225: The Fall of CRM gravity (The Dungeon of martech architecture, part 1)

[00:00:00] Phil: Okay, be honest. When was the last time you pulled up a number in your CRM and you actually trusted it? [00:00:06] Istvan: I consider it this like a CRM tools, like a glorified excel sheet. [00:00:09] Kevin: Passing all these events into the systems that you would use as a marketer, but then like there is, uh, entropy in those end systems. [00:00:15] there's just like so much data and like AI coming into the equation and screwing things up, that I actually think it might just all end up going into like a data warehouse [00:00:23] Sarah: I think the data warehouse has to be the center of information, and all information has to eventually be shared. Get into the warehouse [00:00:33] Lou: The applications are coming to the data and removing copies. [00:00:36] Hope: Everything is going to BigQuery. That's our source of truth. And then we're sending it out to all the different platforms. [00:00:41] I don't wanna be building the same audience five times, right? [00:00:44] I want my audiences in BigQuery. I wanna create them once, and then I wanna be able to send them wherever I need them. [00:00:51] ​ [00:00:57] In This Episode --- [00:00:57] Phil: What's up, folks? Welcome to our four-part [00:01:00] series of crawling through the dungeon of MarTech architecture. You've arrived at part one, the fall of CRM gravity. [00:01:07] ​ [00:01:08] Phil: We'll explore the demotion of the CRM from the center of the MarTech universe to a specialized tool. We'll unpack why CRMs often become dumping grounds for counterfeit truth, why the solution is moving to a warehouse native architecture, and how to achieve portable audiences that belong to no single tool. [00:01:27] All that and a bunch more stuff after a quick word from two of our awesome partners. [00:01:31] Sponsor: MoEngage --- [00:01:31] ​ [00:02:28] Sponsor: Knak --- [00:02:28] ​ [00:03:34] I --- [00:03:34] Phil: Welcome to the descent into the dungeon of MarTech architecture, a four-part journey through the unhinged and constantly expanding world of marketing technology. [00:03:44] this is hopefully gonna be educational and helpful for anyone that's, like, building, working in MarTech. [00:03:49] Uh, hopefully a little bit of fun also, but without a doubt, it's definitely gonna be weird, [00:03:54] Let's begin our descent. [00:03:56] ​ [00:04:00] Phil: Okay, be honest. When was the last time you pulled up a number in your CRM and you actually trusted it? [00:04:06] Like, you didn't have to second guess it, You had total confidence in that one number when someone asked for it. [00:04:12] Or maybe you've actually been in that meeting where you have two people arguing over a number, a metric, and they both pull up the same CRM, they're both logging into their accounts, and somehow they get two completely different answers for what should be the same question, and no one can really explain which number is actually right and why. [00:04:32] you don't have a source of truth if you're dealing with situations like this. [00:04:36] You have a CRM that's turned into a dumping ground of lost updates that have slowly compounded into competing versions of reality. [00:04:44] And the whole point of this isn't to pick on the CRM, um, but this is where our first boss, gaming boss, is hiding [00:04:51] FLOOR 1: Why the CRM Lost Its Authority --- [00:04:51] Phil: Entering floor one, why the CRM lost its authority. [00:05:00] If you're in B2B or B2C, the first floor of our descent looks a bit different, but only because of terminology. [00:05:07] I don't really get this, but in B2B, the two cornerstone platforms are the CRM and the MAP. Makes a lot of sense, the customer relationship management software and the marketing automation platform. Sales works in the former, marketing works in the latter, and then ops and rev ops are kind of stuck making the two talk to each other. [00:05:25] In B2C though, I don't get it. Like for some reason, you all decided that the MAP is actually called a CRM and, and the B2B version of the CRM isn't really needed because there's often no sales team. Instead, it's kind of customer support or product led motion. [00:05:40] But in both scenarios, the same thing happens to that central platform. It gets inherited by teams that weren't its original source, the intended source, and it accumulates data that wasn't designed to hold it in the first place. And it becomes the unofficial source of truth for the whole business, and no one [00:06:00] explicitly decided that that was a good [00:06:03] ​ [00:06:09] Why Every Team Moved Into the CRM (And How It Lost Its Authority) --- [00:06:09] Phil: So how did we get here? CRMs were built for one job: tracking good sales motions. Contacts, deals, stages, activity logs. They were really good at that job, and they still are. Then marketing started moving in, and marketers ruin everything. I love marketers. I'm a marketer, but we, we love to ruin everything. But leadership is actually worse. Leadership started pulling board metrics from the CRM, and then we have the product team that started adding usage data. Then we added ABM and account signals, and we had push notification that had to live somewhere, and then AI interactions needed a home. [00:06:42] We're dealing with a mess in CRMs if, if that's still your source of truth today. [00:06:47] Everyone kind of needed a record of the customer, and the CRM was already there, and it's, like, literally called the customer relationship manager. Any customer relationship data needs to live in that platform. So it became this, like, shared folder that [00:07:00] everyone saved their customer work into, even though it was designed for this very specific kind of sales motion work, not for stuff for the whole company. [00:07:10] Um, the problem is that once data is stored in a CRM, it starts reflecting the team that works there. So sells, sales is editing the contacts, marketing is overwriting a field, customer success is adding notes. Each edit is a local logic, and it's applied to what everyone assumes is the shared source of truth. The data looks, you know, legit, but you know deep down that the authority behind it belongs to whoever was on that customer record last and edited it last. [00:07:40] Meg: I'm used to Salesforce being the single source of truth or HubSpot being the single source of truth. [00:07:45] Phil: That's Meg Gowell, head of marketing at Elly.ai and the former head of marketing at Typeform. She crossed over from a Salesforce-first organization to one where the warehouse had already taken over [00:07:58] Meg: And here it's like our [00:08:00] core business. Is more represented in the data warehouse than it is anywhere else. And then Salesforce more has the sales led part of the business. [00:08:06] And so understanding how those data pieces come together is something that like, to be honest, I'm still working on. I've been here for three and a half, four months, right? It's tricky. And I think that that's what we struggle with the most is. Figuring out how those pieces fit together and how we're not just putting a motion on top of another because they have to, like in PLG, they have to serve one another. [00:08:27] And yet if your tech stack doesn't support that, it makes it really, really hard. And so, you know, we run into challenges sometimes of like, okay, do we have all the data points in the right places for people to action on them? Like, do we know everything we need to know? And I think. I've really experienced here how important that underlying data structure is and consistency across