that actually a problem?

Actually, that's why I would go to the brands is like, hey, if that's what they're telling you, like you should be doing, that's where they're making a lot of margin from. And that's probably should be owning more of that because they're just gouging you on those places. But if it's if it's really, hey, that's not what your, your margins are, then you should go and focus on more, you know, higher margin opportunities than the the compute and storage layer.

Like you could say the same thing about stuff like your Databricks. At the end of the day, we're just, you know, taking EC2 and S3 and putting margin on top of those. But, um, that's my pitch to to vendors out there is like, look, you can actually really reduce your Cogs, offload that cost to somewhere else and focus on the functionality and the what makes your platform differentiated.

Past then reselling, compute and storage. Agreed. I mean, the first time I experienced a tool like segment nearly a decade ago, I could not compute it because like, why would I repay for this? Why would I repay for my own data that we already have or we plan to store somewhere? Why would I? I got maybe it's me being a cheapskate this whole time and where I'm just like, I don't want to spend money on things.

But yeah, something that I've always kind of hope for is you look back at what segment in those platforms first started. Cloud infrastructure was not easy really. Like that was part of part of the point was to make cloud infrastructure for customer. It was trying to make it accessible. And that barrier to entry for marketing to be more analytics and data focused, to be easier.

Correct. But it's it's almost overengineered solutions for a problem in my opinion. I think what happened is that the stuff like the bricks of the world kind of caught up too fast to that business model, really, and just kind of that's why we saw the big composable shift that makes sense. 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, Erica, 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.

00:14:44 — 00:29:47

Awesome. And now back to the hot seat. Okay, I want to dive into Under the Hood with you because you are a part of so many implementations, you've witnessed so many migrations and you've got a lot of opinions. You're a really great survey of one because you've witnessed so many things. And so what do you see as the single most common architectural mistake brands make going warehouse native?

And who sold them that? Bad advice? I'm going to cop out with this one. They were Soul warehouse data without a being warehouse native. I think anybody could guess. And this is one of the reasons why I wrote it up. I wrote about a checkbox features I went a little while ago. I could probably, gosh, I don't know, a year or two years ago.

And this is one of the main reasons that I, that I did it was because I was just starting to see this. We'll call composability or, you know, warehouse Native as a checkbox feature. It's like I you know, I get why. But also it's like, hey, look, if you're going to have a vision and a, you know, stick to it, right?

There are good reasons to say, hey, look, maybe warehouse isn't the best way to go. I'd love for folks to kind of stuck with that, but I think the allure and the pool was was too much just to kind of attach yourself to that narrative in some shape or form to check that box for, for brands. You keep asking about it.

I hear you on that. But I'm all about efficiency. But also just means you need to have the right players. And that's people and vendors. Yes, in the right room. And as you well know, AI keeps getting bolted on to every platform out in the world. How many brands, from your experience and perspective, are claiming AI driven personalization?

Have data that is clean enough to do that in performance? Depends on what you define as AI, right? If you include machine learning within AI, brands have been doing this for a decade easily. Plus it's not new, right? It's not new at all. And I think there are many, many, many. I would say that most brands, the vast majority of brands that I talk to, have already done something in that realm.

I think as you talk, as you think about, especially getting the gen AI in the creative space and doing that kind of personalization, we're adding one, maybe two, you know, there it's still very, very early. I've seen some great experimentation. I actually had, um, I was at a conference, a martech conference earlier this year and actually had a customer come in.

I was at the booth giving out demos, and I had a customer come in and do a reverse demo for me, which is amazing about what they're doing from a concept perspective. And it was truly, really cool to to see an intro. It's still very, very early in the inning to if I had to guess. Yeah, I mean, I would say it's really we're on first base of anyone still for most, and you're lucky if you're on third base of first inning, you see some brands experimenting and failing, right?

Like even just this past weekend, I saw some sort of he broke some World Cup records. Yeah, yeah. So that's something like along those lines. And it was obviously AI generated. It's like this looks like garbage. You know like who knows what brands are actually trying to experiment. I'd rather be at a brand who tries and fails and fails publicly than at a brand who just, you know, too conservative personally, right?

Um, so I at least kudos to the brands who are at least sticking their neck out a little bit and trying to experiment. That makes sense. This next question you might kibosh, but Snowflake's expanded aggressively into cortex native apps data clean rooms. At what point does neutral infrastructure become a competitor to the ISV partners it's supposed to support?

I mean, I'm going to pass on this one. I will just say. I will say, though, and please keep this like it's a fine line, right? It's a there's a very fine line that you have to walk. You know, platforms are on a platform. It's kind of part of what the gig is. I think if you're an ISV out there, I think it's you're the game that you play is and this is actually the game that snowflake has played for a very long time, right?