tools. [00:08:49] And that's something like, you know, we have a great new VP of data here and he's doing a ton of foundational work where in the past we had a really wide spectrum of stuff where it was like, okay, we went really, really deep and we have [00:09:00] this very advanced multi touch attribution system. And yet, like when you go to this other side, it's like, Oh, that's just like very basic reporting. [00:09:06] And so there's just like a very weird mix of like, like, Super deep, super surface level, but like no underlying structure that fully works. And so I think that, and I think that happens to a lot of companies, right? When you're growing fast and you're like taking, you're taking opportunities where you see them and you're moving quickly. [00:09:20] And I think now we're kind of taking a step back and like, we really need to shore up the foundation because that's kind of what supports everything else. Um, so yeah, I don't even know if there are specific tools that I'm using at this point. I'm just really like at the end of the day, I feel like having that data foundation is just so valuable and being able to use it across your tech stack is like one of the most important things I think for anyone just to understand your users and to be able to capture different signals and action on them in a way that like actually will impact revenue. [00:09:48] Phil: So she was on the podcast when she was still at Typeform, and at Typeform, the CRM stopped being the brain for that startup at some point. Not really a startup anymore, but the data warehouse had already taken over that [00:10:00] role when Meg joined. And once you've seen that transition, it's kinda hard to unsee it [00:10:05] Istvan: CRM is an interesting one. I, I I see this, that some companies have this as like a RO truth. Yes. Uh, [00:10:13] Phil: That's Istvan Meszaros, founder and CEO of Mitzu.io. He built warehouse native analytics at scale, and he's arguing that the CRM was always the wrong anchor [00:10:24] Istvan: although I, I consider it this like a CRM tools, like a glorified excel sheet. You're very easy to make a mistake and you never find it out. And, uh, in the data warehouse, it's like a slightly harder to make a mistake. [00:10:38] And, uh, like you have to have like a, a job that updates the table and runs every day. And you somebody review that job. And if you find a mistake, that is then if you make a mistake that it's a mistake for a thousand, uh, records or like a million or like a billion records even, because it's much easier at the spot as you think. Um, yeah, [00:11:00] that's definitely, but it's, um, evangelizing this as a, as a solution. It's, it's hard. [00:11:05] Phil: So access and version control is actually one of the base issues with CRMs when it comes to letting other teams move in and, and consider it the source of truth. When you change a lead status or you bulk upload a list, or you're manually tweaking a lead life cycle stage, nobody really questions it. [00:11:22] Nobody remembers who even made that change. Nobody can see who's doing it. Warehouses demand structure on the other hand, like updates happen through scheduled jobs. They're written in code, version controlled. They're deployed with intention, and in this case, the friction of like accessing it and, and needing permission to do stuff is actually a feature in this case, and the whole thing is kept under lock and key by the data team, and yeah, it's, it's a feature [00:11:48] David: You're right. Marketers don't care how the sausage is made generally, right. And sometimes I think that comes at the expense of their empathy [00:11:55] Phil: That's David Joosten, co-founder at GrowthLoop and the co-author of First-Party Data [00:12:00] Activation. He's seen this from both sides as a platform builder and an enterprise architect, and he's identified two pain points that marketers actually care about [00:12:10] David: having said that, they do very much care about when things go wrong and like they sent out a campaign they didn't intend to target the way that they did. Um, and they certainly care about speed. So how do you solve those two marketing pain points is really what we're talking about [00:12:27] Phil: So I agree with David. Like, marketers may not care how the sausage is made, but they do care deeply when a campaign hits the wrong people, or it uses stale data, or it moves too slow to matter. And David expands on this, and he picks on Marketo a little bit here [00:12:44] David: when you're going and working with, I, I'm gonna pick on Marketo here just because I have a very specific, uh, example here in mind. Um, their APIs into and out of their system are very opaque. So, [00:13:00] um, kind of information you get back to really know like what's in Marketo right now without going record by record is actually quite challenging to do from a data person's perspective. [00:13:11] Um, on the other hand, that's very easy to do in a data warehouse. So the challenge is I have, uh, they, they call it the two generals problem and in computer science. Essentially I'm sending my message to say, Hey, I wanna update customer record number 1, 2, 3 with their most recent activity. How do I know that you got that message and that the messenger didn't die along the way? And so that, that problem still exists in the CRM systems in a way that's been solved in most data warehouses, right? I could look up customer 1, 2, 3, but I can't look up all customers at all times. Uh, in order to verify that those events made it in, they don't give you the sort of observability that you'd want. [00:13:52] ​ [00:13:57] Why Sharing CRM Data Always Breaks It --- [00:13:57] Phil: So when the answer to, "Can we share this with [00:14:00] finance or product or leadership?" is always, "Let me pull a report and send you a spreadsheet," the system of truth is for one team only with a UI that's often really ugly. Uh, here's David Joosten again explaining why the tool itself is actually part of the problem here [00:14:16] David: if you think about, certain tools like Marketo have been optimized more for like a B2B use case. Other tools, uh, in the marketing automation space like Braze or Klaviyo have been optimized for B2C or consumer marketing. So if you're operating in an environment where you have lots of different product areas, you have lots of different, you know, granularities of customer, you have agencies versus SMBs versus enterprise and you have a lot of complexity, it's very unlikely that your marketing automation system can just reflect the truth of your business. [00:14:50] Phil: So both David and Istvan are fantastic pioneers in this space. And yes, obviously, like their perspective comes with bias. Um, they're both [00:15:00] like building in this space and selling a product in this space. But it is the earned kind of bias in my opinion. Like they didn't wake up one day to build a tool and then reverse engineer philosophy around it. [00:15:12] They built a company and a product around this belief because they were already convinced that the warehouse should play a more central role in modern marketing systems. [00:15:23] Let's hear from someone who saw this from the other side of the stack [00:15:26] Kevin: at Segment we were like, okay, we are the source of truth for event data and we're gonna federate that data to all these different destinations. [00:15:33] Phil: That's Kevin White, now head of marketing at Scrunch.ai, and he spent a bunch of years at Segment during its early rise. He was close to the infrastructure story as it kind of unfolded. It was super new in the market still at the time, and now he's had enough distance from it to kinda reflect on what actually happens when data starts moving downstream [00:15:53] Kevin: Uh, and that's gonna be like, your brains are your, like, you know, I'm gonna pass things. Like, it's almost like you're, you're, um, ETL [00:16:00] for. Uh, for like top of funnel or something like that. And so it's just like passing all these events into the systems that you would use as a marketer, but then like there is, uh, entropy in those end systems. [00:16:10] And so like a reverse ETL would say like, oh, we're gonna like take this and turn all that data into like a consistent, um, up to date across all these different systems. And so I think, I think that that was like a, a interesting take, but then. I feel like now, because there's just like so much data and so much unstructured data and like AI coming into the equation and screwing things up, that I actually think it might just all end up going into like a data warehouse as like a data warehouse being the source of truth. [00:16:40] And then from there you can compose and do all these things on top of your data warehouse. Just because like a, a data warehouse, um, or a data lake, or whatever you wanna call it, Databricks or Snowflake, essentially is the only play, the only place I see right now that can handle all that data. And then there's like tools that are built on top of it to structure that data and make it usable and composable in [00:17:00] Salesforce or pull it into common room or, you know, put it into customer a or something like that. [00:17:05] So. Um, I kind of think that that's where the market's going is like if anything's gonna displace Salesforce, it might actually be a data warehouse, which is maybe a, I don't know, counterintuitive or, or, or different take, but I, I don't see anything disrupting the, the CDP market other than like a data warehouse right now. [00:17:23] Phil: I love that Kevin uses entropy in his argument here. Data that travels to downstream tools loses integrity over time, and the CRM compounds those mistakes instead of containing them. [00:17:36] The warehouse solves the entropy problem, in my opinion, in, in a couple of different ways and differently, but, like, every transformation is written, reviewed, deployed explicitly. There's no, "Someone must have changed this at some point." We have that version control. Every change has an author, a timestamp. There's auditability, um, and that is a different relationship between a team and its data, one where the system is [00:18:00] resisting drift instead of accumulating it. [00:18:02] Why CRM Gravity Outlasts the Technical Argument --- [00:18:02] ​ [00:18:07] Phil: So obviously we can't go deep into the dungeon of MarTech architecture without hearing from the legendary figure who built the tunnel by hand. But before we hear from Scott Brinker, I did wanna touch on a recent report that he worked on with Databricks. [00:18:23] Uh, someone quoted in the report, Bryce Peake, who is VP of digital products and technology at Domino's, says in the report, " We've been trying to modify the same MarTech stack we've had since the internet started interneting. Folks, we're going to have to build a new one." So that's a Fortune 500 brand with decades of infrastructure saying the patch has run out of runway. [00:18:47] Meg: The CRM is actually really hard to dethrone, and part of the reason is because teams have built their identity around it. Yeah. You know, I never thought I would miss Salesforce reporting things. You never expected to [00:19:00] say [00:19:00] Phil: Here's Meg Gowell describing her own experience when she crossed over into that warehouse first organization for the first time, and it kinda captures what's really happening for a lot of folks considering that transition [00:19:12] Meg: learning. Looker has been a steep learning curve. Let me tell you. Um, I think part of it is, you know, there's the learning the tool and then there's the learning the fields, right? And learning the underlying data structure. [00:19:24] And I think that those are two different things. And because I did both at the same time, it made the whole thing harder, right? Like if I had just gone into another company where it was Salesforce again, I would have still had a lot of learning to do of like, What's stage one? Does it mean a meeting has been booked? [00:19:37] Does it mean something totally different? Because everybody uses that differently, right? That's hard enough in and of itself. But doing both together, I think, has really been the challenge. Especially when, again, we have these two different motions in the business on the self serve side and the sales led side. [00:19:51] And marketing is responsible for supporting both. And I can't just look at one or the other. Like, I have to look at both. And so, for Understanding [00:20:00] what all the data field means, how they work together is something that I am still working on. Um, I think it's been overall, it's been cool to feel like there are more data points that I can pull than I've previously had access to in a CRM. [00:20:15] Probably. Um, I think there's that level of familiarity though, where like, I feel like it takes me longer to pull stuff still in Looker and so I feel like I'm not quite as. Quick, I guess, as I used to be. And so it makes me not look things up as much as maybe I would in the past. And then I think the other layer on that, that I still haven't figured out is in a CRM world, I would often drill down onto individual contacts and dig into like historical changes and stuff like that. [00:20:46] So that I could really get an idea of like, well, the data is telling me this, but like, is that extra actually true? Like, let me look at a couple of specific examples and that's a lot harder for me to do. From the data warehouse perspective, just because the data isn't quite structured the same [00:21:00] way. And so I think part of it is like figuring that out too, of like, how do I spot check what I'm doing? [00:21:05] Because I'm kind of approaching it from a different way than I, than I used to in a Salesforce or HubSpot world. [00:21:09] Phil: So for a lot of folks like Meg, the CRM feels like the center because that's where the people doing the work are hanging out and logging in. The sales reps work in Salesforce. Marketing managers go in and out of Salesforce and HubSpot. The lifecycle manager maybe lives in MoEngage or Iterable or something like that. The data engineers work in the warehouse, and so the warehouse seems like a weird fit to be the source of truth because none of the go-to-market folks are logging into that tool. Getting the marketing team to trust the warehouse means getting them to trust a system where their day-to-day doesn't happen at all, and that resistance obviously is pretty human. It's a normal reaction, and it's the real reason CRM gravity keeps persisting long past the point where it should have given way. [00:21:57] John: there will always be arguments of [00:22:00] what the source of truth should be [00:22:01] Phil: That's John Saunders, VP of product at Power Digital Marketing. He spent years building Nova, an internal operating system designed to give agencies a single source of truth, and he learned the hard way that the technology was actually the easier half of the problem [00:22:19] John: Um, I think it's more about is the source of truth real for us as an organization? Is everybody aligned that this is the source of truth? And so I actually see, see it as more of a people versus a product problem of if your people aren't aligned, that this is the