You look at AWS has been out there a long time, has built right. Redshift, of course, is much older than that much, but few years old than snowflake is. And snowflake managed to build a better data warehouse, especially early. Like thinking, you know, roughly eight eight years ago or so. Something has to build a much better data warehouse than AWS ever did, right?

If you stay far ahead, enough of the you know of that that more platform approach that's still a winning ball game to play. Not like the world's over when Databricks announces a CDP. I think there are plenty of companies who are still going to be doing fine. Well, and also there is the integral point of you don't just change data warehouse providers quickly, ever.

If anything, the goal is consolidating as much as possible because that's what's needed as opposed to an email. Migration is much more common. Is it still painful and difficult? And don't recommend ever? Yes, but it is much more manageable by comparison. Yeah, I'm thinking it all easier, I think on both fronts.

But yes, I agree with you. It does happen, but it's pretty rare. Usually it's like if both the Snowflake and Databricks are in a an account, they're they're managing different what we would call workloads in the business. Right. Again does that make sense. Data Databricks is there. Core has been you know data pipelines or engineering.

And an ML or snowflake has been more of that serving layer. Right. More of that analyst than BI tool being very SQL heavy, database being very spark heavy. And so and then they've gone the other direction, both trying to get to the other layer. So yeah that's where you would see both in is that, you know just depending on the different workloads.

So yes. Do you see migrations between platforms like that. Yes. And you still will. But I think both of those platforms know that there's a lot more business to be one of getting folks migrated to the cloud right from on prem systems. Then there is so trying to eat each other's business at this point, I don't know, I could be wrong, but that's Luke's personal take is that there's a lot more business there, you know, than there is trying to go in each other.

Well. And also it's different engineering and architectural philosophies as to which approach makes the most sense because they're both extremely valid. Another platform and really integral part of martech that we haven't talked about is the potential transformation with MCC. And I'm curious, what do you predict will change as a result of them, and also what are the limitations that come with them?

Because I got a lot of thoughts. Yeah, I think so. The limitations easier to talk about the limitation right now with there's a lot here. But MCP is effectively like in a not so nice way. It's you sprinkle some llms on APIs and boom, you're right, you have MCP, right? It's like chokes for a long time with colleagues that, you know, MCP is just the new API.

Uh, at the end of the day, of course, there's more nuance to it than that. I mean, the trouble with MCP will be the hard part with MCP is going to be managing the context window and making sure that whichever age like so if you're using MCP in the context of, I have a list of tools that's getting exposed to me, and then I have a client side agent that is in charge of using those tools and a and orchestrating them in a proper order.

The difficult part there is getting the orchestration correct and getting the context, you know, correct and correctly compacted enough, you know, for AI to to use it properly. That's going to be the the limiting factor here. And that's what every company is trying to build that serving layer to help and help our customers do that in an intelligent and meaningful way.

One thing I think about with MCP, though, is like this replace everything, right? And so I've actually thought about a lot about this in the terms of data movement. Right. You know, you look at MCP and you go, oh, actually I can get my, my data from my ad and my marketing platforms via MCP. I can pull those in or now I'm seeing as well that some MCP are exposing tools to push data into them.

And I'm just wondering myself at first I thought, oh, that's an interesting idea. We can actually now switch to MCP to do this. So you don't need specialized tools anymore. I'm like, but does that make sense at scale? Does that make sense for really what it's supposed to be, the more deterministic workflow and how?

The answer that I come to recently is that, no. As much as I wanted it to be maybe a few weeks ago to be the answer for everything, I think for some things you're going to have still a deterministic workflow needs to be done. And that's still of course, natural language can help with orchestration and solving some of that.

But, um, yeah, I don't see, you know, us pulling or pushing data, you know, via MCP and in mass very soon. I think it may be smaller, smaller use cases maybe, but I don't know. Who knows, maybe someone will come out with an additional protocol for it on top of MCP. Yeah, I'm hopeful there is standardization that benefits everyone.

That is thoughtful because there's been a number. Tell me more about that. Would standardization look like in a perfect world for Jacqueline? Yeah. At this point, you will. We will have an episode with me and Ezra going through this. But guardrails governance just as every IT team oversees. So. And snowflake you know what every other instance with data governance you have the same level of privacy and protections with that MCP, for example, with APIs.

Once again, not an engineer. However, I know you set them up to do very specific things. Right now, it's a bit of a free for all where you're like, yeah, you can have access to all of my financial data. And that is not. Most wise, some people do it. I'm not judging. I'm just very risk averse. And there was a study recently where they reviewed roughly about 7000 different MCP servers, and nearly half had zero privacy, security.