source of truth being the platform, um, then you don't have, but if they are aligned, then it is a source of truth for your opinionated approach. [00:22:46] Phil: So Nova's early years focused on consolidating everything into one big platform. Salesforce records, client service metrics, spend data. In practice, it created a different problem for the company though. People didn't want the [00:23:00] data. They realized that they wanted a quick story of what changed and why, and the organizational agreement had to come before the data engineering could matter. Without it, the platform was just another system people had to work around to get the answers that they wanted. [00:23:16] Okay, so let's, let's recap what this floor kind of contains so far. We've got three separate failure modes all-- with the CRM all rooted in kind of the same thing. So number one, CRM was built for the sales motion and was never rebuilt for any one of the other teams that moved in. [00:23:34] Number two, number two, every field looks authoritative, but the authority belongs to whoever ran the last import or whoever edited that lead last. And number three, number three, the teams most anchored to the CRM are the ones most resistant to trusting the warehouse because the warehouse is not where their day starts. [00:23:55] They don't even have a login access to it. So counterfeit truth gets its [00:24:00] grip from the humans who never agreed to stop trusting it. [00:24:04] BOSS BATTLE: The False Truth King --- [00:24:04] ​ [00:24:07] Phil: B- b- b- boss battle. The False Truth King. [00:24:12] The boss on this floor that you need to defeat is the False Truth King. Data with the appearance of authority and none of the substance behind it. [00:24:22] The key to defeating this boss is understanding what the CRM was actually good at in the first place, managing the sales motion, and what it was never built to be, the center of marketing truth, let alone the center of organizational truth. So our end goal isn't to kill the CRM. I'm still a big fan of those platforms for those use cases, but we want to demote it. The end goal is letting the data warehouse take over as the brain or the source of truth. [00:24:49] Let's check our inventory. So before we face this boss, here's what we picked up or learnt on this floor. [00:24:56] There's one big glaring item that we [00:25:00] have in our inventory now, and it's called the data warehouse. Every change has an author, a job, and a timestamp. Updates happen through scheduled version-controlled code. Every change is traceable. Counterfeit truth can't survive a system that audits everything by default. [00:25:19] "Okay, cool, Phil. [00:25:20] Like, thanks for letting me know. Data warehouse should be the source of truth, or our CDP should be the source of truth. H- how do I actually get there?" It's, it's not as simple as just, like, buying Databricks or Snowflake and then just, like, getting one data engineer to do it. There's a whole political battle behind the scenes that needs to happen here. [00:25:38] So not so fast. MarTech Crawler, let's, let's go through how to actually defeat this boss. [00:25:44] Sponsor: GrowthLoop --- [00:25:44] ​ [00:26:48] Sponsor: GrowthBench --- [00:26:48] ​ [00:27:49] Phil: knowing how to defeat this boss is one thing, but actually getting that implemented is actually the bigger battle. Human decision-makers are involved here, and a defiant marketer proclaiming that the data warehouse [00:28:00] needs to exist needs way more than just that proclamation. [00:28:04] So the transition from the CRM to the data warehouse is harder to sell than it sounds. If you've tried to do this, you've felt this for sure. Uh, here's Istvan Meszaros again, who has watched teams make this shift, and he's explaining why it's an uphill battle [00:28:19] Istvan: from the point of view of the user, the end user, it is not really a new category. They are unable to do the same type of analytics they would do in the traditional third party analytics tools. But from point of like. The people or the team that builds these, like marketing, um, infrastructure, data infrastructure for them is a completely new approach. [00:28:43] It's a totally, it's a completely new category as well. I have to admit that it's, it's pretty hard to, uh, companies to switch to a warehouse approach. It takes a lot of time. Getting stuck in one is, is much easier, obviously. Uh, especially, you mentioned [00:29:00] CDPs and now they are out of the box. Doing the warehouse, uh, ingestion. [00:29:04] Uh, so if you have that in your organization immediately collecting a warehouse, uh, customer engagement software or like a, or like an analytics software like ours, it is essentially a couple of minutes only. Uh, so it's very easy to get started to get off of these, uh, third party tools. It's a, it's a different challenge. [00:29:23] uh, yeah, it's, uh, we see struggling in there, but uh, yeah, we see also more and more coming out. Of these older, let's say, let not, I won't call them older. It's like the traditional approach. Yeah. So just to summarize your question, if you think of the, the team that deals this system for them as a new category, if you think of the end user, it is enabling them to do the same thing. [00:29:48] So for them is, is not, I wouldn't see, even say it's a feature, it's completely transparent, should be completely transparent. [00:29:55] Phil: So obviously lots of politics at play here. The revolution happens at the [00:30:00] infrastructure layer in a language that doesn't translate upward to a lot of people that aren't super technical or, or live in those problems day-to-day. The CMO doesn't experience the change, the marketer doesn't see the whole architecture shift. The person who has to fight for it in ops or rev ops or whatever has to win a budget conversation about a thing the beneficiary will never notice in the first place. [00:30:23] It's a doable battle though, and the approach that works is the same as any political change. This isn't gonna become a political episode, don't worry. But the idea is to like find a team that's already in pain, deeply in pain, um, that doesn't wanna live in the CRM anymore, or, you know, maybe they're indoctrinated in there, but they're feeling a pain. There's a bunch of different pains we can cover. Let's say, you know, the analyst who can't answer the same question twice and get the same number, or, uh, the person in sales ops that's manually exporting CSVs every Tuesday. [00:30:59] You [00:31:00] know, win there first with some of those teams, then let that win or that plan to win do the actual selling. Here's David Joosten again explaining how he'd go about it [00:31:11] David: It really comes down to how passionate. The few people that you can rally with you behind it are, [00:31:19] what I mean by that is, um, you're, it's much more effective to have a few really passionate people than a lot of lukewarm supporters. [00:31:26] Phil: Mm-hmm. [00:31:27] David: Um, one of my learnings in working with large enterprise companies was specifically finding the pockets of the most underserved was oftentimes actually the best strategy. So yes, sometimes you get to go in through the front door, work with like the, the biggest geographical market, the biggest product area, uh, and then just like win there. [00:31:47] And that's great. Um, sometimes though you'll get much more passionate supporters by trying to work on the frontier where folks really aren't served. Those folks have no alternative oftentimes, [00:32:00] and so for you to be able to really meet their needs when you're building out these early, uh, 80 20 solutions can mean the world. [00:32:08] And so those people will be your fiercest supporters and advocates. And then the last piece is just proving it. So the quick wins is part of that technical proof of concept. Figuring out a really good way to get everyone in the room to figure out what we need to prove for everyone to feel comfortable with the decision. [00:32:25] Going through a two week process of like running through that with everyone on board and plugged in and engaged at the same time. People aren't on, on on vacation, they're not like at conferences. They're actually plugged in for those two weeks to support the initiative. And then you have a readout at the end to tell you whether you made it. [00:32:41] There's all these like tactics and strategies that help you get the rallying force, uh, almost the same way you build like a little political campaign, right? Uh, you gotta find like the deepest, most passionate supporters to help you convert the majority. [00:32:54] Phil: So you might not be able to win this boss fight with an abstract infrastructure argument during [00:33:00] your RevOps budget meeting, but you can defeat this boss with a before and after story from a team that stopped fighting about numbers. [00:33:09] This might actually be a months-long battle depending on the resources, the size of your team, the priorities, but it's doable. I've been there myself, especially in the age of AI. Like leadership, the execs, they care about this type of project more today because of AI if you can spin it as an AI readiness project. I've actually had success getting this rolled out as a dependency to all the AI projects the exec team wants to launch. We need to do this first. [00:33:37] The bad news is that this boss on this floor doesn't just end with us shining a light on the counterfeit truth element here. Um, you know, this boss that's wielding counterfeit truth at the source is only the first form of this boss. Here's a second version of this that you need to also consider. [00:33:56] Let's say you heroically convince the organization to [00:34:00] build the data warehouse finally. They didn't have one before. Um, a couple of data, uh, engineers are assigned to this work. Maybe you, you had some allies there on the team. The exec team is pumped about a source of truth. You go through an extensive implementation journey, and on the other side, you've got clean data, governed, centralized. [00:34:17] I know, I know. Easier said than done, but let's, for the sake of this, uh, dissent, imagine that we've achieved this. But then someone is asking the team to use this new fresh data in the warehouse to run a campaign. Absolutely. No sweat. Here's your audience. Where do you want this? [00:34:35] Where do you want this? The second form of this boss is a vault you spent years building where every exit leads to a photocopier. [00:34:46] That's no good. [00:34:47] ​ [00:34:56] Why Centralizing Data Only to Copy It Out Defeats the Purpose --- [00:34:56] Phil: Think about what that means. Like you invested the organizational capital, [00:35:00] engineering hours, months of stakeholder negotiations to build the central source of truth, and now you're making copies of it and distributing them every downstream tool in the stack where each copy immediately becomes drifting and is out of date and stale from the original? [00:35:16] It's like building a data treasury and then immediately photocopying everything inside it and mailing copies to different departments. [00:35:25] Obviously, this isn't the right way to approach it. [00:35:27] Lou: You go through this big organizational investment, almost cultural investment of unifying your entire data stack, right? [00:35:35] Phil: That's Lourenco Mello, director of product marketing at Snowflake and he's explaining this paradox really well [00:35:42] Lou: Just having a single source of truth, which, by the way, is a continuous and kind of work in progress. [00:35:47] Um, it's so counterintuitive to then say, Okay, we did it, right? We're where we want to be. Now, to get any sort of A. I. Modeling done or to have any sort of, you know, [00:36:00] campaign execution. Let's copy that data outward and execute it right. You just went through this whole investment. Why are we going back to that? [00:36:09] And so it's a it's a really kind of interesting conversation, but the applications are coming to the data and at least at the very least, removing friction and removing copies. Um, whatever the application model is that customers choose to use. [00:36:22] Phil: This isn't just a B2B thing. For the B2C folks listening, you could actually make the case that this is even more relevant and more important for the B2C orgs. What I hear from the ground is folks and B2C leaders saying some version of, "I've centralized a complete picture of my customer. [00:36:39] We've got purchase behavior, product engagement, support interactions, marketing touchpoints, blah, blah, blah, all living together in our data warehouse. I want to achieve and activate that warehouse data directly and turn it into real-time, highly relevant, deeply personalized experiences across my campaigns, and I can't afford stale data or the mess and friction [00:37:00] that comes from copy it across a bunch of different tools." [00:37:03] So it's a very similar diagnosis across markets and across org types. And, you know, the warehouse owns the source of truth. It holds the truth, but the tools need to come to the data as opposed to the data constantly sharing copies that become outdated to those tools. [00:37:22] Erin: if you've done the work to build your marketing foundation and snowflake. [00:37:26] Phil: That's Erin Foxworthy, Snowflake's Global Industry GTM Lead, marketers and advertisers, and she's describing what the shift looks like from the infrastructure side [00:37:35] Erin: The whole world of applications is now coming to that data, and the whole point is not only from a cost perspective, but a security perspective, right? [00:37:43] Like, you don't want that data to move. So how do we collaborate as an ecosystem right across our consumer sets, but move the data as little as possible? And that's exactly why I think you see the composability conversation coming. And again, You know, a platform like Snowflake probably wasn't, you know, years ago, first of all, it wasn't around. [00:37:59] [00:38:00] Second of all, you have to have the scalability performance to handle that vast amount of data. And the technology is there now, right? And so, like, that's why you start to see this amazing kind of conversation of, like, how do you now build applications and bring those applications like Martech and Adtech to the data. [00:38:14] Phil: So it's a pretty big flip from the old model, especially if you're not, you know, super familiar with the warehouse-native architecture. Instead of data pipelines pushing data to wherever it's needed, platforms are building native connectors that read directly from the warehouse. The mechanism that Erin's describing here is something we've covered in a lot of different episodes. [00:38:34] Um, Snowflake calls it data sharing. A lot of other folks call it zero-copy data. The ability to expose a live view of warehouse data to downstream tools without ever moving it. Um, there's a lot of benefits to this. It's not perfect, um, but you know, there's no export, no sync jobs, no copy that immediately starts drifting or becomes stale. The data stays where it is, and the application comes to it instead of vice versa.