You name it. And that is concerning when this is the exponential growth. It's explosive growth, and I believe in the future of it. However, I think it was brought to market too quickly, too soon just to get ahead. Well, it's it's interesting. I think it's going to lead to whereas APIs led us to data silos, I do think MCP could lead us to context silos.

Right. And that's why I'm hoping that people will be a bit more restrictive about where they hook up MCP to. Right. Like so Canva MCP really cool MCP server that Canva has. I hope that people are not hooking up the Canva MCP server to every single other martech tool that they have, because that seemed like a nightmare to your point, right?

Definitely does with the convergence of CDPs, ISPs, SEPs and warehouses all kind of sprinting towards the middle where it's all collapsing categories. And now that Databricks has really planted their own flag, and it was definitely like the the most known, worst kept secret in the industry. If a tool is built for data scientists, is that actually going to benefit marketers?

Is it coming for marketers? Does that mean, like the warehouse native is now a ceiling that needs to be broken? I would say overall, I don't know. I'm eagerly looking forward to how brands navigate, you know, I hope to help guide that personally. But yeah, it's I'm eager to see how brands to navigate this new world.

I think it's there's more changes to come. It's going to be messy confusing for sure, which is unfortunate, but it's part of the world we live in, so we're trying to make sense of it. So if every tool now becomes every other tool, what's the actual moat? Is it? It can't be data. It can't be the ecosystem. Like what is it?

Most enterprise companies are still, as you've mentioned, running legacy tooling, and they're kind of too invested to rip it out sometimes. Who then is actually winning the innovative players or the incumbents just waiting out of this hype cycle? Oh, I mean, the stock market, I think is very much showing who the winners are.

But is the stock market a good predictor of everything? I think no, not not not always, but I I've had my own personal beliefs here. And that's really just to answer the question directly. It is the innovators. As someone who's been I feel like shouting to avoid for a long time, I feel like the voice finally answered back and said, okay, yes, we're starting to pick this up a bit more.

Well then to push on that, the void is also answering a lot as. And this might just be a black hole, and it might actually be the destruction of the creation of things for how much is change and how fast. It's hard believed to believe that this is all just one giant bubble. I think there are. I think there are aspects of it, of course, that are.

Especially if you look at, you know, all these these companies are just kind of doing big deals with each other. Right. And just like, oh, we're going to pay you billions of dollars, you're going to pay us. Like a bit of a circle jerk. It is. And so, yeah, I try not to think about it too much, because I do see real results with brands like, you know, that's not me putting a sales pitch on it.

It is something that it's fun to see a click and do people get value out of this? I hope that it doesn't just kind of escape away from us and that, you know, it gets it gets too far from us too fast. But yeah, it's right now it feels maintainable and it feels like it's going to be fine. I'm not. There's parts where I've been a little scared.

I would say it in the past, especially with how good AI coding has gotten. Yeah. And so but I think people are adapting. I think things are adapting properly for a while. Didn't we look like it was going to. But I'm less concerned these days than I was maybe 3 or 4 months ago. Well, I'll have to check in another 3 or 4 months and get you on the record for that.

Okay. Luke Ambrosetti city. So last but not least, who is someone we should have on the podcast Aaron Fox with you for those who don't know her. She is amazing. She definitely have her on. I've learned so much from her, especially from, you know, from more of a martech, more of a owned media background, email, push SMS.

Right. I've learned so much from her. She comes from the media and ad side, you know, especially paid media. She's awesome. Okay, well, thank you so much. Where can folks find you and follow along with your greater questions than answers? LinkedIn X, as it's called these days. That's where I spend most of my time on socials for takeaways from that conversation.

First and foremost. Warehouse native has become a checkbox. Vendors have started attaching that label without actually committing to the architecture, which means the thesis Luke Ambrosetti spent his entire career building has been diluted by the very industry that has adopted it. Also, Lorcan didn't disappear.

It's only moved up a level and while Luke Ambrosetti fully rebut it on the record, which is probably the most honest answer in the episode. The real race is the serving layer, not the warehouse. Whoever owns the governed. AI ready data layer wins this next decade of martech, which truly reframes the entire conversion story from everyone's becoming a CDP to everyone's trying to be the foundation AI operates from.

And that's a big challenge. Also, schema ownership is the underrated benefit nobody is talking about. Not the security, not the cost savings, but owning your own definition of what a customer looks like. So you're not force fit into a vendor's data model. And with that, thank you for tuning in to Making Sense of Martech.

See you next time. A special thank you to Christine Murtaugh, who edited this episode into Extra Special. Thank you to Jenna Carter for believing in this passion project meets business. Stay curious.

— End of transcript —