[00:39:00] [00:39:00] So the enforcement of this model is already happening at the buying level. One major holding company, Erin said in the episode that she joined me on, went to market with a firm position. They are actually only accepting data through a data share or zero-copy data going forward, and they were using their buying power to make it stick with the new provider that they're going to. [00:39:20] So when procurement starts enforcing the architecture, the technical argument has to become a commercial one [00:39:28] B-b-b-boss battle. The export Hydra. [00:39:41] The export hydra. The assumption so embedded in how marketing teams operate that it never registers as a choice. It's just an export reflex. Data has to travel to wherever it's needed. We've done it forever. You pull it, clean it, load it in the tool. That's just how it works. [00:39:59] But every [00:40:00] export is a copy. Every copy immediately becomes stale and starts drifting. Defeating this form of the boss on this floor means stopping copies entirely. So for this battle, by this stage of our descent in the crawl, we've achieved four items in our inventory. [00:40:19] Let's check our inventory. So we still have the data warehouse that we used to defeat our first version of the boss, and, you know, every change as an author, job timestamp. [00:40:28] We have nice clean data in there. Um, there's a couple of other things in the inventory that we've picked up that we're gonna unpack how to use it. We've got a data transformation tool. We have reverse ETL, and we have zero-copy data share. Let's unpack how we use those to tackle this, uh, export hydra boss. [00:40:48] ​ [00:40:53] How to Move to a Warehouse-Native Architecture --- [00:40:53] Danny: So the teams that have cleared this floor and beat the Export Hydra boss describe their victory in the same [00:41:00] way, let's give the most tangible roadmap of like, if I'm hearing this and I just want to get started. What does that actually look like for me? [00:41:08] Phil: That's Danny Lambert, the former director of marketing operations at dbt Labs. He's built this stack from scratch at multiple companies, and he gave us the clearest version of the blueprint [00:41:18] Danny: but the first thing you're going to need uh, is extraction, right? So typically this will be Stitch or Fivetran. [00:41:24] Um, and all this lets you do is just go in there, connect your integrations for whatever tool you want the data from. And it just syncs it to your cloud data warehouse. Um, then obviously you'll need a cloud data warehouse. So this could be Snowflake, Redshift, BigQuery, um, you know, Databricks, any one of these, um, if you go to those quick start guys that we talked about, all of them will allow you, I think BigQuery is probably the most flexible in this, will allow you to set it up with. [00:41:49] Low to no cost, right? You'll get a trial period with some consumption credits and then it's very inexpensive at the volume that you'd be doing. It's like, don't worry too much if you're just getting started. You just want to test this out [00:42:00] around cost. Like it's like, I don't know exactly what Fivetran or Stitch are right now, but not super expensive. [00:42:05] Uh, the warehouses have, you know, free to very low cost ways to get in. So set up the extraction. Port dump it into your data warehouse and essentially have just a raw table of data like in the rawest format possible That's where dbt comes in is it lets you do the extra transformation in the warehouse So it'll reference the raw tables that you have dropped in there and you can do that Join these together with basic SQL like that's one of the benefits of dbt is it's not like you're writing custom Python like stored procedure Scripts you're just writing SQL. [00:42:37] So if you have base level SQL, you're great If you don't maybe like brush up on a little bit of SQL, you probably want to know that know that anyway [00:42:44] Phil: Or ask ChatGPT for some help. [00:42:50] Danny: Um, and you can just do basic things, right? Like just select these four columns, clean the name just to get like familiar with, with what it is. [00:42:57] Um, and then once you have that, I think the biggest [00:43:00] last piece is, is the reverse ETL component, right? Census or Hightouch. That's what lets you take. This data that you've transformed into what you want it to do and put it where you want it to go, right? Like having it sit in the warehouse in isolation is not the end result of that, that someone would have to go in there and write the query on it. [00:43:16] It's usually, I'm going to feed it to a BI tool. I'm going to feed it to some operational platform, or I'm going to feed it into an AI model. That's like typically the avenues that you have afterwards. [00:43:26] Phil: So that's a pretty clean skeleton. And, you know, folks that are a bit more technical on the data side will recognize a lot of this as the modern data stack. And to be fair, this hasn't changed a whole ton, but you'd be shocked at how many companies in, from a marketing architecture standpoint haven't implemented this. [00:43:48] But the skeleton is basically three steps: centralize, transform, activate. But for teams that are already deep in the marketing automation and CRM, here's David Joosten again describing the [00:44:00] migration path that he's seen work [00:44:01] David: What they did is they essentially said the, uh, data and the fields that are found in my marketing automation system are sourced from a variety of places, uh, but they represent my source of truth that I trust. Let's rebuild based on that schema, that target output, my marketing data warehouse or data mart. Then I'll have a one-to-one match that I can ensure exists between what I think my of as my source of truth and what exists in this marketing data warehouse. That's how I build confidence with the team that's using it in marketing automation with the data team. [00:44:41] That's the bridge. As soon as I've done that and can start to actually feed the marketing automation system from this single source of truth, instead of all of those sources one by one, I usually increase reliability and observability, which means when something doesn't look right, something [00:45:00] breaks, there's an alert system in place. [00:45:02] It's really built with engineering best practices in mind inside the data warehouse. I can check for duplicates, I can look for, uh, missing data timeframes 'cause there's certain dates that are not populated. All of those things that build confidence in the data. So I'm already starting to see some advantages by, you know, being a one-to-one match, swapping out the sources from, you know, all of the random ones that I've got coming in to this one source that's collecting from those. [00:45:30] Then I can start to think about in my roadmap, where am I gonna go from here? I have an extensible place that I can now add fields as I need. Um, I can start to create calculated fields and aggregations, build my reporting off the same thing that I'm gonna market with. You get all those benefits, uh, as you continue down that roadmap, but you've never lost the thread and forced people to say, yes, I know you have your beloved, trusted source of truth, but I'm gonna overtake [00:46:00] it and I'm gonna win. [00:46:00] It's not about winning. It's about finding like, how do we all walk the path together to like an extensible and brighter future? [00:46:08] Phil: So there you go. That's your skeleton, your recipe to defeat the export hydra boss. You have the tools, the mechanisms to do it. Um, the political battle is a different beast there, but, um, hopefully this was helpful to you if you're living in the CRM world as a source of truth. Hopefully, um, I've convinced you to at least consider alternatives here. [00:46:30] ​ [00:46:35] Phil: And [00:46:36] How to Achieve Portable Audiences --- [00:46:36] Phil: somewhere along this floor, we encounter a few operating realities that are starting to become non-negotiable. Define audiences once , not separately in every tool. Reuse the same segmentation logic across every channel, so the definition of active users is the same in your ad platform as it is in your customer engagement tools. [00:46:57] And minimize the number of handoffs where [00:47:00] data can silently drift. For all the complexities of the long B2B sales journey and account roll-ups and all that crap, I wanted to, like, give a shout-out to the B2C folks here who deal with a challenge on this floor that most B2B marketers only dream of, and that's volume. [00:47:17] One audience definition on every channel at the scale of 30 million user records is a whole different ballgame than your B2B 300K lead database. [00:47:27] Hope: everything is going to BigQuery. That's our source of truth. And then we're sending it out to all the different platforms. [00:47:32] Phil: that's Hope Barrett, the senior director of product management for MarTech at SoundCloud, [00:47:37] Hope: So we're not using what the native. Integrations for SDKs, right? So we, we have the minimums, like I need to have an SDK in my app to get them all engaged user id. But all my events are coming from BigQuery [00:47:51] if I put my Performance marketer hat on, we're actually moving towards a way where we are building audiences, BigQuery itself, and then [00:48:00] using Hightouch to send it to my various endpoints. So like, we're gonna, we're gonna, we're already sending audiences from BigQuery to DV360, Meta, TikTok. So my goal is I wanna, I don't wanna be building the same audience five times, right? [00:48:15] I want my audiences in BigQuery. I wanna create them once, and then I wanna be able to send them wherever I need them. [00:48:21] Phil: So really cool to hear from an awesome company that's got one place, one audience definition, and their reverse ETL sends it everywhere. The native integrations and vendor SDKs get bypassed entirely because everyone is a potential drift point. Here's Blair Bendel, the senior VP of marketing at Foxwoods. [00:48:41] He's inherited the opposite situation with data fragmented across disconnected systems by industry and business unit, and there's no common layer underneath it [00:48:51] Blair: we have an IT team with all kinds of different systems in house. So whether it's our gaming system that captures all our gaming data, our hotel system. [00:49:00] Our food and beverage system. I mean, we have so, so much data that we can aggregate, um, that we needed a partner that could, um, uh, one, ingest all that, um, and, and, and put it into a form to where we can make sense from our, from what our team can execute on, right? [00:49:20] Because if, if we have all this data, um, and we can't execute on it, then, then it's gonna be, uh, certainly counterproductive to, to state the obvious. [00:49:27] So. [00:49:27] Phil: So the vocabulary is different across those two stories. Hope is building audiences in BigQuery and deploying them through Reverse ETL into MoEngage, their customer engagement platform, and Blair is consolidating gaming, hotel, food, and beverage data into a single execution layer. The infrastructure they're both describing is the same though. [00:49:46] Like, we've got one definition of the customer governed in a single place, available everywhere the business needs it, and the outcome is essentially the same, the ability to execute across channels without rebuilding the customer [00:50:00] profile from scratch every time, and no definition drifting, no having to worry about REST API sync time lags. [00:50:08] The thread that's connecting both of those stories is omnichannel execution as a data architecture strategy. The channels matter less than the foundation beneath them, and if that's fragmented or not. Here's David Joosten one last time. Uh, he's almost like our, our Sherpa or our game guide throughout this, uh, first floor on the dungeon here. [00:50:29] Um, but he talks about what that architecture gives marketers [00:50:32] David: a self-serve capability, that cuts out even just the, um, design step of targeting where I can get feedback on, Hey, what if I filter for this or wanna personalize on that? [00:50:45] And then get, you know, numbers back of here's how big the audience is, here's how much that audience spends today, other metrics that I might care about, even before someone on the data team looks at it to actually operationally activate it. That already starts [00:51:00] to cut down on the time it takes to go from ideation to fulfillment in a big way. [00:51:06] And marketers feel much more like fulfilled satiated with their work when they can go from idea to activation much faster. Right? There's no doubt psychologically, like they love that. [00:51:16] Phil: It feels like for years this architecture was just justified on reporting and efficiency grounds, one source of truth, fewer sync jobs, cleaner pipelines. Those arguments were always true, but in our revenue-hungry prioritization sessions, they never won. They were never prioritized. Data warehouse investments got funded by data teams and engineering. [00:51:37] But in 2026, the conversation is different, and we'll talk about this as we go into our second floor here next week. Um, but you could go as far as saying that if you want to do this properly, this, this AI stuff properly, the warehouse-first architecture is almost a prerequisite for AI. Every agent that runs on top of your stack reads from something. [00:51:59] [00:52:00] If what it reads from is a patchwork of copies that are drifted and downstream exports, the agents or your AI is inheriting all of that drift, all of that latency, all of that contradictory logic. If what it reads from is governed, centralized foundation with a consistent definitions and controlled access, blah, blah, blah, the agent can actually do something trustworthy with it. [00:52:23] It's the comeback of data quality, the argument that never moved marketing budgets finally has a forcing function, um, to the rejoice and happiness of ops folks. The funniest part is that, you know, the architecture and the whole modern data stack, like I said, actually hasn't really evolved that much from a foundation standpoint. [00:52:42] What's changed is the tooling and AI that's on top of it. that's your forcing function [00:52:46] New achievements, audiences that belong to no single tool. [00:52:56] That's the biggest achievement of clearing this floor and, and beating both [00:53:00] forms of this boss, that audiences stop belonging to tools. Audiences can be defined once, governed in one place, and deployed anywhere without rebuilding them from scratch or redefining them across segments on other tools. [00:53:13] Uh, we're almost done here, but I wanted to close out by letting Sarah Krasnik Bedell, the founding growth marketer at Railway, former director of growth at Prefect. She's crossed over from data engineering into marketing, and she's seen both sides. So if you're still on the fence about this is the right way to go about it or not and how we do it, um, listen to Sarah [00:53:32] Sarah: I think the data warehouse has to be the center of information, and all information has to eventually be shared. Get into the warehouse and the reason for that is because there's so much data in so many different places that I think it would be close to impossible to actually get a full view without having that data warehouse at the center. [00:53:56] Um, and even if it were possible, it would be really [00:54:00] hard from a, you know, Martech standpoint to make sure that all of the data successfully flows from one place to the other and vice versa without the data warehouse kind of being that central hub. [00:54:12] And so So that's kind of my, uh, my viewpoint and really the tools that are involved or it's, I think it's, it's a two way street, right? [00:54:23] So whether it's product usage or it's Um, sales, behavior, touch points and then surfacing it back into, um, either sometimes those tools to be able to kind of create that, um, flywheel of being able to learn and create segments in something like common room or more simply, it's then right forwarded to something like a, right, some sort of CRM where there's a, a one stop shop for right. [00:54:55] I'm in right b2b world right now. So one stop shop for sales reps or, [00:55:00] um, a marketing team or a dev, uh, dev rel team or something like that to really understand, okay, what is the kind of full view of who this person is and what they've done? Um, The, there's kind of a lot of pieces in that. Um, there's two different approaches in, in how to get data from one place to the other. [00:55:21] The first is, um, just kind of dumping everything into a data warehouse and using something like dbt to transform it, to store it, right? and then using, right, something like Census, for example, to write it out into your email marketing platform, into a CRM, et cetera. [00:55:37] But there's also a second way, which is relying more on webhooks. So, for example, right, someone does something, it triggers a webhook into another system, and that's a little bit more, I hate this word because I think it's really loaded, but it's a little more real time, right? If you want to, um, have, not rely on a batch process that runs, Um, every few [00:56:00] hours, but if you really want that, like, okay, one thing happened, now someone gets notified, um, that's a little bit more of a, of a straight shot to, to that world. [00:56:08] So, that's kind of the two different ways I think about it. [00:56:11] Phil: Such a cool perspective from Sarah coming from a data eng background, uh, leading marketing and growth now. And it feels like this new achievement is an operating model as much as an architecture, right? Like audiences defined in one place, governed consistently, available everywhere. Marketing, sales, customer success, all GTM teams working from the same shared definition of the customer. [00:56:31] So we're starting to make our way down to the second floor. We're going down in the dungeon of MarTech architecture, and we start getting a preview of what the next battle is ahead for us because our jobs are far from over [00:56:46] ​ [00:56:52] How CLI/MCP Servers Are Changing Marketing Stack Integration --- [00:56:52] Phil: The CLI and the MCP servers are becoming the new integration language between AI tools and marketing [00:57:00] platforms. Activation has expanded beyond batch jobs syncing data to downstream tools on a schedule. In 2026, we're talking about a continuous conversational process where AI tools read from and write back to the stack in real time. [00:57:15] The warehouse stops being the storage layer that you've pushed all of your data into, and it becomes infrastructure that agents can act on. [00:57:24] Scott: part of the biggest problem people have had in MarTech, you know, from the beginning, from that very first like landscape of like 150 technologies, [00:57:32] they're [00:57:32] Phil: That's the legendary Scott Brinker, who shows up as our game guide in the next few episodes [00:57:39] Scott: well, it's great you got all these things, but do they actually work together well? Okay. No. You know, and now that we have like 15,000 of 'em, you know, do they all work together? [00:57:47] Well, not so much. So integration has always been the challenge, you know, and the way in which we've solved it to date or solved is probably over the way we've made progress to address it to date is like, okay, well you have these [00:58:00] major platforms, you know, like a Salesforce, like a HubSpot, you know, like a market or whatever it is. [00:58:05] That's like, okay, we've got enough momentum around us that we will, you know, gravitationally pull other companies to build bespoke integrations to our particular set of APIs. [00:58:18] Phil: Right. [00:58:19] Scott: So that that got us further down the road. So if you pick, you know, again, like whatever it is, Salesforce, HubSpot, you know, Braze, whatever it is, you know, you then have a set of things that, you know, you can pull from in their ecosystem that are integrate. [00:58:33] But there's probably a whole bunch of other things that you might wanna do that haven't integrated to get to a place where, you know, we've now got this emerging standard to say, well listen, if I just make something available as an MCP server, you know, and some other piece of software, you know, implements an MCP client, that kind of, again, a little hand wavy for a moment, but any MCP client being able to like, interact with any MCP server without anyone having to like, build, [00:59:00] you know, custom bespoke integrations, I think that's gonna be a massive unlock. [00:59:04] And again, I'm sure a lot of vendors individually, if they could say like, the love of God, don't do this. They can, it's, it's, it, it, it's almost hit this tipping button. Mean this is the beauty of standards is once something becomes a standard and people are like, yeah, no, we kind of have to do it. Um, and so I actually think composability is going to be, gonna be almost like the defacto, uh, you know, um, architectural, uh, property, uh, of MarTech here moving forward. [00:59:38] Phil: Scott's recently co-authored The State of MarTech for 2026 report released in May this year, and he's got the new MarTech landscape. [00:59:45] The categories gaining ground are CMS, web experience, e-commerce platform, iPaaS, data integration. You know, every growing category is benefiting from disruption hitting the rest of the stack. [00:59:57] And then there's MCP. Over [01:00:00] 29,000 MCP servers are now listed across public registries in only 18 months. It's insane. The composability that Scott's been describing for years now seems to be standard infrastructure. [01:00:13] The report's argument that MarTech is metamorphosizing from apps humans operate into infrastructure that agents can actually use, and in this last era, marketing platforms completed to be where marketers work. In the next era, they'll be competing to be what agents work with. so Scott's point here is almost giving us a map for the next, um, floor as we're descending down the dungeon here. The first two fights on this first floor were about location. The CRM had become too messy to trust as a system of record. We were exporting data everywhere, and that made every downstream tool its own drifting copy of the customer. MCP is changing that integration layer, but it does not magically fix everything underneath it. [01:00:57] Agents can talk to more tools now. [01:01:00] Great. Awesome. But if the data is messy, incomplete, stale, or missing the context behind why something happened, all we've done is give machines a faster way to be wrong. So the next floor is about the raw material itself. [01:01:14] So we've got one floor, one boss with two forms down, but we'll see you on the next floor [01:01:19] ​ [01:01:19] ​